]>
Commit | Line | Data |
---|---|---|
1b1d2e6d BP |
1 | /* |
2 | * Copyright (c) 2017, 2018 Nicira, Inc. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | ||
19 | #include "raft.h" | |
20 | #include "raft-private.h" | |
21 | ||
22 | #include <errno.h> | |
23 | #include <unistd.h> | |
24 | ||
25 | #include "hash.h" | |
26 | #include "jsonrpc.h" | |
27 | #include "lockfile.h" | |
28 | #include "openvswitch/dynamic-string.h" | |
29 | #include "openvswitch/hmap.h" | |
30 | #include "openvswitch/json.h" | |
31 | #include "openvswitch/list.h" | |
32 | #include "openvswitch/poll-loop.h" | |
33 | #include "openvswitch/vlog.h" | |
34 | #include "ovsdb-error.h" | |
35 | #include "ovsdb-parser.h" | |
36 | #include "ovsdb/log.h" | |
37 | #include "raft-rpc.h" | |
38 | #include "random.h" | |
3423cd97 | 39 | #include "simap.h" |
1b1d2e6d BP |
40 | #include "socket-util.h" |
41 | #include "stream.h" | |
42 | #include "timeval.h" | |
43 | #include "unicode.h" | |
44 | #include "unixctl.h" | |
45 | #include "util.h" | |
46 | #include "uuid.h" | |
47 | ||
48 | VLOG_DEFINE_THIS_MODULE(raft); | |
49 | ||
50 | /* Roles for a Raft server: | |
51 | * | |
52 | * - Followers: Servers in touch with the current leader. | |
53 | * | |
54 | * - Candidate: Servers unaware of a current leader and seeking election to | |
55 | * leader. | |
56 | * | |
57 | * - Leader: Handles all client requests. At most one at a time. | |
58 | * | |
59 | * In normal operation there is exactly one leader and all of the other servers | |
60 | * are followers. */ | |
61 | enum raft_role { | |
62 | RAFT_FOLLOWER, | |
63 | RAFT_CANDIDATE, | |
64 | RAFT_LEADER | |
65 | }; | |
66 | ||
67dba070 HZ |
67 | /* Flags for unit tests. */ |
68 | enum raft_failure_test { | |
69 | FT_NO_TEST, | |
70 | FT_CRASH_BEFORE_SEND_APPEND_REQ, | |
71 | FT_CRASH_AFTER_SEND_APPEND_REQ, | |
72 | FT_CRASH_BEFORE_SEND_EXEC_REP, | |
73 | FT_CRASH_AFTER_SEND_EXEC_REP, | |
74 | FT_CRASH_BEFORE_SEND_EXEC_REQ, | |
75 | FT_CRASH_AFTER_SEND_EXEC_REQ, | |
76 | FT_CRASH_AFTER_RECV_APPEND_REQ_UPDATE, | |
93ee4209 HZ |
77 | FT_DELAY_ELECTION, |
78 | FT_DONT_SEND_VOTE_REQUEST | |
67dba070 HZ |
79 | }; |
80 | static enum raft_failure_test failure_test; | |
81 | ||
1b1d2e6d BP |
82 | /* A connection between this Raft server and another one. */ |
83 | struct raft_conn { | |
84 | struct ovs_list list_node; /* In struct raft's 'conns' list. */ | |
85 | struct jsonrpc_session *js; /* JSON-RPC connection. */ | |
86 | struct uuid sid; /* This server's unique ID. */ | |
87 | char *nickname; /* Short name for use in log messages. */ | |
88 | bool incoming; /* True if incoming, false if outgoing. */ | |
89 | unsigned int js_seqno; /* Seqno for noticing (re)connections. */ | |
90 | }; | |
91 | ||
92 | static void raft_conn_close(struct raft_conn *); | |
93 | ||
94 | /* A "command", that is, a request to append an entry to the log. | |
95 | * | |
96 | * The Raft specification only allows clients to issue commands to the leader. | |
97 | * With this implementation, clients may issue a command on any server, which | |
98 | * then relays the command to the leader if necessary. | |
99 | * | |
100 | * This structure is thus used in three cases: | |
101 | * | |
102 | * 1. We are the leader and the command was issued to us directly. | |
103 | * | |
104 | * 2. We are a follower and relayed the command to the leader. | |
105 | * | |
106 | * 3. We are the leader and a follower relayed the command to us. | |
107 | */ | |
108 | struct raft_command { | |
109 | /* All cases. */ | |
110 | struct hmap_node hmap_node; /* In struct raft's 'commands' hmap. */ | |
111 | unsigned int n_refs; /* Reference count. */ | |
112 | enum raft_command_status status; /* Execution status. */ | |
7ef36089 | 113 | struct uuid eid; /* Entry ID of result. */ |
1b1d2e6d BP |
114 | |
115 | /* Case 1 only. */ | |
116 | uint64_t index; /* Index in log (0 if being relayed). */ | |
117 | ||
1b1d2e6d BP |
118 | /* Case 2 only. */ |
119 | long long int timestamp; /* Issue or last ping time, for expiration. */ | |
120 | ||
121 | /* Case 3 only. */ | |
122 | struct uuid sid; /* The follower (otherwise UUID_ZERO). */ | |
123 | }; | |
124 | ||
125 | static void raft_command_complete(struct raft *, struct raft_command *, | |
126 | enum raft_command_status); | |
127 | ||
128 | static void raft_complete_all_commands(struct raft *, | |
129 | enum raft_command_status); | |
130 | ||
131 | /* Type of deferred action, see struct raft_waiter. */ | |
132 | enum raft_waiter_type { | |
133 | RAFT_W_ENTRY, | |
134 | RAFT_W_TERM, | |
135 | RAFT_W_RPC, | |
136 | }; | |
137 | ||
138 | /* An action deferred until a log write commits to disk. */ | |
139 | struct raft_waiter { | |
140 | struct ovs_list list_node; | |
141 | uint64_t commit_ticket; | |
142 | ||
143 | enum raft_waiter_type type; | |
144 | union { | |
145 | /* RAFT_W_ENTRY. | |
146 | * | |
147 | * Waits for a RAFT_REC_ENTRY write to our local log to commit. Upon | |
148 | * completion, updates 'log_synced' to indicate that the new log entry | |
149 | * or entries are committed and, if we are leader, also updates our | |
150 | * local 'match_index'. */ | |
151 | struct { | |
152 | uint64_t index; | |
153 | } entry; | |
154 | ||
155 | /* RAFT_W_TERM. | |
156 | * | |
157 | * Waits for a RAFT_REC_TERM or RAFT_REC_VOTE record write to commit. | |
158 | * Upon completion, updates 'synced_term' and 'synced_vote', which | |
159 | * triggers sending RPCs deferred by the uncommitted 'term' and | |
160 | * 'vote'. */ | |
161 | struct { | |
162 | uint64_t term; | |
163 | struct uuid vote; | |
164 | } term; | |
165 | ||
166 | /* RAFT_W_RPC. | |
167 | * | |
168 | * Sometimes, sending an RPC to a peer must be delayed until an entry, | |
169 | * a term, or a vote mentioned in the RPC is synced to disk. This | |
170 | * waiter keeps a copy of such an RPC until the previous waiters have | |
171 | * committed. */ | |
172 | union raft_rpc *rpc; | |
173 | }; | |
174 | }; | |
175 | ||
176 | static struct raft_waiter *raft_waiter_create(struct raft *, | |
177 | enum raft_waiter_type, | |
178 | bool start_commit); | |
179 | static void raft_waiters_destroy(struct raft *); | |
180 | ||
181 | /* The Raft state machine. */ | |
182 | struct raft { | |
183 | struct hmap_node hmap_node; /* In 'all_rafts'. */ | |
184 | struct ovsdb_log *log; | |
185 | ||
186 | /* Persistent derived state. | |
187 | * | |
188 | * This must be updated on stable storage before responding to RPCs. It can be | |
189 | * derived from the header, snapshot, and log in 'log'. */ | |
190 | ||
191 | struct uuid cid; /* Cluster ID (immutable for the cluster). */ | |
192 | struct uuid sid; /* Server ID (immutable for the server). */ | |
193 | char *local_address; /* Local address (immutable for the server). */ | |
194 | char *local_nickname; /* Used for local server in log messages. */ | |
195 | char *name; /* Schema name (immutable for the cluster). */ | |
196 | ||
197 | /* Contains "struct raft_server"s and represents the server configuration | |
198 | * most recently added to 'log'. */ | |
199 | struct hmap servers; | |
200 | ||
8e354614 HZ |
201 | #define ELECTION_BASE_MSEC 1000 |
202 | #define ELECTION_RANGE_MSEC 1000 | |
203 | /* The election timeout base value for leader election, in milliseconds. | |
204 | * It can be set by unixctl cluster/change-election-timer. Default value is | |
205 | * ELECTION_BASE_MSEC. */ | |
206 | uint64_t election_timer; | |
207 | /* If not 0, it is the new value of election_timer being proposed. */ | |
208 | uint64_t election_timer_new; | |
209 | ||
1b1d2e6d BP |
210 | /* Persistent state on all servers. |
211 | * | |
212 | * Must be updated on stable storage before responding to RPCs. */ | |
213 | ||
214 | /* Current term and the vote for that term. These might be on the way to | |
215 | * disk now. */ | |
216 | uint64_t term; /* Initialized to 0 and only increases. */ | |
217 | struct uuid vote; /* All-zeros if no vote yet in 'term'. */ | |
218 | ||
219 | /* The term and vote that have been synced to disk. */ | |
220 | uint64_t synced_term; | |
221 | struct uuid synced_vote; | |
222 | ||
223 | /* The log. | |
224 | * | |
225 | * A log entry with index 1 never really exists; the initial snapshot for a | |
226 | * Raft is considered to include this index. The first real log entry has | |
227 | * index 2. | |
228 | * | |
229 | * A new Raft instance contains an empty log: log_start=2, log_end=2. | |
230 | * Over time, the log grows: log_start=2, log_end=N. | |
231 | * At some point, the server takes a snapshot: log_start=N, log_end=N. | |
232 | * The log continues to grow: log_start=N, log_end=N+1... | |
233 | * | |
234 | * Must be updated on stable storage before responding to RPCs. */ | |
235 | struct raft_entry *entries; /* Log entry i is in log[i - log_start]. */ | |
236 | uint64_t log_start; /* Index of first entry in log. */ | |
237 | uint64_t log_end; /* Index of last entry in log, plus 1. */ | |
238 | uint64_t log_synced; /* Index of last synced entry. */ | |
239 | size_t allocated_log; /* Allocated entries in 'log'. */ | |
240 | ||
241 | /* Snapshot state (see Figure 5.1) | |
242 | * | |
243 | * This is the state of the cluster as of the last discarded log entry, | |
244 | * that is, at log index 'log_start - 1' (called prevIndex in Figure 5.1). | |
245 | * Only committed log entries can be included in a snapshot. */ | |
246 | struct raft_entry snap; | |
247 | ||
248 | /* Volatile state. | |
249 | * | |
250 | * The snapshot is always committed, but the rest of the log might not be yet. | |
251 | * 'last_applied' tracks what entries have been passed to the client. If the | |
252 | * client hasn't yet read the latest snapshot, then even the snapshot isn't | |
253 | * applied yet. Thus, the invariants are different for these members: | |
254 | * | |
255 | * log_start - 2 <= last_applied <= commit_index < log_end. | |
256 | * log_start - 1 <= commit_index < log_end. | |
257 | */ | |
258 | ||
259 | enum raft_role role; /* Current role. */ | |
260 | uint64_t commit_index; /* Max log index known to be committed. */ | |
261 | uint64_t last_applied; /* Max log index applied to state machine. */ | |
262 | struct uuid leader_sid; /* Server ID of leader (zero, if unknown). */ | |
263 | ||
1b1d2e6d BP |
264 | long long int election_base; /* Time of last heartbeat from leader. */ |
265 | long long int election_timeout; /* Time at which we start an election. */ | |
266 | ||
267 | /* Used for joining a cluster. */ | |
268 | bool joining; /* Attempting to join the cluster? */ | |
269 | struct sset remote_addresses; /* Addresses to try to find other servers. */ | |
270 | long long int join_timeout; /* Time to re-send add server request. */ | |
271 | ||
272 | /* Used for leaving a cluster. */ | |
273 | bool leaving; /* True if we are leaving the cluster. */ | |
274 | bool left; /* True if we have finished leaving. */ | |
275 | long long int leave_timeout; /* Time to re-send remove server request. */ | |
276 | ||
277 | /* Failure. */ | |
278 | bool failed; /* True if unrecoverable error has occurred. */ | |
279 | ||
280 | /* File synchronization. */ | |
281 | struct ovs_list waiters; /* Contains "struct raft_waiter"s. */ | |
282 | ||
283 | /* Network connections. */ | |
284 | struct pstream *listener; /* For connections from other Raft servers. */ | |
285 | long long int listen_backoff; /* For retrying creating 'listener'. */ | |
286 | struct ovs_list conns; /* Contains struct raft_conns. */ | |
287 | ||
288 | /* Leaders only. Reinitialized after becoming leader. */ | |
289 | struct hmap add_servers; /* Contains "struct raft_server"s to add. */ | |
290 | struct raft_server *remove_server; /* Server being removed. */ | |
291 | struct hmap commands; /* Contains "struct raft_command"s. */ | |
1b1d2e6d BP |
292 | long long int ping_timeout; /* Time at which to send a heartbeat */ |
293 | ||
294 | /* Candidates only. Reinitialized at start of election. */ | |
295 | int n_votes; /* Number of votes for me. */ | |
923f01ca HZ |
296 | |
297 | /* Followers and candidates only. */ | |
298 | bool candidate_retrying; /* The earlier election timed-out and we are | |
299 | now retrying. */ | |
300 | bool had_leader; /* There has been leader elected since last | |
301 | election initiated. This is to help setting | |
302 | candidate_retrying. */ | |
2833885f HZ |
303 | |
304 | /* For all. */ | |
305 | bool ever_had_leader; /* There has been leader elected since the raft | |
306 | is initialized, meaning it is ever | |
307 | connected. */ | |
1b1d2e6d BP |
308 | }; |
309 | ||
310 | /* All Raft structures. */ | |
311 | static struct hmap all_rafts = HMAP_INITIALIZER(&all_rafts); | |
312 | ||
313 | static void raft_init(void); | |
314 | ||
315 | static struct ovsdb_error *raft_read_header(struct raft *) | |
316 | OVS_WARN_UNUSED_RESULT; | |
317 | ||
318 | static void raft_send_execute_command_reply(struct raft *, | |
319 | const struct uuid *sid, | |
320 | const struct uuid *eid, | |
321 | enum raft_command_status, | |
322 | uint64_t commit_index); | |
323 | ||
324 | static void raft_update_our_match_index(struct raft *, uint64_t min_index); | |
325 | ||
326 | static void raft_send_remove_server_reply__( | |
327 | struct raft *, const struct uuid *target_sid, | |
328 | const struct uuid *requester_sid, struct unixctl_conn *requester_conn, | |
329 | bool success, const char *comment); | |
17bd4149 | 330 | static void raft_finished_leaving_cluster(struct raft *); |
1b1d2e6d BP |
331 | |
332 | static void raft_server_init_leader(struct raft *, struct raft_server *); | |
333 | ||
334 | static bool raft_rpc_is_heartbeat(const union raft_rpc *); | |
335 | static bool raft_is_rpc_synced(const struct raft *, const union raft_rpc *); | |
336 | ||
337 | static void raft_handle_rpc(struct raft *, const union raft_rpc *); | |
17bd4149 | 338 | |
02acb41a BP |
339 | static bool raft_send_at(struct raft *, const union raft_rpc *, |
340 | int line_number); | |
341 | #define raft_send(raft, rpc) raft_send_at(raft, rpc, __LINE__) | |
342 | ||
343 | static bool raft_send_to_conn_at(struct raft *, const union raft_rpc *, | |
344 | struct raft_conn *, int line_number); | |
345 | #define raft_send_to_conn(raft, rpc, conn) \ | |
346 | raft_send_to_conn_at(raft, rpc, conn, __LINE__) | |
347 | ||
1b1d2e6d BP |
348 | static void raft_send_append_request(struct raft *, |
349 | struct raft_server *, unsigned int n, | |
350 | const char *comment); | |
351 | ||
352 | static void raft_become_leader(struct raft *); | |
353 | static void raft_become_follower(struct raft *); | |
0f954f32 HZ |
354 | static void raft_reset_election_timer(struct raft *); |
355 | static void raft_reset_ping_timer(struct raft *); | |
1b1d2e6d BP |
356 | static void raft_send_heartbeats(struct raft *); |
357 | static void raft_start_election(struct raft *, bool leadership_transfer); | |
358 | static bool raft_truncate(struct raft *, uint64_t new_end); | |
359 | static void raft_get_servers_from_log(struct raft *, enum vlog_level); | |
8e354614 | 360 | static void raft_get_election_timer_from_log(struct raft *); |
1b1d2e6d BP |
361 | |
362 | static bool raft_handle_write_error(struct raft *, struct ovsdb_error *); | |
363 | ||
364 | static void raft_run_reconfigure(struct raft *); | |
365 | ||
923f01ca | 366 | static void raft_set_leader(struct raft *, const struct uuid *sid); |
1b1d2e6d BP |
367 | static struct raft_server * |
368 | raft_find_server(const struct raft *raft, const struct uuid *sid) | |
369 | { | |
370 | return raft_server_find(&raft->servers, sid); | |
371 | } | |
372 | ||
373 | static char * | |
374 | raft_make_address_passive(const char *address_) | |
375 | { | |
376 | if (!strncmp(address_, "unix:", 5)) { | |
377 | return xasprintf("p%s", address_); | |
378 | } else { | |
379 | char *address = xstrdup(address_); | |
0b043300 BP |
380 | char *host, *port; |
381 | inet_parse_host_port_tokens(strchr(address, ':') + 1, &host, &port); | |
1b1d2e6d BP |
382 | |
383 | struct ds paddr = DS_EMPTY_INITIALIZER; | |
384 | ds_put_format(&paddr, "p%.3s:%s:", address, port); | |
385 | if (strchr(host, ':')) { | |
386 | ds_put_format(&paddr, "[%s]", host); | |
387 | } else { | |
388 | ds_put_cstr(&paddr, host); | |
389 | } | |
390 | free(address); | |
391 | return ds_steal_cstr(&paddr); | |
392 | } | |
393 | } | |
394 | ||
395 | static struct raft * | |
396 | raft_alloc(void) | |
397 | { | |
398 | raft_init(); | |
399 | ||
400 | struct raft *raft = xzalloc(sizeof *raft); | |
401 | hmap_node_nullify(&raft->hmap_node); | |
402 | hmap_init(&raft->servers); | |
403 | raft->log_start = raft->log_end = 1; | |
404 | raft->role = RAFT_FOLLOWER; | |
405 | sset_init(&raft->remote_addresses); | |
406 | raft->join_timeout = LLONG_MAX; | |
407 | ovs_list_init(&raft->waiters); | |
408 | raft->listen_backoff = LLONG_MIN; | |
409 | ovs_list_init(&raft->conns); | |
410 | hmap_init(&raft->add_servers); | |
411 | hmap_init(&raft->commands); | |
412 | ||
8e354614 | 413 | raft->election_timer = ELECTION_BASE_MSEC; |
1b1d2e6d BP |
414 | |
415 | return raft; | |
416 | } | |
417 | ||
418 | /* Creates an on-disk file that represents a new Raft cluster and initializes | |
419 | * it to consist of a single server, the one on which this function is called. | |
420 | * | |
421 | * Creates the local copy of the cluster's log in 'file_name', which must not | |
422 | * already exist. Gives it the name 'name', which should be the database | |
423 | * schema name and which is used only to match up this database with the server | |
424 | * added to the cluster later if the cluster ID is unavailable. | |
425 | * | |
426 | * The new server is located at 'local_address', which must take one of the | |
427 | * forms "tcp:IP:PORT" or "ssl:IP:PORT", where IP is an IPv4 address or a | |
428 | * square bracket enclosed IPv6 address and PORT is a TCP port number. | |
429 | * | |
430 | * This only creates the on-disk file. Use raft_open() to start operating the | |
431 | * new server. | |
432 | * | |
433 | * Returns null if successful, otherwise an ovsdb_error describing the | |
434 | * problem. */ | |
435 | struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
436 | raft_create_cluster(const char *file_name, const char *name, | |
437 | const char *local_address, const struct json *data) | |
438 | { | |
439 | /* Parse and verify validity of the local address. */ | |
440 | struct ovsdb_error *error = raft_address_validate(local_address); | |
441 | if (error) { | |
442 | return error; | |
443 | } | |
444 | ||
445 | /* Create log file. */ | |
446 | struct ovsdb_log *log; | |
447 | error = ovsdb_log_open(file_name, RAFT_MAGIC, OVSDB_LOG_CREATE_EXCL, | |
448 | -1, &log); | |
449 | if (error) { | |
450 | return error; | |
451 | } | |
452 | ||
453 | /* Write log file. */ | |
454 | struct raft_header h = { | |
455 | .sid = uuid_random(), | |
456 | .cid = uuid_random(), | |
457 | .name = xstrdup(name), | |
458 | .local_address = xstrdup(local_address), | |
459 | .joining = false, | |
460 | .remote_addresses = SSET_INITIALIZER(&h.remote_addresses), | |
461 | .snap_index = 1, | |
462 | .snap = { | |
463 | .term = 1, | |
464 | .data = json_nullable_clone(data), | |
465 | .eid = uuid_random(), | |
466 | .servers = json_object_create(), | |
467 | }, | |
468 | }; | |
469 | shash_add_nocopy(json_object(h.snap.servers), | |
470 | xasprintf(UUID_FMT, UUID_ARGS(&h.sid)), | |
471 | json_string_create(local_address)); | |
472 | error = ovsdb_log_write_and_free(log, raft_header_to_json(&h)); | |
473 | raft_header_uninit(&h); | |
474 | if (!error) { | |
475 | error = ovsdb_log_commit_block(log); | |
476 | } | |
477 | ovsdb_log_close(log); | |
478 | ||
479 | return error; | |
480 | } | |
481 | ||
482 | /* Creates a database file that represents a new server in an existing Raft | |
483 | * cluster. | |
484 | * | |
485 | * Creates the local copy of the cluster's log in 'file_name', which must not | |
486 | * already exist. Gives it the name 'name', which must be the same name | |
487 | * passed in to raft_create_cluster() earlier. | |
488 | * | |
489 | * 'cid' is optional. If specified, the new server will join only the cluster | |
490 | * with the given cluster ID. | |
491 | * | |
492 | * The new server is located at 'local_address', which must take one of the | |
493 | * forms "tcp:IP:PORT" or "ssl:IP:PORT", where IP is an IPv4 address or a | |
494 | * square bracket enclosed IPv6 address and PORT is a TCP port number. | |
495 | * | |
496 | * Joining the cluster requires contacting it. Thus, 'remote_addresses' | |
497 | * specifies the addresses of existing servers in the cluster. One server out | |
498 | * of the existing cluster is sufficient, as long as that server is reachable | |
499 | * and not partitioned from the current cluster leader. If multiple servers | |
500 | * from the cluster are specified, then it is sufficient for any of them to | |
501 | * meet this criterion. | |
502 | * | |
503 | * This only creates the on-disk file and does no network access. Use | |
504 | * raft_open() to start operating the new server. (Until this happens, the | |
505 | * new server has not joined the cluster.) | |
506 | * | |
507 | * Returns null if successful, otherwise an ovsdb_error describing the | |
508 | * problem. */ | |
509 | struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
510 | raft_join_cluster(const char *file_name, | |
511 | const char *name, const char *local_address, | |
512 | const struct sset *remote_addresses, | |
513 | const struct uuid *cid) | |
514 | { | |
515 | ovs_assert(!sset_is_empty(remote_addresses)); | |
516 | ||
517 | /* Parse and verify validity of the addresses. */ | |
518 | struct ovsdb_error *error = raft_address_validate(local_address); | |
519 | if (error) { | |
520 | return error; | |
521 | } | |
522 | const char *addr; | |
523 | SSET_FOR_EACH (addr, remote_addresses) { | |
524 | error = raft_address_validate(addr); | |
525 | if (error) { | |
526 | return error; | |
527 | } | |
528 | if (!strcmp(addr, local_address)) { | |
529 | return ovsdb_error(NULL, "remote addresses cannot be the same " | |
530 | "as the local address"); | |
531 | } | |
532 | } | |
533 | ||
534 | /* Verify validity of the cluster ID (if provided). */ | |
535 | if (cid && uuid_is_zero(cid)) { | |
536 | return ovsdb_error(NULL, "all-zero UUID is not valid cluster ID"); | |
537 | } | |
538 | ||
539 | /* Create log file. */ | |
540 | struct ovsdb_log *log; | |
541 | error = ovsdb_log_open(file_name, RAFT_MAGIC, OVSDB_LOG_CREATE_EXCL, | |
542 | -1, &log); | |
543 | if (error) { | |
544 | return error; | |
545 | } | |
546 | ||
547 | /* Write log file. */ | |
548 | struct raft_header h = { | |
549 | .sid = uuid_random(), | |
550 | .cid = cid ? *cid : UUID_ZERO, | |
551 | .name = xstrdup(name), | |
552 | .local_address = xstrdup(local_address), | |
553 | .joining = true, | |
554 | /* No snapshot yet. */ | |
555 | }; | |
556 | sset_clone(&h.remote_addresses, remote_addresses); | |
557 | error = ovsdb_log_write_and_free(log, raft_header_to_json(&h)); | |
558 | raft_header_uninit(&h); | |
559 | if (!error) { | |
560 | error = ovsdb_log_commit_block(log); | |
561 | } | |
562 | ovsdb_log_close(log); | |
563 | ||
564 | return error; | |
565 | } | |
566 | ||
567 | /* Reads the initial header record from 'log', which must be a Raft clustered | |
568 | * database log, and populates '*md' with the information read from it. The | |
569 | * caller must eventually destroy 'md' with raft_metadata_destroy(). | |
570 | * | |
571 | * On success, returns NULL. On failure, returns an error that the caller must | |
572 | * eventually destroy and zeros '*md'. */ | |
573 | struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
574 | raft_read_metadata(struct ovsdb_log *log, struct raft_metadata *md) | |
575 | { | |
576 | struct raft *raft = raft_alloc(); | |
577 | raft->log = log; | |
578 | ||
579 | struct ovsdb_error *error = raft_read_header(raft); | |
580 | if (!error) { | |
581 | md->sid = raft->sid; | |
582 | md->name = xstrdup(raft->name); | |
583 | md->local = xstrdup(raft->local_address); | |
584 | md->cid = raft->cid; | |
585 | } else { | |
586 | memset(md, 0, sizeof *md); | |
587 | } | |
588 | ||
589 | raft->log = NULL; | |
590 | raft_close(raft); | |
591 | return error; | |
592 | } | |
593 | ||
594 | /* Frees the metadata in 'md'. */ | |
595 | void | |
596 | raft_metadata_destroy(struct raft_metadata *md) | |
597 | { | |
598 | if (md) { | |
599 | free(md->name); | |
600 | free(md->local); | |
601 | } | |
602 | } | |
603 | ||
604 | static const struct raft_entry * | |
605 | raft_get_entry(const struct raft *raft, uint64_t index) | |
606 | { | |
607 | ovs_assert(index >= raft->log_start); | |
608 | ovs_assert(index < raft->log_end); | |
609 | return &raft->entries[index - raft->log_start]; | |
610 | } | |
611 | ||
612 | static uint64_t | |
613 | raft_get_term(const struct raft *raft, uint64_t index) | |
614 | { | |
615 | return (index == raft->log_start - 1 | |
616 | ? raft->snap.term | |
617 | : raft_get_entry(raft, index)->term); | |
618 | } | |
619 | ||
620 | static const struct json * | |
621 | raft_servers_for_index(const struct raft *raft, uint64_t index) | |
622 | { | |
623 | ovs_assert(index >= raft->log_start - 1); | |
624 | ovs_assert(index < raft->log_end); | |
625 | ||
626 | const struct json *servers = raft->snap.servers; | |
627 | for (uint64_t i = raft->log_start; i <= index; i++) { | |
628 | const struct raft_entry *e = raft_get_entry(raft, i); | |
629 | if (e->servers) { | |
630 | servers = e->servers; | |
631 | } | |
632 | } | |
633 | return servers; | |
634 | } | |
635 | ||
636 | static void | |
637 | raft_set_servers(struct raft *raft, const struct hmap *new_servers, | |
638 | enum vlog_level level) | |
639 | { | |
640 | struct raft_server *s, *next; | |
641 | HMAP_FOR_EACH_SAFE (s, next, hmap_node, &raft->servers) { | |
642 | if (!raft_server_find(new_servers, &s->sid)) { | |
643 | ovs_assert(s != raft->remove_server); | |
644 | ||
645 | hmap_remove(&raft->servers, &s->hmap_node); | |
646 | VLOG(level, "server %s removed from configuration", s->nickname); | |
647 | raft_server_destroy(s); | |
648 | } | |
649 | } | |
650 | ||
651 | HMAP_FOR_EACH_SAFE (s, next, hmap_node, new_servers) { | |
652 | if (!raft_find_server(raft, &s->sid)) { | |
653 | VLOG(level, "server %s added to configuration", s->nickname); | |
654 | ||
655 | struct raft_server *new | |
656 | = raft_server_add(&raft->servers, &s->sid, s->address); | |
657 | raft_server_init_leader(raft, new); | |
658 | } | |
659 | } | |
660 | } | |
661 | ||
662 | static uint64_t | |
663 | raft_add_entry(struct raft *raft, | |
664 | uint64_t term, struct json *data, const struct uuid *eid, | |
8e354614 | 665 | struct json *servers, uint64_t election_timer) |
1b1d2e6d BP |
666 | { |
667 | if (raft->log_end - raft->log_start >= raft->allocated_log) { | |
668 | raft->entries = x2nrealloc(raft->entries, &raft->allocated_log, | |
669 | sizeof *raft->entries); | |
670 | } | |
671 | ||
672 | uint64_t index = raft->log_end++; | |
673 | struct raft_entry *entry = &raft->entries[index - raft->log_start]; | |
674 | entry->term = term; | |
675 | entry->data = data; | |
676 | entry->eid = eid ? *eid : UUID_ZERO; | |
677 | entry->servers = servers; | |
8e354614 | 678 | entry->election_timer = election_timer; |
1b1d2e6d BP |
679 | return index; |
680 | } | |
681 | ||
8e354614 HZ |
682 | /* Writes a RAFT_REC_ENTRY record for 'term', 'data', 'eid', 'servers', |
683 | * 'election_timer' to * 'raft''s log and returns an error indication. */ | |
1b1d2e6d BP |
684 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT |
685 | raft_write_entry(struct raft *raft, uint64_t term, struct json *data, | |
8e354614 HZ |
686 | const struct uuid *eid, struct json *servers, |
687 | uint64_t election_timer) | |
1b1d2e6d BP |
688 | { |
689 | struct raft_record r = { | |
690 | .type = RAFT_REC_ENTRY, | |
691 | .term = term, | |
692 | .entry = { | |
8e354614 HZ |
693 | .index = raft_add_entry(raft, term, data, eid, servers, |
694 | election_timer), | |
1b1d2e6d BP |
695 | .data = data, |
696 | .servers = servers, | |
8e354614 | 697 | .election_timer = election_timer, |
1b1d2e6d BP |
698 | .eid = eid ? *eid : UUID_ZERO, |
699 | }, | |
700 | }; | |
701 | return ovsdb_log_write_and_free(raft->log, raft_record_to_json(&r)); | |
702 | } | |
703 | ||
704 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
705 | raft_write_state(struct ovsdb_log *log, | |
706 | uint64_t term, const struct uuid *vote) | |
707 | { | |
708 | struct raft_record r = { .term = term }; | |
709 | if (vote && !uuid_is_zero(vote)) { | |
710 | r.type = RAFT_REC_VOTE; | |
711 | r.sid = *vote; | |
712 | } else { | |
713 | r.type = RAFT_REC_TERM; | |
714 | } | |
715 | return ovsdb_log_write_and_free(log, raft_record_to_json(&r)); | |
716 | } | |
717 | ||
718 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
719 | raft_apply_record(struct raft *raft, unsigned long long int rec_idx, | |
720 | const struct raft_record *r) | |
721 | { | |
722 | /* Apply "term", which is present in most kinds of records (and otherwise | |
723 | * 0). | |
724 | * | |
725 | * A Raft leader can replicate entries from previous terms to the other | |
726 | * servers in the cluster, retaining the original terms on those entries | |
727 | * (see section 3.6.2 "Committing entries from previous terms" for more | |
728 | * information), so it's OK for the term in a log record to precede the | |
729 | * current term. */ | |
730 | if (r->term > raft->term) { | |
731 | raft->term = raft->synced_term = r->term; | |
732 | raft->vote = raft->synced_vote = UUID_ZERO; | |
733 | } | |
734 | ||
735 | switch (r->type) { | |
736 | case RAFT_REC_ENTRY: | |
737 | if (r->entry.index < raft->commit_index) { | |
738 | return ovsdb_error(NULL, "record %llu attempts to truncate log " | |
739 | "from %"PRIu64" to %"PRIu64" entries, but " | |
740 | "commit index is already %"PRIu64, | |
741 | rec_idx, raft->log_end, r->entry.index, | |
742 | raft->commit_index); | |
743 | } else if (r->entry.index > raft->log_end) { | |
744 | return ovsdb_error(NULL, "record %llu with index %"PRIu64" skips " | |
745 | "past expected index %"PRIu64, | |
746 | rec_idx, r->entry.index, raft->log_end); | |
747 | } | |
748 | ||
749 | if (r->entry.index < raft->log_end) { | |
750 | /* This can happen, but it is notable. */ | |
751 | VLOG_DBG("record %llu truncates log from %"PRIu64" to %"PRIu64 | |
752 | " entries", rec_idx, raft->log_end, r->entry.index); | |
753 | raft_truncate(raft, r->entry.index); | |
754 | } | |
755 | ||
756 | uint64_t prev_term = (raft->log_end > raft->log_start | |
757 | ? raft->entries[raft->log_end | |
758 | - raft->log_start - 1].term | |
759 | : raft->snap.term); | |
760 | if (r->term < prev_term) { | |
761 | return ovsdb_error(NULL, "record %llu with index %"PRIu64" term " | |
762 | "%"PRIu64" precedes previous entry's term " | |
763 | "%"PRIu64, | |
764 | rec_idx, r->entry.index, r->term, prev_term); | |
765 | } | |
766 | ||
767 | raft->log_synced = raft_add_entry( | |
768 | raft, r->term, | |
769 | json_nullable_clone(r->entry.data), &r->entry.eid, | |
8e354614 HZ |
770 | json_nullable_clone(r->entry.servers), |
771 | r->entry.election_timer); | |
1b1d2e6d BP |
772 | return NULL; |
773 | ||
774 | case RAFT_REC_TERM: | |
775 | return NULL; | |
776 | ||
777 | case RAFT_REC_VOTE: | |
778 | if (r->term < raft->term) { | |
779 | return ovsdb_error(NULL, "record %llu votes for term %"PRIu64" " | |
780 | "but current term is %"PRIu64, | |
781 | rec_idx, r->term, raft->term); | |
782 | } else if (!uuid_is_zero(&raft->vote) | |
783 | && !uuid_equals(&raft->vote, &r->sid)) { | |
784 | return ovsdb_error(NULL, "record %llu votes for "SID_FMT" in term " | |
785 | "%"PRIu64" but a previous record for the " | |
786 | "same term voted for "SID_FMT, rec_idx, | |
787 | SID_ARGS(&raft->vote), r->term, | |
788 | SID_ARGS(&r->sid)); | |
789 | } else { | |
790 | raft->vote = raft->synced_vote = r->sid; | |
791 | return NULL; | |
792 | } | |
793 | break; | |
794 | ||
795 | case RAFT_REC_NOTE: | |
796 | if (!strcmp(r->note, "left")) { | |
797 | return ovsdb_error(NULL, "record %llu indicates server has left " | |
798 | "the cluster; it cannot be added back (use " | |
799 | "\"ovsdb-tool join-cluster\" to add a new " | |
800 | "server)", rec_idx); | |
801 | } | |
802 | return NULL; | |
803 | ||
804 | case RAFT_REC_COMMIT_INDEX: | |
805 | if (r->commit_index < raft->commit_index) { | |
806 | return ovsdb_error(NULL, "record %llu regresses commit index " | |
807 | "from %"PRIu64 " to %"PRIu64, | |
808 | rec_idx, raft->commit_index, r->commit_index); | |
809 | } else if (r->commit_index >= raft->log_end) { | |
810 | return ovsdb_error(NULL, "record %llu advances commit index to " | |
811 | "%"PRIu64 " but last log index is %"PRIu64, | |
812 | rec_idx, r->commit_index, raft->log_end - 1); | |
813 | } else { | |
814 | raft->commit_index = r->commit_index; | |
815 | return NULL; | |
816 | } | |
817 | break; | |
818 | ||
819 | case RAFT_REC_LEADER: | |
820 | /* XXX we could use this to take back leadership for quick restart */ | |
821 | return NULL; | |
822 | ||
823 | default: | |
824 | OVS_NOT_REACHED(); | |
825 | } | |
826 | } | |
827 | ||
828 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
829 | raft_read_header(struct raft *raft) | |
830 | { | |
831 | /* Read header record. */ | |
832 | struct json *json; | |
833 | struct ovsdb_error *error = ovsdb_log_read(raft->log, &json); | |
834 | if (error || !json) { | |
835 | /* Report error or end-of-file. */ | |
836 | return error; | |
837 | } | |
838 | ovsdb_log_mark_base(raft->log); | |
839 | ||
840 | struct raft_header h; | |
841 | error = raft_header_from_json(&h, json); | |
842 | json_destroy(json); | |
843 | if (error) { | |
844 | return error; | |
845 | } | |
846 | ||
847 | raft->sid = h.sid; | |
848 | raft->cid = h.cid; | |
849 | raft->name = xstrdup(h.name); | |
850 | raft->local_address = xstrdup(h.local_address); | |
851 | raft->local_nickname = raft_address_to_nickname(h.local_address, &h.sid); | |
852 | raft->joining = h.joining; | |
853 | ||
854 | if (h.joining) { | |
855 | sset_clone(&raft->remote_addresses, &h.remote_addresses); | |
856 | } else { | |
857 | raft_entry_clone(&raft->snap, &h.snap); | |
858 | raft->log_start = raft->log_end = h.snap_index + 1; | |
7efc6980 | 859 | raft->log_synced = raft->commit_index = h.snap_index; |
1b1d2e6d BP |
860 | raft->last_applied = h.snap_index - 1; |
861 | } | |
862 | ||
863 | raft_header_uninit(&h); | |
864 | ||
865 | return NULL; | |
866 | } | |
867 | ||
868 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
869 | raft_read_log(struct raft *raft) | |
870 | { | |
871 | for (unsigned long long int i = 1; ; i++) { | |
872 | struct json *json; | |
873 | struct ovsdb_error *error = ovsdb_log_read(raft->log, &json); | |
874 | if (!json) { | |
875 | if (error) { | |
876 | /* We assume that the error is due to a partial write while | |
877 | * appending to the file before a crash, so log it and | |
878 | * continue. */ | |
879 | char *error_string = ovsdb_error_to_string_free(error); | |
880 | VLOG_WARN("%s", error_string); | |
881 | free(error_string); | |
882 | error = NULL; | |
883 | } | |
884 | break; | |
885 | } | |
886 | ||
887 | struct raft_record r; | |
888 | error = raft_record_from_json(&r, json); | |
889 | if (!error) { | |
890 | error = raft_apply_record(raft, i, &r); | |
891 | raft_record_uninit(&r); | |
892 | } | |
ce52f1ee | 893 | json_destroy(json); |
1b1d2e6d BP |
894 | if (error) { |
895 | return ovsdb_wrap_error(error, "error reading record %llu from " | |
896 | "%s log", i, raft->name); | |
897 | } | |
898 | } | |
899 | ||
900 | /* Set the most recent servers. */ | |
901 | raft_get_servers_from_log(raft, VLL_DBG); | |
902 | ||
8e354614 HZ |
903 | /* Set the most recent election_timer. */ |
904 | raft_get_election_timer_from_log(raft); | |
905 | ||
1b1d2e6d BP |
906 | return NULL; |
907 | } | |
908 | ||
909 | static void | |
0f954f32 | 910 | raft_reset_election_timer(struct raft *raft) |
1b1d2e6d | 911 | { |
8e354614 | 912 | unsigned int duration = (raft->election_timer |
1b1d2e6d BP |
913 | + random_range(ELECTION_RANGE_MSEC)); |
914 | raft->election_base = time_msec(); | |
67dba070 HZ |
915 | if (failure_test == FT_DELAY_ELECTION) { |
916 | /* Slow down this node so that it won't win the next election. */ | |
8e354614 | 917 | duration += raft->election_timer; |
67dba070 | 918 | } |
1b1d2e6d BP |
919 | raft->election_timeout = raft->election_base + duration; |
920 | } | |
921 | ||
0f954f32 HZ |
922 | static void |
923 | raft_reset_ping_timer(struct raft *raft) | |
924 | { | |
8e354614 | 925 | raft->ping_timeout = time_msec() + raft->election_timer / 3; |
0f954f32 HZ |
926 | } |
927 | ||
1b1d2e6d BP |
928 | static void |
929 | raft_add_conn(struct raft *raft, struct jsonrpc_session *js, | |
930 | const struct uuid *sid, bool incoming) | |
931 | { | |
932 | struct raft_conn *conn = xzalloc(sizeof *conn); | |
933 | ovs_list_push_back(&raft->conns, &conn->list_node); | |
934 | conn->js = js; | |
935 | if (sid) { | |
936 | conn->sid = *sid; | |
937 | } | |
938 | conn->nickname = raft_address_to_nickname(jsonrpc_session_get_name(js), | |
939 | &conn->sid); | |
940 | conn->incoming = incoming; | |
941 | conn->js_seqno = jsonrpc_session_get_seqno(conn->js); | |
db5a066c | 942 | jsonrpc_session_set_probe_interval(js, 0); |
1b1d2e6d BP |
943 | } |
944 | ||
945 | /* Starts the local server in an existing Raft cluster, using the local copy of | |
946 | * the cluster's log in 'file_name'. Takes ownership of 'log', whether | |
947 | * successful or not. */ | |
948 | struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
949 | raft_open(struct ovsdb_log *log, struct raft **raftp) | |
950 | { | |
951 | struct raft *raft = raft_alloc(); | |
952 | raft->log = log; | |
953 | ||
954 | struct ovsdb_error *error = raft_read_header(raft); | |
955 | if (error) { | |
956 | goto error; | |
957 | } | |
958 | ||
959 | if (!raft->joining) { | |
960 | error = raft_read_log(raft); | |
961 | if (error) { | |
962 | goto error; | |
963 | } | |
964 | ||
965 | /* Find our own server. */ | |
966 | if (!raft_find_server(raft, &raft->sid)) { | |
967 | error = ovsdb_error(NULL, "server does not belong to cluster"); | |
968 | goto error; | |
969 | } | |
970 | ||
971 | /* If there's only one server, start an election right away so that the | |
972 | * cluster bootstraps quickly. */ | |
973 | if (hmap_count(&raft->servers) == 1) { | |
974 | raft_start_election(raft, false); | |
975 | } | |
976 | } else { | |
977 | raft->join_timeout = time_msec() + 1000; | |
978 | } | |
979 | ||
ac3ba8c6 HZ |
980 | raft_reset_ping_timer(raft); |
981 | raft_reset_election_timer(raft); | |
982 | ||
1b1d2e6d BP |
983 | *raftp = raft; |
984 | hmap_insert(&all_rafts, &raft->hmap_node, hash_string(raft->name, 0)); | |
985 | return NULL; | |
986 | ||
987 | error: | |
988 | raft_close(raft); | |
989 | *raftp = NULL; | |
990 | return error; | |
991 | } | |
992 | ||
993 | /* Returns the name of 'raft', which in OVSDB is the database schema name. */ | |
994 | const char * | |
995 | raft_get_name(const struct raft *raft) | |
996 | { | |
997 | return raft->name; | |
998 | } | |
999 | ||
1000 | /* Returns the cluster ID of 'raft'. If 'raft' has not yet completed joining | |
1001 | * its cluster, then 'cid' will be all-zeros (unless the administrator | |
1002 | * specified a cluster ID running "ovsdb-tool join-cluster"). | |
1003 | * | |
1004 | * Each cluster has a unique cluster ID. */ | |
1005 | const struct uuid * | |
1006 | raft_get_cid(const struct raft *raft) | |
1007 | { | |
1008 | return &raft->cid; | |
1009 | } | |
1010 | ||
1011 | /* Returns the server ID of 'raft'. Each server has a unique server ID. */ | |
1012 | const struct uuid * | |
1013 | raft_get_sid(const struct raft *raft) | |
1014 | { | |
1015 | return &raft->sid; | |
1016 | } | |
1017 | ||
3423cd97 IM |
1018 | /* Adds memory consumption info to 'usage' for later use by memory_report(). */ |
1019 | void | |
1020 | raft_get_memory_usage(const struct raft *raft, struct simap *usage) | |
1021 | { | |
1022 | struct raft_conn *conn; | |
1023 | int cnt = 0; | |
1024 | ||
1025 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1026 | simap_increase(usage, "raft-backlog", | |
1027 | jsonrpc_session_get_backlog(conn->js)); | |
1028 | cnt++; | |
1029 | } | |
1030 | simap_increase(usage, "raft-connections", cnt); | |
1031 | } | |
1032 | ||
1b1d2e6d BP |
1033 | /* Returns true if 'raft' has completed joining its cluster, has not left or |
1034 | * initiated leaving the cluster, does not have failed disk storage, and is | |
1035 | * apparently connected to the leader in a healthy way (or is itself the | |
4d9b28cb HZ |
1036 | * leader). |
1037 | * | |
1038 | * If 'raft' is candidate: | |
1039 | * a) if it is the first round of election, consider it as connected, hoping | |
1040 | * it will successfully elect a new leader soon. | |
1041 | * b) if it is already retrying, consider it as disconnected (so that clients | |
1042 | * may decide to reconnect to other members). */ | |
1b1d2e6d BP |
1043 | bool |
1044 | raft_is_connected(const struct raft *raft) | |
1045 | { | |
923f01ca | 1046 | bool ret = (!raft->candidate_retrying |
1b1d2e6d BP |
1047 | && !raft->joining |
1048 | && !raft->leaving | |
1049 | && !raft->left | |
2833885f HZ |
1050 | && !raft->failed |
1051 | && raft->ever_had_leader); | |
923f01ca HZ |
1052 | VLOG_DBG("raft_is_connected: %s\n", ret? "true": "false"); |
1053 | return ret; | |
1b1d2e6d BP |
1054 | } |
1055 | ||
1056 | /* Returns true if 'raft' is the cluster leader. */ | |
1057 | bool | |
1058 | raft_is_leader(const struct raft *raft) | |
1059 | { | |
1060 | return raft->role == RAFT_LEADER; | |
1061 | } | |
1062 | ||
1063 | /* Returns true if 'raft' is the process of joining its cluster. */ | |
1064 | bool | |
1065 | raft_is_joining(const struct raft *raft) | |
1066 | { | |
1067 | return raft->joining; | |
1068 | } | |
1069 | ||
1070 | /* Only returns *connected* connections. */ | |
1071 | static struct raft_conn * | |
1072 | raft_find_conn_by_sid(struct raft *raft, const struct uuid *sid) | |
1073 | { | |
1074 | if (!uuid_is_zero(sid)) { | |
1075 | struct raft_conn *conn; | |
1076 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1077 | if (uuid_equals(sid, &conn->sid) | |
1078 | && jsonrpc_session_is_connected(conn->js)) { | |
1079 | return conn; | |
1080 | } | |
1081 | } | |
1082 | } | |
1083 | return NULL; | |
1084 | } | |
1085 | ||
1086 | static struct raft_conn * | |
1087 | raft_find_conn_by_address(struct raft *raft, const char *address) | |
1088 | { | |
1089 | struct raft_conn *conn; | |
1090 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1091 | if (!strcmp(jsonrpc_session_get_name(conn->js), address)) { | |
1092 | return conn; | |
1093 | } | |
1094 | } | |
1095 | return NULL; | |
1096 | } | |
1097 | ||
1098 | static void OVS_PRINTF_FORMAT(3, 4) | |
1099 | raft_record_note(struct raft *raft, const char *note, | |
1100 | const char *comment_format, ...) | |
1101 | { | |
1102 | va_list args; | |
1103 | va_start(args, comment_format); | |
1104 | char *comment = xvasprintf(comment_format, args); | |
1105 | va_end(args); | |
1106 | ||
1107 | struct raft_record r = { | |
1108 | .type = RAFT_REC_NOTE, | |
1109 | .comment = comment, | |
1110 | .note = CONST_CAST(char *, note), | |
1111 | }; | |
1112 | ignore(ovsdb_log_write_and_free(raft->log, raft_record_to_json(&r))); | |
1113 | ||
1114 | free(comment); | |
1115 | } | |
1116 | ||
1117 | /* If we're leader, try to transfer leadership to another server, logging | |
1118 | * 'reason' as the human-readable reason (it should be a phrase suitable for | |
1119 | * following "because") . */ | |
1120 | void | |
1121 | raft_transfer_leadership(struct raft *raft, const char *reason) | |
1122 | { | |
1123 | if (raft->role != RAFT_LEADER) { | |
1124 | return; | |
1125 | } | |
1126 | ||
1127 | struct raft_server *s; | |
1128 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
1129 | if (!uuid_equals(&raft->sid, &s->sid) | |
1130 | && s->phase == RAFT_PHASE_STABLE) { | |
1131 | struct raft_conn *conn = raft_find_conn_by_sid(raft, &s->sid); | |
1132 | if (!conn) { | |
1133 | continue; | |
1134 | } | |
1135 | ||
1136 | union raft_rpc rpc = { | |
1137 | .become_leader = { | |
1138 | .common = { | |
1139 | .comment = CONST_CAST(char *, reason), | |
1140 | .type = RAFT_RPC_BECOME_LEADER, | |
1141 | .sid = s->sid, | |
1142 | }, | |
1143 | .term = raft->term, | |
1144 | } | |
1145 | }; | |
02acb41a | 1146 | raft_send_to_conn(raft, &rpc, conn); |
1b1d2e6d BP |
1147 | |
1148 | raft_record_note(raft, "transfer leadership", | |
1149 | "transferring leadership to %s because %s", | |
1150 | s->nickname, reason); | |
1151 | break; | |
1152 | } | |
1153 | } | |
1154 | } | |
1155 | ||
1156 | /* Send a RemoveServerRequest to the rest of the servers in the cluster. | |
1157 | * | |
1158 | * If we know which server is the leader, we can just send the request to it. | |
1159 | * However, we might not know which server is the leader, and we might never | |
1160 | * find out if the remove request was actually previously committed by a | |
1161 | * majority of the servers (because in that case the new leader will not send | |
1162 | * AppendRequests or heartbeats to us). Therefore, we instead send | |
1163 | * RemoveRequests to every server. This theoretically has the same problem, if | |
1164 | * the current cluster leader was not previously a member of the cluster, but | |
1165 | * it seems likely to be more robust in practice. */ | |
1166 | static void | |
1167 | raft_send_remove_server_requests(struct raft *raft) | |
1168 | { | |
1169 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
1170 | VLOG_INFO_RL(&rl, "sending remove request (joining=%s, leaving=%s)", | |
1171 | raft->joining ? "true" : "false", | |
1172 | raft->leaving ? "true" : "false"); | |
1173 | const struct raft_server *s; | |
1174 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
1175 | if (!uuid_equals(&s->sid, &raft->sid)) { | |
1176 | union raft_rpc rpc = (union raft_rpc) { | |
1177 | .remove_server_request = { | |
1178 | .common = { | |
1179 | .type = RAFT_RPC_REMOVE_SERVER_REQUEST, | |
1180 | .sid = s->sid, | |
1181 | }, | |
1182 | .sid = raft->sid, | |
1183 | }, | |
1184 | }; | |
1185 | raft_send(raft, &rpc); | |
1186 | } | |
1187 | } | |
1188 | ||
8e354614 | 1189 | raft->leave_timeout = time_msec() + raft->election_timer; |
1b1d2e6d BP |
1190 | } |
1191 | ||
1192 | /* Attempts to start 'raft' leaving its cluster. The caller can check progress | |
1193 | * using raft_is_leaving() and raft_left(). */ | |
1194 | void | |
1195 | raft_leave(struct raft *raft) | |
1196 | { | |
1197 | if (raft->joining || raft->failed || raft->leaving || raft->left) { | |
1198 | return; | |
1199 | } | |
1200 | VLOG_INFO(SID_FMT": starting to leave cluster "CID_FMT, | |
1201 | SID_ARGS(&raft->sid), CID_ARGS(&raft->cid)); | |
1202 | raft->leaving = true; | |
1203 | raft_transfer_leadership(raft, "this server is leaving the cluster"); | |
1204 | raft_become_follower(raft); | |
1205 | raft_send_remove_server_requests(raft); | |
8e354614 | 1206 | raft->leave_timeout = time_msec() + raft->election_timer; |
1b1d2e6d BP |
1207 | } |
1208 | ||
1209 | /* Returns true if 'raft' is currently attempting to leave its cluster. */ | |
1210 | bool | |
1211 | raft_is_leaving(const struct raft *raft) | |
1212 | { | |
1213 | return raft->leaving; | |
1214 | } | |
1215 | ||
1216 | /* Returns true if 'raft' successfully left its cluster. */ | |
1217 | bool | |
1218 | raft_left(const struct raft *raft) | |
1219 | { | |
1220 | return raft->left; | |
1221 | } | |
1222 | ||
1223 | /* Returns true if 'raft' has experienced a disk I/O failure. When this | |
1224 | * returns true, only closing and reopening 'raft' allows for recovery. */ | |
1225 | bool | |
1226 | raft_failed(const struct raft *raft) | |
1227 | { | |
1228 | return raft->failed; | |
1229 | } | |
1230 | ||
1231 | /* Forces 'raft' to attempt to take leadership of the cluster by deposing the | |
1232 | * current cluster. */ | |
1233 | void | |
1234 | raft_take_leadership(struct raft *raft) | |
1235 | { | |
1236 | if (raft->role != RAFT_LEADER) { | |
1237 | raft_start_election(raft, true); | |
1238 | } | |
1239 | } | |
1240 | ||
1241 | /* Closes everything owned by 'raft' that might be visible outside the process: | |
1242 | * network connections, commands, etc. This is part of closing 'raft'; it is | |
1243 | * also used if 'raft' has failed in an unrecoverable way. */ | |
1244 | static void | |
1245 | raft_close__(struct raft *raft) | |
1246 | { | |
1247 | if (!hmap_node_is_null(&raft->hmap_node)) { | |
1248 | hmap_remove(&all_rafts, &raft->hmap_node); | |
1249 | hmap_node_nullify(&raft->hmap_node); | |
1250 | } | |
1251 | ||
1252 | raft_complete_all_commands(raft, RAFT_CMD_SHUTDOWN); | |
1253 | ||
1254 | struct raft_server *rs = raft->remove_server; | |
1255 | if (rs) { | |
1256 | raft_send_remove_server_reply__(raft, &rs->sid, &rs->requester_sid, | |
1257 | rs->requester_conn, false, | |
1258 | RAFT_SERVER_SHUTDOWN); | |
1259 | raft_server_destroy(raft->remove_server); | |
1260 | raft->remove_server = NULL; | |
1261 | } | |
1262 | ||
1263 | struct raft_conn *conn, *next; | |
1264 | LIST_FOR_EACH_SAFE (conn, next, list_node, &raft->conns) { | |
1265 | raft_conn_close(conn); | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | /* Closes and frees 'raft'. | |
1270 | * | |
1271 | * A server's cluster membership is independent of whether the server is | |
1272 | * actually running. When a server that is a member of a cluster closes, the | |
1273 | * cluster treats this as a server failure. */ | |
1274 | void | |
1275 | raft_close(struct raft *raft) | |
1276 | { | |
1277 | if (!raft) { | |
1278 | return; | |
1279 | } | |
1280 | ||
1281 | raft_transfer_leadership(raft, "this server is shutting down"); | |
1282 | ||
1283 | raft_close__(raft); | |
1284 | ||
1285 | ovsdb_log_close(raft->log); | |
1286 | ||
1287 | raft_servers_destroy(&raft->servers); | |
1288 | ||
1289 | for (uint64_t index = raft->log_start; index < raft->log_end; index++) { | |
1290 | struct raft_entry *e = &raft->entries[index - raft->log_start]; | |
1291 | raft_entry_uninit(e); | |
1292 | } | |
1293 | free(raft->entries); | |
1294 | ||
1295 | raft_entry_uninit(&raft->snap); | |
1296 | ||
1297 | raft_waiters_destroy(raft); | |
1298 | ||
1299 | raft_servers_destroy(&raft->add_servers); | |
1300 | ||
1301 | hmap_destroy(&raft->commands); | |
1302 | ||
1303 | pstream_close(raft->listener); | |
1304 | ||
1305 | sset_destroy(&raft->remote_addresses); | |
1306 | free(raft->local_address); | |
1307 | free(raft->local_nickname); | |
1308 | free(raft->name); | |
1309 | ||
1310 | free(raft); | |
1311 | } | |
1312 | ||
1313 | static bool | |
1314 | raft_conn_receive(struct raft *raft, struct raft_conn *conn, | |
1315 | union raft_rpc *rpc) | |
1316 | { | |
1317 | struct jsonrpc_msg *msg = jsonrpc_session_recv(conn->js); | |
1318 | if (!msg) { | |
1319 | return false; | |
1320 | } | |
1321 | ||
1322 | struct ovsdb_error *error = raft_rpc_from_jsonrpc(&raft->cid, &raft->sid, | |
1323 | msg, rpc); | |
1324 | jsonrpc_msg_destroy(msg); | |
1325 | if (error) { | |
1326 | char *s = ovsdb_error_to_string_free(error); | |
1327 | VLOG_INFO("%s: %s", jsonrpc_session_get_name(conn->js), s); | |
1328 | free(s); | |
1329 | return false; | |
1330 | } | |
1331 | ||
1332 | if (uuid_is_zero(&conn->sid)) { | |
1333 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(50, 50); | |
1334 | conn->sid = rpc->common.sid; | |
1335 | VLOG_INFO_RL(&rl, "%s: learned server ID "SID_FMT, | |
1336 | jsonrpc_session_get_name(conn->js), SID_ARGS(&conn->sid)); | |
1337 | } else if (!uuid_equals(&conn->sid, &rpc->common.sid)) { | |
1338 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1339 | VLOG_WARN_RL(&rl, "%s: ignoring message with unexpected server ID " | |
1340 | SID_FMT" (expected "SID_FMT")", | |
1341 | jsonrpc_session_get_name(conn->js), | |
1342 | SID_ARGS(&rpc->common.sid), SID_ARGS(&conn->sid)); | |
1343 | raft_rpc_uninit(rpc); | |
1344 | return false; | |
1345 | } | |
1346 | ||
1347 | const char *address = (rpc->type == RAFT_RPC_HELLO_REQUEST | |
1348 | ? rpc->hello_request.address | |
1349 | : rpc->type == RAFT_RPC_ADD_SERVER_REQUEST | |
1350 | ? rpc->add_server_request.address | |
1351 | : NULL); | |
1352 | if (address) { | |
1353 | char *new_nickname = raft_address_to_nickname(address, &conn->sid); | |
1354 | if (strcmp(conn->nickname, new_nickname)) { | |
1355 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(50, 50); | |
1356 | VLOG_INFO_RL(&rl, "%s: learned remote address %s", | |
1357 | jsonrpc_session_get_name(conn->js), address); | |
1358 | ||
1359 | free(conn->nickname); | |
1360 | conn->nickname = new_nickname; | |
1361 | } else { | |
1362 | free(new_nickname); | |
1363 | } | |
1364 | } | |
1365 | ||
1366 | return true; | |
1367 | } | |
1368 | ||
1369 | static const char * | |
1370 | raft_get_nickname(const struct raft *raft, const struct uuid *sid, | |
1371 | char buf[SID_LEN + 1], size_t bufsize) | |
1372 | { | |
1373 | if (uuid_equals(sid, &raft->sid)) { | |
1374 | return raft->local_nickname; | |
1375 | } | |
1376 | ||
1377 | const char *s = raft_servers_get_nickname__(&raft->servers, sid); | |
1378 | if (s) { | |
1379 | return s; | |
1380 | } | |
1381 | ||
1382 | return raft_servers_get_nickname(&raft->add_servers, sid, buf, bufsize); | |
1383 | } | |
1384 | ||
1385 | static void | |
02acb41a BP |
1386 | log_rpc(const union raft_rpc *rpc, const char *direction, |
1387 | const struct raft_conn *conn, int line_number) | |
1b1d2e6d BP |
1388 | { |
1389 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(600, 600); | |
1390 | if (!raft_rpc_is_heartbeat(rpc) && !VLOG_DROP_DBG(&rl)) { | |
1391 | struct ds s = DS_EMPTY_INITIALIZER; | |
02acb41a BP |
1392 | if (line_number) { |
1393 | ds_put_format(&s, "raft.c:%d ", line_number); | |
1394 | } | |
1395 | ds_put_format(&s, "%s%s ", direction, conn->nickname); | |
1b1d2e6d | 1396 | raft_rpc_format(rpc, &s); |
02acb41a | 1397 | VLOG_DBG("%s", ds_cstr(&s)); |
1b1d2e6d BP |
1398 | ds_destroy(&s); |
1399 | } | |
1400 | } | |
1401 | ||
1402 | static void | |
1403 | raft_send_add_server_request(struct raft *raft, struct raft_conn *conn) | |
1404 | { | |
1405 | union raft_rpc rq = { | |
1406 | .add_server_request = { | |
1407 | .common = { | |
1408 | .type = RAFT_RPC_ADD_SERVER_REQUEST, | |
1409 | .sid = UUID_ZERO, | |
1410 | .comment = NULL, | |
1411 | }, | |
1412 | .address = raft->local_address, | |
1413 | }, | |
1414 | }; | |
02acb41a | 1415 | raft_send_to_conn(raft, &rq, conn); |
1b1d2e6d BP |
1416 | } |
1417 | ||
1418 | static void | |
1419 | raft_conn_run(struct raft *raft, struct raft_conn *conn) | |
1420 | { | |
1421 | jsonrpc_session_run(conn->js); | |
1422 | ||
1423 | unsigned int new_seqno = jsonrpc_session_get_seqno(conn->js); | |
8c2c503b IM |
1424 | bool reconnected = new_seqno != conn->js_seqno; |
1425 | bool just_connected = (reconnected | |
1b1d2e6d | 1426 | && jsonrpc_session_is_connected(conn->js)); |
8c2c503b IM |
1427 | |
1428 | if (reconnected) { | |
1429 | /* Clear 'last_install_snapshot_request' since it might not reach the | |
1430 | * destination or server was restarted. */ | |
1431 | struct raft_server *server = raft_find_server(raft, &conn->sid); | |
1432 | if (server) { | |
1433 | free(server->last_install_snapshot_request); | |
1434 | server->last_install_snapshot_request = NULL; | |
1435 | } | |
1436 | } | |
1437 | ||
1b1d2e6d BP |
1438 | conn->js_seqno = new_seqno; |
1439 | if (just_connected) { | |
1440 | if (raft->joining) { | |
1441 | raft_send_add_server_request(raft, conn); | |
1442 | } else if (raft->leaving) { | |
1443 | union raft_rpc rq = { | |
1444 | .remove_server_request = { | |
1445 | .common = { | |
1446 | .type = RAFT_RPC_REMOVE_SERVER_REQUEST, | |
1447 | .sid = conn->sid, | |
1448 | }, | |
1449 | .sid = raft->sid, | |
1450 | }, | |
1451 | }; | |
02acb41a | 1452 | raft_send_to_conn(raft, &rq, conn); |
1b1d2e6d BP |
1453 | } else { |
1454 | union raft_rpc rq = (union raft_rpc) { | |
1455 | .hello_request = { | |
1456 | .common = { | |
1457 | .type = RAFT_RPC_HELLO_REQUEST, | |
1458 | .sid = conn->sid, | |
1459 | }, | |
1460 | .address = raft->local_address, | |
1461 | }, | |
1462 | }; | |
02acb41a | 1463 | raft_send_to_conn(raft, &rq, conn); |
1b1d2e6d BP |
1464 | } |
1465 | } | |
1466 | ||
1467 | for (size_t i = 0; i < 50; i++) { | |
1468 | union raft_rpc rpc; | |
1469 | if (!raft_conn_receive(raft, conn, &rpc)) { | |
1470 | break; | |
1471 | } | |
1472 | ||
02acb41a | 1473 | log_rpc(&rpc, "<--", conn, 0); |
1b1d2e6d BP |
1474 | raft_handle_rpc(raft, &rpc); |
1475 | raft_rpc_uninit(&rpc); | |
1476 | } | |
1477 | } | |
1478 | ||
1479 | static void | |
1480 | raft_waiter_complete_rpc(struct raft *raft, const union raft_rpc *rpc) | |
1481 | { | |
1482 | uint64_t term = raft_rpc_get_term(rpc); | |
1483 | if (term && term < raft->term) { | |
1484 | /* Drop the message because it's for an expired term. */ | |
1485 | return; | |
1486 | } | |
1487 | ||
1488 | if (!raft_is_rpc_synced(raft, rpc)) { | |
1489 | /* This is a bug. A reply message is deferred because some state in | |
1490 | * the message, such as a term or index, has not been committed to | |
1491 | * disk, and they should only be completed when that commit is done. | |
1492 | * But this message is being completed before the commit is finished. | |
1493 | * Complain, and hope that someone reports the bug. */ | |
1494 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
1495 | if (VLOG_DROP_ERR(&rl)) { | |
1496 | return; | |
1497 | } | |
1498 | ||
1499 | struct ds s = DS_EMPTY_INITIALIZER; | |
1500 | ||
1501 | if (term > raft->synced_term) { | |
1502 | ds_put_format(&s, " because message term %"PRIu64" is " | |
1503 | "past synced term %"PRIu64, | |
1504 | term, raft->synced_term); | |
1505 | } | |
1506 | ||
1507 | uint64_t index = raft_rpc_get_min_sync_index(rpc); | |
1508 | if (index > raft->log_synced) { | |
1509 | ds_put_format(&s, " %s message index %"PRIu64" is past last " | |
1510 | "synced index %"PRIu64, | |
1511 | s.length ? "and" : "because", | |
1512 | index, raft->log_synced); | |
1513 | } | |
1514 | ||
1515 | const struct uuid *vote = raft_rpc_get_vote(rpc); | |
1516 | if (vote && !uuid_equals(vote, &raft->synced_vote)) { | |
1517 | char buf1[SID_LEN + 1]; | |
1518 | char buf2[SID_LEN + 1]; | |
1519 | ds_put_format(&s, " %s vote %s differs from synced vote %s", | |
1520 | s.length ? "and" : "because", | |
1521 | raft_get_nickname(raft, vote, buf1, sizeof buf1), | |
1522 | raft_get_nickname(raft, &raft->synced_vote, | |
1523 | buf2, sizeof buf2)); | |
1524 | } | |
1525 | ||
1526 | char buf[SID_LEN + 1]; | |
1527 | ds_put_format(&s, ": %s ", | |
1528 | raft_get_nickname(raft, &rpc->common.sid, | |
1529 | buf, sizeof buf)); | |
1530 | raft_rpc_format(rpc, &s); | |
1531 | VLOG_ERR("internal error: deferred %s message completed " | |
1532 | "but not ready to send%s", | |
1533 | raft_rpc_type_to_string(rpc->type), ds_cstr(&s)); | |
1534 | ds_destroy(&s); | |
1535 | ||
1536 | return; | |
1537 | } | |
1538 | ||
1539 | struct raft_conn *dst = raft_find_conn_by_sid(raft, &rpc->common.sid); | |
1540 | if (dst) { | |
02acb41a | 1541 | raft_send_to_conn(raft, rpc, dst); |
1b1d2e6d BP |
1542 | } |
1543 | } | |
1544 | ||
1545 | static void | |
1546 | raft_waiter_complete(struct raft *raft, struct raft_waiter *w) | |
1547 | { | |
1548 | switch (w->type) { | |
1549 | case RAFT_W_ENTRY: | |
1550 | if (raft->role == RAFT_LEADER) { | |
1551 | raft_update_our_match_index(raft, w->entry.index); | |
1552 | } | |
1553 | raft->log_synced = w->entry.index; | |
1554 | break; | |
1555 | ||
1556 | case RAFT_W_TERM: | |
1557 | raft->synced_term = w->term.term; | |
1558 | raft->synced_vote = w->term.vote; | |
1559 | break; | |
1560 | ||
1561 | case RAFT_W_RPC: | |
1562 | raft_waiter_complete_rpc(raft, w->rpc); | |
1563 | break; | |
1564 | } | |
1565 | } | |
1566 | ||
1567 | static void | |
1568 | raft_waiter_destroy(struct raft_waiter *w) | |
1569 | { | |
1570 | if (!w) { | |
1571 | return; | |
1572 | } | |
1573 | ||
1574 | ovs_list_remove(&w->list_node); | |
1575 | ||
1576 | switch (w->type) { | |
1577 | case RAFT_W_ENTRY: | |
1578 | case RAFT_W_TERM: | |
1579 | break; | |
1580 | ||
1581 | case RAFT_W_RPC: | |
1582 | raft_rpc_uninit(w->rpc); | |
1583 | free(w->rpc); | |
1584 | break; | |
1585 | } | |
1586 | free(w); | |
1587 | } | |
1588 | ||
1589 | static void | |
1590 | raft_waiters_run(struct raft *raft) | |
1591 | { | |
1592 | if (ovs_list_is_empty(&raft->waiters)) { | |
1593 | return; | |
1594 | } | |
1595 | ||
1596 | uint64_t cur = ovsdb_log_commit_progress(raft->log); | |
1597 | struct raft_waiter *w, *next; | |
1598 | LIST_FOR_EACH_SAFE (w, next, list_node, &raft->waiters) { | |
1599 | if (cur < w->commit_ticket) { | |
1600 | break; | |
1601 | } | |
1602 | raft_waiter_complete(raft, w); | |
1603 | raft_waiter_destroy(w); | |
1604 | } | |
1605 | } | |
1606 | ||
1607 | static void | |
1608 | raft_waiters_wait(struct raft *raft) | |
1609 | { | |
1610 | struct raft_waiter *w; | |
1611 | LIST_FOR_EACH (w, list_node, &raft->waiters) { | |
1612 | ovsdb_log_commit_wait(raft->log, w->commit_ticket); | |
1613 | break; | |
1614 | } | |
1615 | } | |
1616 | ||
1617 | static void | |
1618 | raft_waiters_destroy(struct raft *raft) | |
1619 | { | |
1620 | struct raft_waiter *w, *next; | |
1621 | LIST_FOR_EACH_SAFE (w, next, list_node, &raft->waiters) { | |
1622 | raft_waiter_destroy(w); | |
1623 | } | |
1624 | } | |
1625 | ||
1626 | static bool OVS_WARN_UNUSED_RESULT | |
1627 | raft_set_term(struct raft *raft, uint64_t term, const struct uuid *vote) | |
1628 | { | |
1629 | struct ovsdb_error *error = raft_write_state(raft->log, term, vote); | |
1630 | if (!raft_handle_write_error(raft, error)) { | |
1631 | return false; | |
1632 | } | |
1633 | ||
1634 | struct raft_waiter *w = raft_waiter_create(raft, RAFT_W_TERM, true); | |
1635 | raft->term = w->term.term = term; | |
1636 | raft->vote = w->term.vote = vote ? *vote : UUID_ZERO; | |
1637 | return true; | |
1638 | } | |
1639 | ||
1640 | static void | |
1641 | raft_accept_vote(struct raft *raft, struct raft_server *s, | |
1642 | const struct uuid *vote) | |
1643 | { | |
1644 | if (uuid_equals(&s->vote, vote)) { | |
1645 | return; | |
1646 | } | |
1647 | if (!uuid_is_zero(&s->vote)) { | |
1648 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
1649 | char buf1[SID_LEN + 1]; | |
1650 | char buf2[SID_LEN + 1]; | |
1651 | VLOG_WARN_RL(&rl, "server %s changed its vote from %s to %s", | |
1652 | s->nickname, | |
1653 | raft_get_nickname(raft, &s->vote, buf1, sizeof buf1), | |
1654 | raft_get_nickname(raft, vote, buf2, sizeof buf2)); | |
1655 | } | |
1656 | s->vote = *vote; | |
1657 | if (uuid_equals(vote, &raft->sid) | |
1658 | && ++raft->n_votes > hmap_count(&raft->servers) / 2) { | |
1659 | raft_become_leader(raft); | |
1660 | } | |
1661 | } | |
1662 | ||
1663 | static void | |
1664 | raft_start_election(struct raft *raft, bool leadership_transfer) | |
1665 | { | |
1666 | if (raft->leaving) { | |
1667 | return; | |
1668 | } | |
1669 | ||
1670 | struct raft_server *me = raft_find_server(raft, &raft->sid); | |
1671 | if (!me) { | |
1672 | return; | |
1673 | } | |
1674 | ||
1675 | if (!raft_set_term(raft, raft->term + 1, &raft->sid)) { | |
1676 | return; | |
1677 | } | |
1678 | ||
1b1d2e6d | 1679 | ovs_assert(raft->role != RAFT_LEADER); |
93ee4209 | 1680 | |
cdae6100 | 1681 | raft->leader_sid = UUID_ZERO; |
1b1d2e6d | 1682 | raft->role = RAFT_CANDIDATE; |
923f01ca HZ |
1683 | /* If there was no leader elected since last election, we know we are |
1684 | * retrying now. */ | |
1685 | raft->candidate_retrying = !raft->had_leader; | |
1686 | raft->had_leader = false; | |
1b1d2e6d BP |
1687 | |
1688 | raft->n_votes = 0; | |
1689 | ||
1690 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1691 | if (!VLOG_DROP_INFO(&rl)) { | |
1692 | long long int now = time_msec(); | |
1693 | if (now >= raft->election_timeout) { | |
1694 | VLOG_INFO("term %"PRIu64": %lld ms timeout expired, " | |
1695 | "starting election", | |
1696 | raft->term, now - raft->election_base); | |
1697 | } else { | |
1698 | VLOG_INFO("term %"PRIu64": starting election", raft->term); | |
1699 | } | |
1700 | } | |
0f954f32 | 1701 | raft_reset_election_timer(raft); |
1b1d2e6d BP |
1702 | |
1703 | struct raft_server *peer; | |
1704 | HMAP_FOR_EACH (peer, hmap_node, &raft->servers) { | |
1705 | peer->vote = UUID_ZERO; | |
1706 | if (uuid_equals(&raft->sid, &peer->sid)) { | |
1707 | continue; | |
1708 | } | |
1709 | ||
1710 | union raft_rpc rq = { | |
1711 | .vote_request = { | |
1712 | .common = { | |
1713 | .type = RAFT_RPC_VOTE_REQUEST, | |
1714 | .sid = peer->sid, | |
1715 | }, | |
1716 | .term = raft->term, | |
1717 | .last_log_index = raft->log_end - 1, | |
1718 | .last_log_term = ( | |
1719 | raft->log_end > raft->log_start | |
1720 | ? raft->entries[raft->log_end - raft->log_start - 1].term | |
1721 | : raft->snap.term), | |
1722 | .leadership_transfer = leadership_transfer, | |
1723 | }, | |
1724 | }; | |
93ee4209 HZ |
1725 | if (failure_test != FT_DONT_SEND_VOTE_REQUEST) { |
1726 | raft_send(raft, &rq); | |
1727 | } | |
1b1d2e6d BP |
1728 | } |
1729 | ||
1730 | /* Vote for ourselves. */ | |
1731 | raft_accept_vote(raft, me, &raft->sid); | |
1732 | } | |
1733 | ||
1734 | static void | |
1735 | raft_open_conn(struct raft *raft, const char *address, const struct uuid *sid) | |
1736 | { | |
1737 | if (strcmp(address, raft->local_address) | |
1738 | && !raft_find_conn_by_address(raft, address)) { | |
1739 | raft_add_conn(raft, jsonrpc_session_open(address, true), sid, false); | |
1740 | } | |
1741 | } | |
1742 | ||
1743 | static void | |
1744 | raft_conn_close(struct raft_conn *conn) | |
1745 | { | |
1746 | jsonrpc_session_close(conn->js); | |
1747 | ovs_list_remove(&conn->list_node); | |
1748 | free(conn->nickname); | |
1749 | free(conn); | |
1750 | } | |
1751 | ||
1752 | /* Returns true if 'conn' should stay open, false if it should be closed. */ | |
1753 | static bool | |
1754 | raft_conn_should_stay_open(struct raft *raft, struct raft_conn *conn) | |
1755 | { | |
1756 | /* Close the connection if it's actually dead. If necessary, we'll | |
1757 | * initiate a new session later. */ | |
1758 | if (!jsonrpc_session_is_alive(conn->js)) { | |
1759 | return false; | |
1760 | } | |
1761 | ||
1762 | /* Keep incoming sessions. We trust the originator to decide to drop | |
1763 | * it. */ | |
1764 | if (conn->incoming) { | |
1765 | return true; | |
1766 | } | |
1767 | ||
1768 | /* If we are joining the cluster, keep sessions to the remote addresses | |
1769 | * that are supposed to be part of the cluster we're joining. */ | |
1770 | if (raft->joining && sset_contains(&raft->remote_addresses, | |
1771 | jsonrpc_session_get_name(conn->js))) { | |
1772 | return true; | |
1773 | } | |
1774 | ||
1775 | /* We have joined the cluster. If we did that "recently", then there is a | |
1776 | * chance that we do not have the most recent server configuration log | |
1777 | * entry. If so, it's a waste to disconnect from the servers that were in | |
1778 | * remote_addresses and that will probably appear in the configuration, | |
1779 | * just to reconnect to them a moment later when we do get the | |
1780 | * configuration update. If we are not ourselves in the configuration, | |
1781 | * then we know that there must be a new configuration coming up, so in | |
1782 | * that case keep the connection. */ | |
1783 | if (!raft_find_server(raft, &raft->sid)) { | |
1784 | return true; | |
1785 | } | |
1786 | ||
1787 | /* Keep the connection only if the server is part of the configuration. */ | |
1788 | return raft_find_server(raft, &conn->sid); | |
1789 | } | |
1790 | ||
1791 | /* Allows 'raft' to maintain the distributed log. Call this function as part | |
1792 | * of the process's main loop. */ | |
1793 | void | |
1794 | raft_run(struct raft *raft) | |
1795 | { | |
1796 | if (raft->left || raft->failed) { | |
1797 | return; | |
1798 | } | |
1799 | ||
1800 | raft_waiters_run(raft); | |
1801 | ||
1802 | if (!raft->listener && time_msec() >= raft->listen_backoff) { | |
1803 | char *paddr = raft_make_address_passive(raft->local_address); | |
1804 | int error = pstream_open(paddr, &raft->listener, DSCP_DEFAULT); | |
1805 | if (error) { | |
1806 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
1807 | VLOG_WARN_RL(&rl, "%s: listen failed (%s)", | |
1808 | paddr, ovs_strerror(error)); | |
1809 | raft->listen_backoff = time_msec() + 1000; | |
1810 | } | |
1811 | free(paddr); | |
1812 | } | |
1813 | ||
1814 | if (raft->listener) { | |
1815 | struct stream *stream; | |
1816 | int error = pstream_accept(raft->listener, &stream); | |
1817 | if (!error) { | |
1818 | raft_add_conn(raft, jsonrpc_session_open_unreliably( | |
1819 | jsonrpc_open(stream), DSCP_DEFAULT), NULL, | |
1820 | true); | |
1821 | } else if (error != EAGAIN) { | |
1822 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
1823 | VLOG_WARN_RL(&rl, "%s: accept failed: %s", | |
1824 | pstream_get_name(raft->listener), | |
1825 | ovs_strerror(error)); | |
1826 | } | |
1827 | } | |
1828 | ||
1829 | /* Run RPCs for all open sessions. */ | |
1830 | struct raft_conn *conn; | |
1831 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1832 | raft_conn_run(raft, conn); | |
1833 | } | |
1834 | ||
1835 | /* Close unneeded sessions. */ | |
1836 | struct raft_conn *next; | |
1837 | LIST_FOR_EACH_SAFE (conn, next, list_node, &raft->conns) { | |
1838 | if (!raft_conn_should_stay_open(raft, conn)) { | |
1839 | raft_conn_close(conn); | |
1840 | } | |
1841 | } | |
1842 | ||
1843 | /* Open needed sessions. */ | |
1844 | struct raft_server *server; | |
1845 | HMAP_FOR_EACH (server, hmap_node, &raft->servers) { | |
1846 | raft_open_conn(raft, server->address, &server->sid); | |
1847 | } | |
1848 | if (raft->joining) { | |
1849 | const char *address; | |
1850 | SSET_FOR_EACH (address, &raft->remote_addresses) { | |
1851 | raft_open_conn(raft, address, NULL); | |
1852 | } | |
1853 | } | |
1854 | ||
1855 | if (!raft->joining && time_msec() >= raft->election_timeout) { | |
89771c1e HZ |
1856 | if (raft->role == RAFT_LEADER) { |
1857 | /* Check if majority of followers replied, then reset | |
1858 | * election_timeout and reset s->replied. Otherwise, become | |
1859 | * follower. | |
1860 | * | |
1861 | * Raft paper section 6.2: Leaders: A server might be in the leader | |
1862 | * state, but if it isn’t the current leader, it could be | |
1863 | * needlessly delaying client requests. For example, suppose a | |
1864 | * leader is partitioned from the rest of the cluster, but it can | |
1865 | * still communicate with a particular client. Without additional | |
1866 | * mechanism, it could delay a request from that client forever, | |
1867 | * being unable to replicate a log entry to any other servers. | |
1868 | * Meanwhile, there might be another leader of a newer term that is | |
1869 | * able to communicate with a majority of the cluster and would be | |
1870 | * able to commit the client’s request. Thus, a leader in Raft | |
1871 | * steps down if an election timeout elapses without a successful | |
1872 | * round of heartbeats to a majority of its cluster; this allows | |
1873 | * clients to retry their requests with another server. */ | |
1874 | int count = 0; | |
1875 | HMAP_FOR_EACH (server, hmap_node, &raft->servers) { | |
1876 | if (server->replied) { | |
1877 | count ++; | |
1878 | } | |
1879 | } | |
1880 | if (count >= hmap_count(&raft->servers) / 2) { | |
1881 | HMAP_FOR_EACH (server, hmap_node, &raft->servers) { | |
1882 | server->replied = false; | |
1883 | } | |
1884 | raft_reset_election_timer(raft); | |
1885 | } else { | |
1886 | raft_become_follower(raft); | |
1887 | raft_start_election(raft, false); | |
1888 | } | |
1889 | } else { | |
1890 | raft_start_election(raft, false); | |
1891 | } | |
1892 | ||
1b1d2e6d BP |
1893 | } |
1894 | ||
1895 | if (raft->leaving && time_msec() >= raft->leave_timeout) { | |
1896 | raft_send_remove_server_requests(raft); | |
1897 | } | |
1898 | ||
1899 | if (raft->joining && time_msec() >= raft->join_timeout) { | |
1900 | raft->join_timeout = time_msec() + 1000; | |
1901 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1902 | raft_send_add_server_request(raft, conn); | |
1903 | } | |
1904 | } | |
1905 | ||
5a9b53a5 HZ |
1906 | long long int now = time_msec(); |
1907 | if (now >= raft->ping_timeout) { | |
1b1d2e6d BP |
1908 | if (raft->role == RAFT_LEADER) { |
1909 | raft_send_heartbeats(raft); | |
5a9b53a5 HZ |
1910 | } |
1911 | /* Check if any commands timeout. Timeout is set to twice the time of | |
1912 | * election base time so that commands can complete properly during | |
1913 | * leader election. E.g. a leader crashed and current node with pending | |
1914 | * commands becomes new leader: the pending commands can still complete | |
1915 | * if the crashed leader has replicated the transactions to majority of | |
1916 | * followers before it crashed. */ | |
1917 | struct raft_command *cmd, *next_cmd; | |
1918 | HMAP_FOR_EACH_SAFE (cmd, next_cmd, hmap_node, &raft->commands) { | |
1919 | if (cmd->timestamp | |
8e354614 | 1920 | && now - cmd->timestamp > raft->election_timer * 2) { |
5a9b53a5 | 1921 | raft_command_complete(raft, cmd, RAFT_CMD_TIMEOUT); |
1b1d2e6d BP |
1922 | } |
1923 | } | |
f4bef8b3 | 1924 | raft_reset_ping_timer(raft); |
1b1d2e6d BP |
1925 | } |
1926 | ||
1927 | /* Do this only at the end; if we did it as soon as we set raft->left or | |
1928 | * raft->failed in handling the RemoveServerReply, then it could easily | |
1929 | * cause references to freed memory in RPC sessions, etc. */ | |
1930 | if (raft->left || raft->failed) { | |
1931 | raft_close__(raft); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | static void | |
1936 | raft_wait_session(struct jsonrpc_session *js) | |
1937 | { | |
1938 | if (js) { | |
1939 | jsonrpc_session_wait(js); | |
1940 | jsonrpc_session_recv_wait(js); | |
1941 | } | |
1942 | } | |
1943 | ||
1944 | /* Causes the next call to poll_block() to wake up when 'raft' needs to do | |
1945 | * something. */ | |
1946 | void | |
1947 | raft_wait(struct raft *raft) | |
1948 | { | |
1949 | if (raft->left || raft->failed) { | |
1950 | return; | |
1951 | } | |
1952 | ||
1953 | raft_waiters_wait(raft); | |
1954 | ||
1955 | if (raft->listener) { | |
1956 | pstream_wait(raft->listener); | |
1957 | } else { | |
1958 | poll_timer_wait_until(raft->listen_backoff); | |
1959 | } | |
1960 | ||
1961 | struct raft_conn *conn; | |
1962 | LIST_FOR_EACH (conn, list_node, &raft->conns) { | |
1963 | raft_wait_session(conn->js); | |
1964 | } | |
1965 | ||
1966 | if (!raft->joining) { | |
1967 | poll_timer_wait_until(raft->election_timeout); | |
1968 | } else { | |
1969 | poll_timer_wait_until(raft->join_timeout); | |
1970 | } | |
1971 | if (raft->leaving) { | |
1972 | poll_timer_wait_until(raft->leave_timeout); | |
1973 | } | |
1974 | if (raft->role == RAFT_LEADER || !hmap_is_empty(&raft->commands)) { | |
1975 | poll_timer_wait_until(raft->ping_timeout); | |
1976 | } | |
1977 | } | |
1978 | ||
1979 | static struct raft_waiter * | |
1980 | raft_waiter_create(struct raft *raft, enum raft_waiter_type type, | |
1981 | bool start_commit) | |
1982 | { | |
1983 | struct raft_waiter *w = xzalloc(sizeof *w); | |
1984 | ovs_list_push_back(&raft->waiters, &w->list_node); | |
1985 | w->commit_ticket = start_commit ? ovsdb_log_commit_start(raft->log) : 0; | |
1986 | w->type = type; | |
1987 | return w; | |
1988 | } | |
1989 | ||
1990 | /* Returns a human-readable representation of 'status' (or NULL if 'status' is | |
1991 | * invalid). */ | |
1992 | const char * | |
1993 | raft_command_status_to_string(enum raft_command_status status) | |
1994 | { | |
1995 | switch (status) { | |
1996 | case RAFT_CMD_INCOMPLETE: | |
1997 | return "operation still in progress"; | |
1998 | case RAFT_CMD_SUCCESS: | |
1999 | return "success"; | |
2000 | case RAFT_CMD_NOT_LEADER: | |
2001 | return "not leader"; | |
2002 | case RAFT_CMD_BAD_PREREQ: | |
2003 | return "prerequisite check failed"; | |
2004 | case RAFT_CMD_LOST_LEADERSHIP: | |
2005 | return "lost leadership"; | |
2006 | case RAFT_CMD_SHUTDOWN: | |
2007 | return "server shutdown"; | |
2008 | case RAFT_CMD_IO_ERROR: | |
2009 | return "I/O error"; | |
2010 | case RAFT_CMD_TIMEOUT: | |
2011 | return "timeout"; | |
2012 | default: | |
2013 | return NULL; | |
2014 | } | |
2015 | } | |
2016 | ||
2017 | /* Converts human-readable status in 's' into status code in '*statusp'. | |
2018 | * Returns true if successful, false if 's' is unknown. */ | |
2019 | bool | |
2020 | raft_command_status_from_string(const char *s, | |
2021 | enum raft_command_status *statusp) | |
2022 | { | |
2023 | for (enum raft_command_status status = 0; ; status++) { | |
2024 | const char *s2 = raft_command_status_to_string(status); | |
2025 | if (!s2) { | |
2026 | *statusp = 0; | |
2027 | return false; | |
2028 | } else if (!strcmp(s, s2)) { | |
2029 | *statusp = status; | |
2030 | return true; | |
2031 | } | |
2032 | } | |
2033 | } | |
2034 | ||
2035 | static const struct uuid * | |
2036 | raft_get_eid(const struct raft *raft, uint64_t index) | |
2037 | { | |
2038 | for (; index >= raft->log_start; index--) { | |
2039 | const struct raft_entry *e = raft_get_entry(raft, index); | |
2040 | if (e->data) { | |
2041 | return &e->eid; | |
2042 | } | |
2043 | } | |
2044 | return &raft->snap.eid; | |
2045 | } | |
2046 | ||
2cd62f75 | 2047 | const struct uuid * |
1b1d2e6d BP |
2048 | raft_current_eid(const struct raft *raft) |
2049 | { | |
2050 | return raft_get_eid(raft, raft->log_end - 1); | |
2051 | } | |
2052 | ||
2053 | static struct raft_command * | |
2054 | raft_command_create_completed(enum raft_command_status status) | |
2055 | { | |
2056 | ovs_assert(status != RAFT_CMD_INCOMPLETE); | |
2057 | ||
2058 | struct raft_command *cmd = xzalloc(sizeof *cmd); | |
2059 | cmd->n_refs = 1; | |
2060 | cmd->status = status; | |
2061 | return cmd; | |
2062 | } | |
2063 | ||
2064 | static struct raft_command * | |
2065 | raft_command_create_incomplete(struct raft *raft, uint64_t index) | |
2066 | { | |
2067 | struct raft_command *cmd = xzalloc(sizeof *cmd); | |
2068 | cmd->n_refs = 2; /* One for client, one for raft->commands. */ | |
2069 | cmd->index = index; | |
2070 | cmd->status = RAFT_CMD_INCOMPLETE; | |
2071 | hmap_insert(&raft->commands, &cmd->hmap_node, cmd->index); | |
2072 | return cmd; | |
2073 | } | |
2074 | ||
2075 | static struct raft_command * OVS_WARN_UNUSED_RESULT | |
2076 | raft_command_initiate(struct raft *raft, | |
2077 | const struct json *data, const struct json *servers, | |
8e354614 | 2078 | uint64_t election_timer, const struct uuid *eid) |
1b1d2e6d BP |
2079 | { |
2080 | /* Write to local log. */ | |
2081 | uint64_t index = raft->log_end; | |
2082 | if (!raft_handle_write_error( | |
2083 | raft, raft_write_entry( | |
2084 | raft, raft->term, json_nullable_clone(data), eid, | |
8e354614 HZ |
2085 | json_nullable_clone(servers), |
2086 | election_timer))) { | |
1b1d2e6d BP |
2087 | return raft_command_create_completed(RAFT_CMD_IO_ERROR); |
2088 | } | |
2089 | ||
2090 | struct raft_command *cmd = raft_command_create_incomplete(raft, index); | |
7ef36089 HZ |
2091 | ovs_assert(eid); |
2092 | cmd->eid = *eid; | |
5a9b53a5 | 2093 | cmd->timestamp = time_msec(); |
1b1d2e6d BP |
2094 | |
2095 | raft_waiter_create(raft, RAFT_W_ENTRY, true)->entry.index = cmd->index; | |
2096 | ||
67dba070 HZ |
2097 | if (failure_test == FT_CRASH_BEFORE_SEND_APPEND_REQ) { |
2098 | ovs_fatal(0, "Raft test: crash before sending append_request."); | |
2099 | } | |
1b1d2e6d BP |
2100 | /* Write to remote logs. */ |
2101 | struct raft_server *s; | |
2102 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
2103 | if (!uuid_equals(&s->sid, &raft->sid) && s->next_index == index) { | |
2104 | raft_send_append_request(raft, s, 1, "execute command"); | |
2105 | s->next_index++; | |
2106 | } | |
2107 | } | |
67dba070 HZ |
2108 | if (failure_test == FT_CRASH_AFTER_SEND_APPEND_REQ) { |
2109 | ovs_fatal(0, "Raft test: crash after sending append_request."); | |
2110 | } | |
0f954f32 | 2111 | raft_reset_ping_timer(raft); |
1b1d2e6d BP |
2112 | |
2113 | return cmd; | |
2114 | } | |
2115 | ||
5a9b53a5 HZ |
2116 | static void |
2117 | log_all_commands(struct raft *raft) | |
2118 | { | |
2119 | struct raft_command *cmd, *next; | |
2120 | HMAP_FOR_EACH_SAFE (cmd, next, hmap_node, &raft->commands) { | |
2121 | VLOG_DBG("raft command eid: "UUID_FMT, UUID_ARGS(&cmd->eid)); | |
2122 | } | |
2123 | } | |
2124 | ||
1b1d2e6d | 2125 | static struct raft_command * OVS_WARN_UNUSED_RESULT |
8e354614 HZ |
2126 | raft_command_execute__(struct raft *raft, const struct json *data, |
2127 | const struct json *servers, uint64_t election_timer, | |
1b1d2e6d BP |
2128 | const struct uuid *prereq, struct uuid *result) |
2129 | { | |
2130 | if (raft->joining || raft->leaving || raft->left || raft->failed) { | |
2131 | return raft_command_create_completed(RAFT_CMD_SHUTDOWN); | |
2132 | } | |
2133 | ||
2134 | if (raft->role != RAFT_LEADER) { | |
2135 | /* Consider proxying the command to the leader. We can only do that if | |
2136 | * we know the leader and the command does not change the set of | |
2137 | * servers. We do not proxy commands without prerequisites, even | |
2138 | * though we could, because in an OVSDB context a log entry doesn't | |
2139 | * make sense without context. */ | |
8e354614 | 2140 | if (servers || election_timer || !data |
1b1d2e6d BP |
2141 | || raft->role != RAFT_FOLLOWER || uuid_is_zero(&raft->leader_sid) |
2142 | || !prereq) { | |
2143 | return raft_command_create_completed(RAFT_CMD_NOT_LEADER); | |
2144 | } | |
2145 | } | |
2146 | ||
2147 | struct uuid eid = data ? uuid_random() : UUID_ZERO; | |
2148 | if (result) { | |
2149 | *result = eid; | |
2150 | } | |
2151 | ||
2152 | if (raft->role != RAFT_LEADER) { | |
2153 | const union raft_rpc rpc = { | |
2154 | .execute_command_request = { | |
2155 | .common = { | |
2156 | .type = RAFT_RPC_EXECUTE_COMMAND_REQUEST, | |
2157 | .sid = raft->leader_sid, | |
2158 | }, | |
2159 | .data = CONST_CAST(struct json *, data), | |
2160 | .prereq = *prereq, | |
2161 | .result = eid, | |
2162 | } | |
2163 | }; | |
67dba070 HZ |
2164 | if (failure_test == FT_CRASH_BEFORE_SEND_EXEC_REQ) { |
2165 | ovs_fatal(0, "Raft test: crash before sending " | |
2166 | "execute_command_request"); | |
2167 | } | |
1b1d2e6d BP |
2168 | if (!raft_send(raft, &rpc)) { |
2169 | /* Couldn't send command, so it definitely failed. */ | |
2170 | return raft_command_create_completed(RAFT_CMD_NOT_LEADER); | |
2171 | } | |
67dba070 HZ |
2172 | if (failure_test == FT_CRASH_AFTER_SEND_EXEC_REQ) { |
2173 | ovs_fatal(0, "Raft test: crash after sending " | |
2174 | "execute_command_request"); | |
2175 | } | |
1b1d2e6d BP |
2176 | |
2177 | struct raft_command *cmd = raft_command_create_incomplete(raft, 0); | |
2178 | cmd->timestamp = time_msec(); | |
2179 | cmd->eid = eid; | |
5a9b53a5 | 2180 | log_all_commands(raft); |
1b1d2e6d BP |
2181 | return cmd; |
2182 | } | |
2183 | ||
2184 | const struct uuid *current_eid = raft_current_eid(raft); | |
2185 | if (prereq && !uuid_equals(prereq, current_eid)) { | |
2186 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
2187 | VLOG_INFO_RL(&rl, "current entry eid "UUID_FMT" does not match " | |
2188 | "prerequisite "UUID_FMT, | |
2189 | UUID_ARGS(current_eid), UUID_ARGS(prereq)); | |
2190 | return raft_command_create_completed(RAFT_CMD_BAD_PREREQ); | |
2191 | } | |
2192 | ||
8e354614 | 2193 | return raft_command_initiate(raft, data, servers, election_timer, &eid); |
1b1d2e6d BP |
2194 | } |
2195 | ||
2196 | /* Initiates appending a log entry to 'raft'. The log entry consists of 'data' | |
2197 | * and, if 'prereq' is nonnull, it is only added to the log if the previous | |
2198 | * entry in the log has entry ID 'prereq'. If 'result' is nonnull, it is | |
2199 | * populated with the entry ID for the new log entry. | |
2200 | * | |
2201 | * Returns a "struct raft_command" that may be used to track progress adding | |
2202 | * the log entry. The caller must eventually free the returned structure, with | |
2203 | * raft_command_unref(). */ | |
2204 | struct raft_command * OVS_WARN_UNUSED_RESULT | |
2205 | raft_command_execute(struct raft *raft, const struct json *data, | |
2206 | const struct uuid *prereq, struct uuid *result) | |
2207 | { | |
8e354614 | 2208 | return raft_command_execute__(raft, data, NULL, 0, prereq, result); |
1b1d2e6d BP |
2209 | } |
2210 | ||
2211 | /* Returns the status of 'cmd'. */ | |
2212 | enum raft_command_status | |
2213 | raft_command_get_status(const struct raft_command *cmd) | |
2214 | { | |
2215 | ovs_assert(cmd->n_refs > 0); | |
2216 | return cmd->status; | |
2217 | } | |
2218 | ||
2219 | /* Returns the index of the log entry at which 'cmd' was committed. | |
2220 | * | |
2221 | * This function works only with successful commands. */ | |
2222 | uint64_t | |
2223 | raft_command_get_commit_index(const struct raft_command *cmd) | |
2224 | { | |
2225 | ovs_assert(cmd->n_refs > 0); | |
2226 | ovs_assert(cmd->status == RAFT_CMD_SUCCESS); | |
2227 | return cmd->index; | |
2228 | } | |
2229 | ||
2230 | /* Frees 'cmd'. */ | |
2231 | void | |
2232 | raft_command_unref(struct raft_command *cmd) | |
2233 | { | |
2234 | if (cmd) { | |
2235 | ovs_assert(cmd->n_refs > 0); | |
2236 | if (!--cmd->n_refs) { | |
2237 | free(cmd); | |
2238 | } | |
2239 | } | |
2240 | } | |
2241 | ||
2242 | /* Causes poll_block() to wake up when 'cmd' has status to report. */ | |
2243 | void | |
2244 | raft_command_wait(const struct raft_command *cmd) | |
2245 | { | |
2246 | if (cmd->status != RAFT_CMD_INCOMPLETE) { | |
2247 | poll_immediate_wake(); | |
2248 | } | |
2249 | } | |
2250 | ||
2251 | static void | |
2252 | raft_command_complete(struct raft *raft, | |
2253 | struct raft_command *cmd, | |
2254 | enum raft_command_status status) | |
2255 | { | |
5a9b53a5 HZ |
2256 | VLOG_DBG("raft_command_complete eid "UUID_FMT" status: %s", |
2257 | UUID_ARGS(&cmd->eid), raft_command_status_to_string(status)); | |
1b1d2e6d BP |
2258 | if (!uuid_is_zero(&cmd->sid)) { |
2259 | uint64_t commit_index = status == RAFT_CMD_SUCCESS ? cmd->index : 0; | |
2260 | raft_send_execute_command_reply(raft, &cmd->sid, &cmd->eid, status, | |
2261 | commit_index); | |
2262 | } | |
2263 | ||
2264 | ovs_assert(cmd->status == RAFT_CMD_INCOMPLETE); | |
2265 | ovs_assert(cmd->n_refs > 0); | |
2266 | hmap_remove(&raft->commands, &cmd->hmap_node); | |
2267 | cmd->status = status; | |
2268 | raft_command_unref(cmd); | |
2269 | } | |
2270 | ||
2271 | static void | |
2272 | raft_complete_all_commands(struct raft *raft, enum raft_command_status status) | |
2273 | { | |
2274 | struct raft_command *cmd, *next; | |
2275 | HMAP_FOR_EACH_SAFE (cmd, next, hmap_node, &raft->commands) { | |
2276 | raft_command_complete(raft, cmd, status); | |
2277 | } | |
2278 | } | |
2279 | ||
1b1d2e6d BP |
2280 | static struct raft_command * |
2281 | raft_find_command_by_eid(struct raft *raft, const struct uuid *eid) | |
2282 | { | |
2283 | struct raft_command *cmd; | |
2284 | ||
2285 | HMAP_FOR_EACH (cmd, hmap_node, &raft->commands) { | |
2286 | if (uuid_equals(&cmd->eid, eid)) { | |
2287 | return cmd; | |
2288 | } | |
2289 | } | |
2290 | return NULL; | |
2291 | } | |
2292 | \f | |
2293 | #define RAFT_RPC(ENUM, NAME) \ | |
2294 | static void raft_handle_##NAME(struct raft *, const struct raft_##NAME *); | |
2295 | RAFT_RPC_TYPES | |
2296 | #undef RAFT_RPC | |
2297 | ||
2298 | static void | |
2299 | raft_handle_hello_request(struct raft *raft OVS_UNUSED, | |
2300 | const struct raft_hello_request *hello OVS_UNUSED) | |
2301 | { | |
2302 | } | |
2303 | ||
2304 | /* 'sid' is the server being added. */ | |
2305 | static void | |
2306 | raft_send_add_server_reply__(struct raft *raft, const struct uuid *sid, | |
2307 | const char *address, | |
2308 | bool success, const char *comment) | |
2309 | { | |
2310 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); | |
2311 | if (!VLOG_DROP_INFO(&rl)) { | |
2312 | struct ds s = DS_EMPTY_INITIALIZER; | |
2313 | char buf[SID_LEN + 1]; | |
2314 | ds_put_format(&s, "adding %s ("SID_FMT" at %s) " | |
2315 | "to cluster "CID_FMT" %s", | |
2316 | raft_get_nickname(raft, sid, buf, sizeof buf), | |
2317 | SID_ARGS(sid), address, CID_ARGS(&raft->cid), | |
2318 | success ? "succeeded" : "failed"); | |
2319 | if (comment) { | |
2320 | ds_put_format(&s, " (%s)", comment); | |
2321 | } | |
2322 | VLOG_INFO("%s", ds_cstr(&s)); | |
2323 | ds_destroy(&s); | |
2324 | } | |
2325 | ||
2326 | union raft_rpc rpy = { | |
2327 | .add_server_reply = { | |
2328 | .common = { | |
2329 | .type = RAFT_RPC_ADD_SERVER_REPLY, | |
2330 | .sid = *sid, | |
2331 | .comment = CONST_CAST(char *, comment), | |
2332 | }, | |
2333 | .success = success, | |
2334 | } | |
2335 | }; | |
2336 | ||
2337 | struct sset *remote_addresses = &rpy.add_server_reply.remote_addresses; | |
2338 | sset_init(remote_addresses); | |
2339 | if (!raft->joining) { | |
2340 | struct raft_server *s; | |
2341 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
2342 | if (!uuid_equals(&s->sid, &raft->sid)) { | |
2343 | sset_add(remote_addresses, s->address); | |
2344 | } | |
2345 | } | |
2346 | } | |
2347 | ||
2348 | raft_send(raft, &rpy); | |
2349 | ||
2350 | sset_destroy(remote_addresses); | |
2351 | } | |
2352 | ||
2353 | static void | |
17bd4149 BP |
2354 | raft_send_remove_server_reply_rpc(struct raft *raft, |
2355 | const struct uuid *dst_sid, | |
2356 | const struct uuid *target_sid, | |
1b1d2e6d BP |
2357 | bool success, const char *comment) |
2358 | { | |
17bd4149 BP |
2359 | if (uuid_equals(&raft->sid, dst_sid)) { |
2360 | if (success && uuid_equals(&raft->sid, target_sid)) { | |
2361 | raft_finished_leaving_cluster(raft); | |
2362 | } | |
2363 | return; | |
2364 | } | |
2365 | ||
1b1d2e6d BP |
2366 | const union raft_rpc rpy = { |
2367 | .remove_server_reply = { | |
2368 | .common = { | |
2369 | .type = RAFT_RPC_REMOVE_SERVER_REPLY, | |
17bd4149 | 2370 | .sid = *dst_sid, |
1b1d2e6d BP |
2371 | .comment = CONST_CAST(char *, comment), |
2372 | }, | |
17bd4149 BP |
2373 | .target_sid = (uuid_equals(dst_sid, target_sid) |
2374 | ? UUID_ZERO | |
2375 | : *target_sid), | |
1b1d2e6d BP |
2376 | .success = success, |
2377 | } | |
2378 | }; | |
2379 | raft_send(raft, &rpy); | |
2380 | } | |
2381 | ||
2382 | static void | |
2383 | raft_send_remove_server_reply__(struct raft *raft, | |
2384 | const struct uuid *target_sid, | |
2385 | const struct uuid *requester_sid, | |
2386 | struct unixctl_conn *requester_conn, | |
2387 | bool success, const char *comment) | |
2388 | { | |
2389 | struct ds s = DS_EMPTY_INITIALIZER; | |
2390 | ds_put_format(&s, "request "); | |
2391 | if (!uuid_is_zero(requester_sid)) { | |
2392 | char buf[SID_LEN + 1]; | |
2393 | ds_put_format(&s, "by %s", | |
2394 | raft_get_nickname(raft, requester_sid, buf, sizeof buf)); | |
2395 | } else { | |
2396 | ds_put_cstr(&s, "via unixctl"); | |
2397 | } | |
2398 | ds_put_cstr(&s, " to remove "); | |
2399 | if (!requester_conn && uuid_equals(target_sid, requester_sid)) { | |
2400 | ds_put_cstr(&s, "itself"); | |
2401 | } else { | |
2402 | char buf[SID_LEN + 1]; | |
2403 | ds_put_cstr(&s, raft_get_nickname(raft, target_sid, buf, sizeof buf)); | |
17bd4149 BP |
2404 | if (uuid_equals(target_sid, &raft->sid)) { |
2405 | ds_put_cstr(&s, " (ourselves)"); | |
2406 | } | |
1b1d2e6d BP |
2407 | } |
2408 | ds_put_format(&s, " from cluster "CID_FMT" %s", | |
2409 | CID_ARGS(&raft->cid), | |
2410 | success ? "succeeded" : "failed"); | |
2411 | if (comment) { | |
2412 | ds_put_format(&s, " (%s)", comment); | |
2413 | } | |
2414 | ||
2415 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); | |
2416 | VLOG_INFO_RL(&rl, "%s", ds_cstr(&s)); | |
2417 | ||
2418 | /* Send RemoveServerReply to the requester (which could be a server or a | |
2419 | * unixctl connection. Also always send it to the removed server; this | |
2420 | * allows it to be sure that it's really removed and update its log and | |
2421 | * disconnect permanently. */ | |
2422 | if (!uuid_is_zero(requester_sid)) { | |
17bd4149 | 2423 | raft_send_remove_server_reply_rpc(raft, requester_sid, target_sid, |
1b1d2e6d BP |
2424 | success, comment); |
2425 | } | |
2426 | if (!uuid_equals(requester_sid, target_sid)) { | |
17bd4149 BP |
2427 | raft_send_remove_server_reply_rpc(raft, target_sid, target_sid, |
2428 | success, comment); | |
1b1d2e6d BP |
2429 | } |
2430 | if (requester_conn) { | |
2431 | if (success) { | |
2432 | unixctl_command_reply(requester_conn, ds_cstr(&s)); | |
2433 | } else { | |
2434 | unixctl_command_reply_error(requester_conn, ds_cstr(&s)); | |
2435 | } | |
2436 | } | |
2437 | ||
2438 | ds_destroy(&s); | |
2439 | } | |
2440 | ||
2441 | static void | |
2442 | raft_send_add_server_reply(struct raft *raft, | |
2443 | const struct raft_add_server_request *rq, | |
2444 | bool success, const char *comment) | |
2445 | { | |
2446 | return raft_send_add_server_reply__(raft, &rq->common.sid, rq->address, | |
2447 | success, comment); | |
2448 | } | |
2449 | ||
2450 | static void | |
2451 | raft_send_remove_server_reply(struct raft *raft, | |
2452 | const struct raft_remove_server_request *rq, | |
2453 | bool success, const char *comment) | |
2454 | { | |
2455 | return raft_send_remove_server_reply__(raft, &rq->sid, &rq->common.sid, | |
2456 | rq->requester_conn, success, | |
2457 | comment); | |
2458 | } | |
2459 | ||
2460 | static void | |
2461 | raft_become_follower(struct raft *raft) | |
2462 | { | |
2463 | raft->leader_sid = UUID_ZERO; | |
2464 | if (raft->role == RAFT_FOLLOWER) { | |
2465 | return; | |
2466 | } | |
2467 | ||
2468 | raft->role = RAFT_FOLLOWER; | |
0f954f32 | 2469 | raft_reset_election_timer(raft); |
1b1d2e6d BP |
2470 | |
2471 | /* Notify clients about lost leadership. | |
2472 | * | |
2473 | * We do not reverse our changes to 'raft->servers' because the new | |
2474 | * configuration is already part of the log. Possibly the configuration | |
2475 | * log entry will not be committed, but until we know that we must use the | |
2476 | * new configuration. Our AppendEntries processing will properly update | |
2477 | * the server configuration later, if necessary. */ | |
2478 | struct raft_server *s; | |
2479 | HMAP_FOR_EACH (s, hmap_node, &raft->add_servers) { | |
2480 | raft_send_add_server_reply__(raft, &s->sid, s->address, false, | |
2481 | RAFT_SERVER_LOST_LEADERSHIP); | |
2482 | } | |
2483 | if (raft->remove_server) { | |
2484 | raft_send_remove_server_reply__(raft, &raft->remove_server->sid, | |
2485 | &raft->remove_server->requester_sid, | |
2486 | raft->remove_server->requester_conn, | |
2487 | false, RAFT_SERVER_LOST_LEADERSHIP); | |
2488 | raft_server_destroy(raft->remove_server); | |
2489 | raft->remove_server = NULL; | |
2490 | } | |
2491 | ||
2492 | raft_complete_all_commands(raft, RAFT_CMD_LOST_LEADERSHIP); | |
2493 | } | |
2494 | ||
2495 | static void | |
2496 | raft_send_append_request(struct raft *raft, | |
2497 | struct raft_server *peer, unsigned int n, | |
2498 | const char *comment) | |
2499 | { | |
2500 | ovs_assert(raft->role == RAFT_LEADER); | |
2501 | ||
2502 | const union raft_rpc rq = { | |
2503 | .append_request = { | |
2504 | .common = { | |
2505 | .type = RAFT_RPC_APPEND_REQUEST, | |
2506 | .sid = peer->sid, | |
2507 | .comment = CONST_CAST(char *, comment), | |
2508 | }, | |
2509 | .term = raft->term, | |
2510 | .prev_log_index = peer->next_index - 1, | |
2511 | .prev_log_term = (peer->next_index - 1 >= raft->log_start | |
2512 | ? raft->entries[peer->next_index - 1 | |
2513 | - raft->log_start].term | |
2514 | : raft->snap.term), | |
2515 | .leader_commit = raft->commit_index, | |
2516 | .entries = &raft->entries[peer->next_index - raft->log_start], | |
2517 | .n_entries = n, | |
2518 | }, | |
2519 | }; | |
2520 | raft_send(raft, &rq); | |
2521 | } | |
2522 | ||
2523 | static void | |
2524 | raft_send_heartbeats(struct raft *raft) | |
2525 | { | |
2526 | struct raft_server *s; | |
2527 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
2528 | if (!uuid_equals(&raft->sid, &s->sid)) { | |
2529 | raft_send_append_request(raft, s, 0, "heartbeat"); | |
2530 | } | |
2531 | } | |
2532 | ||
2533 | /* Send anyone waiting for a command to complete a ping to let them | |
2534 | * know we're still working on it. */ | |
2535 | struct raft_command *cmd; | |
2536 | HMAP_FOR_EACH (cmd, hmap_node, &raft->commands) { | |
2537 | if (!uuid_is_zero(&cmd->sid)) { | |
2538 | raft_send_execute_command_reply(raft, &cmd->sid, | |
2539 | &cmd->eid, | |
2540 | RAFT_CMD_INCOMPLETE, 0); | |
2541 | } | |
2542 | } | |
0f954f32 HZ |
2543 | |
2544 | raft_reset_ping_timer(raft); | |
1b1d2e6d BP |
2545 | } |
2546 | ||
2547 | /* Initializes the fields in 's' that represent the leader's view of the | |
2548 | * server. */ | |
2549 | static void | |
2550 | raft_server_init_leader(struct raft *raft, struct raft_server *s) | |
2551 | { | |
2552 | s->next_index = raft->log_end; | |
2553 | s->match_index = 0; | |
2554 | s->phase = RAFT_PHASE_STABLE; | |
89771c1e | 2555 | s->replied = false; |
1b1d2e6d BP |
2556 | } |
2557 | ||
923f01ca HZ |
2558 | static void |
2559 | raft_set_leader(struct raft *raft, const struct uuid *sid) | |
2560 | { | |
2561 | raft->leader_sid = *sid; | |
2833885f | 2562 | raft->ever_had_leader = raft->had_leader = true; |
923f01ca HZ |
2563 | raft->candidate_retrying = false; |
2564 | } | |
2565 | ||
1b1d2e6d BP |
2566 | static void |
2567 | raft_become_leader(struct raft *raft) | |
2568 | { | |
5a9b53a5 | 2569 | log_all_commands(raft); |
1b1d2e6d BP |
2570 | |
2571 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5); | |
2572 | VLOG_INFO_RL(&rl, "term %"PRIu64": elected leader by %d+ of " | |
2573 | "%"PRIuSIZE" servers", raft->term, | |
2574 | raft->n_votes, hmap_count(&raft->servers)); | |
2575 | ||
2576 | ovs_assert(raft->role != RAFT_LEADER); | |
2577 | raft->role = RAFT_LEADER; | |
923f01ca | 2578 | raft_set_leader(raft, &raft->sid); |
89771c1e | 2579 | raft_reset_election_timer(raft); |
0f954f32 | 2580 | raft_reset_ping_timer(raft); |
1b1d2e6d BP |
2581 | |
2582 | struct raft_server *s; | |
2583 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
2584 | raft_server_init_leader(raft, s); | |
2585 | } | |
2586 | ||
8e354614 HZ |
2587 | raft->election_timer_new = 0; |
2588 | ||
1b1d2e6d | 2589 | raft_update_our_match_index(raft, raft->log_end - 1); |
1b1d2e6d BP |
2590 | |
2591 | /* Write the fact that we are leader to the log. This is not used by the | |
2592 | * algorithm (although it could be, for quick restart), but it is used for | |
2593 | * offline analysis to check for conformance with the properties that Raft | |
2594 | * guarantees. */ | |
2595 | struct raft_record r = { | |
2596 | .type = RAFT_REC_LEADER, | |
2597 | .term = raft->term, | |
2598 | .sid = raft->sid, | |
2599 | }; | |
2600 | ignore(ovsdb_log_write_and_free(raft->log, raft_record_to_json(&r))); | |
2601 | ||
2602 | /* Initiate a no-op commit. Otherwise we might never find out what's in | |
2603 | * the log. See section 6.4 item 1: | |
2604 | * | |
2605 | * The Leader Completeness Property guarantees that a leader has all | |
2606 | * committed entries, but at the start of its term, it may not know | |
2607 | * which those are. To find out, it needs to commit an entry from its | |
2608 | * term. Raft handles this by having each leader commit a blank no-op | |
2609 | * entry into the log at the start of its term. As soon as this no-op | |
2610 | * entry is committed, the leader’s commit index will be at least as | |
2611 | * large as any other servers’ during its term. | |
2612 | */ | |
8e354614 HZ |
2613 | raft_command_unref(raft_command_execute__(raft, NULL, NULL, 0, NULL, |
2614 | NULL)); | |
1b1d2e6d BP |
2615 | } |
2616 | ||
2617 | /* Processes term 'term' received as part of RPC 'common'. Returns true if the | |
2618 | * caller should continue processing the RPC, false if the caller should reject | |
2619 | * it due to a stale term. */ | |
2620 | static bool | |
2621 | raft_receive_term__(struct raft *raft, const struct raft_rpc_common *common, | |
2622 | uint64_t term) | |
2623 | { | |
2624 | /* Section 3.3 says: | |
2625 | * | |
2626 | * Current terms are exchanged whenever servers communicate; if one | |
2627 | * server’s current term is smaller than the other’s, then it updates | |
2628 | * its current term to the larger value. If a candidate or leader | |
2629 | * discovers that its term is out of date, it immediately reverts to | |
2630 | * follower state. If a server receives a request with a stale term | |
2631 | * number, it rejects the request. | |
2632 | */ | |
2633 | if (term > raft->term) { | |
2634 | if (!raft_set_term(raft, term, NULL)) { | |
2635 | /* Failed to update the term to 'term'. */ | |
2636 | return false; | |
2637 | } | |
2638 | raft_become_follower(raft); | |
2639 | } else if (term < raft->term) { | |
2640 | char buf[SID_LEN + 1]; | |
2641 | VLOG_INFO("rejecting term %"PRIu64" < current term %"PRIu64" received " | |
2642 | "in %s message from server %s", | |
2643 | term, raft->term, | |
2644 | raft_rpc_type_to_string(common->type), | |
2645 | raft_get_nickname(raft, &common->sid, buf, sizeof buf)); | |
2646 | return false; | |
2647 | } | |
2648 | return true; | |
2649 | } | |
2650 | ||
2651 | static void | |
2652 | raft_get_servers_from_log(struct raft *raft, enum vlog_level level) | |
2653 | { | |
2654 | const struct json *servers_json = raft->snap.servers; | |
2655 | for (uint64_t index = raft->log_end - 1; index >= raft->log_start; | |
2656 | index--) { | |
2657 | struct raft_entry *e = &raft->entries[index - raft->log_start]; | |
2658 | if (e->servers) { | |
2659 | servers_json = e->servers; | |
2660 | break; | |
2661 | } | |
2662 | } | |
2663 | ||
2664 | struct hmap servers; | |
2665 | struct ovsdb_error *error = raft_servers_from_json(servers_json, &servers); | |
2666 | ovs_assert(!error); | |
2667 | raft_set_servers(raft, &servers, level); | |
2668 | raft_servers_destroy(&servers); | |
2669 | } | |
2670 | ||
2671 | /* Truncates the log, so that raft->log_end becomes 'new_end'. | |
2672 | * | |
2673 | * Doesn't write anything to disk. In theory, we could truncate the on-disk | |
2674 | * log file, but we don't have the right information to know how long it should | |
2675 | * be. What we actually do is to append entries for older indexes to the | |
2676 | * on-disk log; when we re-read it later, these entries truncate the log. | |
2677 | * | |
2678 | * Returns true if any of the removed log entries were server configuration | |
2679 | * entries, false otherwise. */ | |
2680 | static bool | |
2681 | raft_truncate(struct raft *raft, uint64_t new_end) | |
2682 | { | |
2683 | ovs_assert(new_end >= raft->log_start); | |
2684 | if (raft->log_end > new_end) { | |
2685 | char buf[SID_LEN + 1]; | |
2686 | VLOG_INFO("%s truncating %"PRIu64 " entries from end of log", | |
2687 | raft_get_nickname(raft, &raft->sid, buf, sizeof buf), | |
2688 | raft->log_end - new_end); | |
2689 | } | |
2690 | ||
2691 | bool servers_changed = false; | |
2692 | while (raft->log_end > new_end) { | |
2693 | struct raft_entry *entry = &raft->entries[--raft->log_end | |
2694 | - raft->log_start]; | |
2695 | if (entry->servers) { | |
2696 | servers_changed = true; | |
2697 | } | |
2698 | raft_entry_uninit(entry); | |
2699 | } | |
2700 | return servers_changed; | |
2701 | } | |
2702 | ||
2703 | static const struct json * | |
2704 | raft_peek_next_entry(struct raft *raft, struct uuid *eid) | |
2705 | { | |
2706 | /* Invariant: log_start - 2 <= last_applied <= commit_index < log_end. */ | |
2707 | ovs_assert(raft->log_start <= raft->last_applied + 2); | |
2708 | ovs_assert(raft->last_applied <= raft->commit_index); | |
2709 | ovs_assert(raft->commit_index < raft->log_end); | |
2710 | ||
2711 | if (raft->joining || raft->failed) { | |
2712 | return NULL; | |
2713 | } | |
2714 | ||
2715 | if (raft->log_start == raft->last_applied + 2) { | |
2716 | *eid = raft->snap.eid; | |
2717 | return raft->snap.data; | |
2718 | } | |
2719 | ||
2720 | while (raft->last_applied < raft->commit_index) { | |
2721 | const struct raft_entry *e = raft_get_entry(raft, | |
2722 | raft->last_applied + 1); | |
2723 | if (e->data) { | |
2724 | *eid = e->eid; | |
2725 | return e->data; | |
2726 | } | |
2727 | raft->last_applied++; | |
2728 | } | |
2729 | return NULL; | |
2730 | } | |
2731 | ||
2732 | static const struct json * | |
2733 | raft_get_next_entry(struct raft *raft, struct uuid *eid) | |
2734 | { | |
2735 | const struct json *data = raft_peek_next_entry(raft, eid); | |
2736 | if (data) { | |
2737 | raft->last_applied++; | |
2738 | } | |
2739 | return data; | |
2740 | } | |
2741 | ||
0f954f32 HZ |
2742 | /* Updates commit index in raft log. If commit index is already up-to-date |
2743 | * it does nothing and return false, otherwise, returns true. */ | |
2744 | static bool | |
1b1d2e6d BP |
2745 | raft_update_commit_index(struct raft *raft, uint64_t new_commit_index) |
2746 | { | |
2747 | if (new_commit_index <= raft->commit_index) { | |
0f954f32 | 2748 | return false; |
1b1d2e6d BP |
2749 | } |
2750 | ||
2751 | if (raft->role == RAFT_LEADER) { | |
2752 | while (raft->commit_index < new_commit_index) { | |
2753 | uint64_t index = ++raft->commit_index; | |
2754 | const struct raft_entry *e = raft_get_entry(raft, index); | |
1b1d2e6d BP |
2755 | if (e->data) { |
2756 | struct raft_command *cmd | |
5a9b53a5 | 2757 | = raft_find_command_by_eid(raft, &e->eid); |
1b1d2e6d | 2758 | if (cmd) { |
5a9b53a5 HZ |
2759 | if (!cmd->index) { |
2760 | VLOG_DBG("Command completed after role change from" | |
2761 | " follower to leader "UUID_FMT, | |
2762 | UUID_ARGS(&e->eid)); | |
2763 | cmd->index = index; | |
2764 | } | |
1b1d2e6d BP |
2765 | raft_command_complete(raft, cmd, RAFT_CMD_SUCCESS); |
2766 | } | |
2767 | } | |
8e354614 HZ |
2768 | if (e->election_timer) { |
2769 | VLOG_INFO("Election timer changed from %"PRIu64" to %"PRIu64, | |
2770 | raft->election_timer, e->election_timer); | |
2771 | raft->election_timer = e->election_timer; | |
2772 | raft->election_timer_new = 0; | |
2773 | } | |
8b37ed75 BP |
2774 | if (e->servers) { |
2775 | /* raft_run_reconfigure() can write a new Raft entry, which can | |
2776 | * reallocate raft->entries, which would invalidate 'e', so | |
2777 | * this case must be last, after the one for 'e->data'. */ | |
2778 | raft_run_reconfigure(raft); | |
2779 | } | |
1b1d2e6d BP |
2780 | } |
2781 | } else { | |
8e354614 HZ |
2782 | while (raft->commit_index < new_commit_index) { |
2783 | uint64_t index = ++raft->commit_index; | |
2784 | const struct raft_entry *e = raft_get_entry(raft, index); | |
2785 | if (e->election_timer) { | |
2786 | VLOG_INFO("Election timer changed from %"PRIu64" to %"PRIu64, | |
2787 | raft->election_timer, e->election_timer); | |
2788 | raft->election_timer = e->election_timer; | |
2789 | } | |
2790 | } | |
5a9b53a5 HZ |
2791 | /* Check if any pending command can be completed, and complete it. |
2792 | * This can happen when leader fail-over before sending | |
2793 | * execute_command_reply. */ | |
2794 | const struct uuid *eid = raft_get_eid(raft, new_commit_index); | |
2795 | struct raft_command *cmd = raft_find_command_by_eid(raft, eid); | |
2796 | if (cmd) { | |
2797 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
2798 | VLOG_INFO_RL(&rl, | |
2799 | "Command completed without reply (eid: "UUID_FMT", " | |
2800 | "commit index: %"PRIu64")", | |
2801 | UUID_ARGS(eid), new_commit_index); | |
2802 | cmd->index = new_commit_index; | |
2803 | raft_command_complete(raft, cmd, RAFT_CMD_SUCCESS); | |
2804 | } | |
1b1d2e6d BP |
2805 | } |
2806 | ||
2807 | /* Write the commit index to the log. The next time we restart, this | |
2808 | * allows us to start exporting a reasonably fresh log, instead of a log | |
2809 | * that only contains the snapshot. */ | |
2810 | struct raft_record r = { | |
2811 | .type = RAFT_REC_COMMIT_INDEX, | |
2812 | .commit_index = raft->commit_index, | |
2813 | }; | |
2814 | ignore(ovsdb_log_write_and_free(raft->log, raft_record_to_json(&r))); | |
0f954f32 | 2815 | return true; |
1b1d2e6d BP |
2816 | } |
2817 | ||
2818 | /* This doesn't use rq->entries (but it does use rq->n_entries). */ | |
2819 | static void | |
2820 | raft_send_append_reply(struct raft *raft, const struct raft_append_request *rq, | |
2821 | enum raft_append_result result, const char *comment) | |
2822 | { | |
2823 | /* Figure 3.1: "If leaderCommit > commitIndex, set commitIndex = | |
2824 | * min(leaderCommit, index of last new entry)" */ | |
2825 | if (result == RAFT_APPEND_OK && rq->leader_commit > raft->commit_index) { | |
2826 | raft_update_commit_index( | |
2827 | raft, MIN(rq->leader_commit, rq->prev_log_index + rq->n_entries)); | |
2828 | } | |
2829 | ||
2830 | /* Send reply. */ | |
2831 | union raft_rpc reply = { | |
2832 | .append_reply = { | |
2833 | .common = { | |
2834 | .type = RAFT_RPC_APPEND_REPLY, | |
2835 | .sid = rq->common.sid, | |
2836 | .comment = CONST_CAST(char *, comment), | |
2837 | }, | |
2838 | .term = raft->term, | |
2839 | .log_end = raft->log_end, | |
2840 | .prev_log_index = rq->prev_log_index, | |
2841 | .prev_log_term = rq->prev_log_term, | |
2842 | .n_entries = rq->n_entries, | |
2843 | .result = result, | |
2844 | } | |
2845 | }; | |
2846 | raft_send(raft, &reply); | |
2847 | } | |
2848 | ||
2849 | /* If 'prev_log_index' exists in 'raft''s log, in term 'prev_log_term', returns | |
2850 | * NULL. Otherwise, returns an explanation for the mismatch. */ | |
2851 | static const char * | |
2852 | match_index_and_term(const struct raft *raft, | |
2853 | uint64_t prev_log_index, uint64_t prev_log_term) | |
2854 | { | |
2855 | if (prev_log_index < raft->log_start - 1) { | |
2856 | return "mismatch before start of log"; | |
2857 | } else if (prev_log_index == raft->log_start - 1) { | |
2858 | if (prev_log_term != raft->snap.term) { | |
2859 | return "prev_term mismatch"; | |
2860 | } | |
2861 | } else if (prev_log_index < raft->log_end) { | |
2862 | if (raft->entries[prev_log_index - raft->log_start].term | |
2863 | != prev_log_term) { | |
2864 | return "term mismatch"; | |
2865 | } | |
2866 | } else { | |
2867 | /* prev_log_index >= raft->log_end */ | |
2868 | return "mismatch past end of log"; | |
2869 | } | |
2870 | return NULL; | |
2871 | } | |
2872 | ||
2873 | static void | |
2874 | raft_handle_append_entries(struct raft *raft, | |
2875 | const struct raft_append_request *rq, | |
2876 | uint64_t prev_log_index, uint64_t prev_log_term, | |
2877 | const struct raft_entry *entries, | |
2878 | unsigned int n_entries) | |
2879 | { | |
2880 | /* Section 3.5: "When sending an AppendEntries RPC, the leader includes | |
2881 | * the index and term of the entry in its log that immediately precedes | |
2882 | * the new entries. If the follower does not find an entry in its log | |
2883 | * with the same index and term, then it refuses the new entries." */ | |
2884 | const char *mismatch = match_index_and_term(raft, prev_log_index, | |
2885 | prev_log_term); | |
2886 | if (mismatch) { | |
2887 | VLOG_INFO("rejecting append_request because previous entry " | |
2888 | "%"PRIu64",%"PRIu64" not in local log (%s)", | |
2889 | prev_log_term, prev_log_index, mismatch); | |
2890 | raft_send_append_reply(raft, rq, RAFT_APPEND_INCONSISTENCY, mismatch); | |
2891 | return; | |
2892 | } | |
2893 | ||
2894 | /* Figure 3.1: "If an existing entry conflicts with a new one (same | |
2895 | * index but different terms), delete the existing entry and all that | |
2896 | * follow it." */ | |
2897 | unsigned int i; | |
2898 | bool servers_changed = false; | |
2899 | for (i = 0; ; i++) { | |
2900 | if (i >= n_entries) { | |
2901 | /* No change. */ | |
2902 | if (rq->common.comment | |
2903 | && !strcmp(rq->common.comment, "heartbeat")) { | |
2904 | raft_send_append_reply(raft, rq, RAFT_APPEND_OK, "heartbeat"); | |
2905 | } else { | |
2906 | raft_send_append_reply(raft, rq, RAFT_APPEND_OK, "no change"); | |
2907 | } | |
2908 | return; | |
2909 | } | |
2910 | ||
2911 | uint64_t log_index = (prev_log_index + 1) + i; | |
2912 | if (log_index >= raft->log_end) { | |
2913 | break; | |
2914 | } | |
2915 | if (raft->entries[log_index - raft->log_start].term | |
2916 | != entries[i].term) { | |
2917 | if (raft_truncate(raft, log_index)) { | |
2918 | servers_changed = true; | |
2919 | } | |
2920 | break; | |
2921 | } | |
2922 | } | |
2923 | ||
67dba070 HZ |
2924 | if (failure_test == FT_CRASH_AFTER_RECV_APPEND_REQ_UPDATE) { |
2925 | ovs_fatal(0, "Raft test: crash after receiving append_request with " | |
2926 | "update."); | |
2927 | } | |
1b1d2e6d BP |
2928 | /* Figure 3.1: "Append any entries not already in the log." */ |
2929 | struct ovsdb_error *error = NULL; | |
2930 | bool any_written = false; | |
2931 | for (; i < n_entries; i++) { | |
2932 | const struct raft_entry *e = &entries[i]; | |
2933 | error = raft_write_entry(raft, e->term, | |
2934 | json_nullable_clone(e->data), &e->eid, | |
8e354614 HZ |
2935 | json_nullable_clone(e->servers), |
2936 | e->election_timer); | |
1b1d2e6d BP |
2937 | if (error) { |
2938 | break; | |
2939 | } | |
2940 | any_written = true; | |
2941 | if (e->servers) { | |
2942 | servers_changed = true; | |
2943 | } | |
2944 | } | |
2945 | ||
2946 | if (any_written) { | |
2947 | raft_waiter_create(raft, RAFT_W_ENTRY, true)->entry.index | |
2948 | = raft->log_end - 1; | |
2949 | } | |
2950 | if (servers_changed) { | |
2951 | /* The set of servers might have changed; check. */ | |
2952 | raft_get_servers_from_log(raft, VLL_INFO); | |
2953 | } | |
2954 | ||
2955 | if (error) { | |
2956 | char *s = ovsdb_error_to_string_free(error); | |
2957 | VLOG_ERR("%s", s); | |
2958 | free(s); | |
2959 | raft_send_append_reply(raft, rq, RAFT_APPEND_IO_ERROR, "I/O error"); | |
2960 | return; | |
2961 | } | |
2962 | ||
2963 | raft_send_append_reply(raft, rq, RAFT_APPEND_OK, "log updated"); | |
2964 | } | |
2965 | ||
2966 | static bool | |
2967 | raft_update_leader(struct raft *raft, const struct uuid *sid) | |
2968 | { | |
2969 | if (raft->role == RAFT_LEADER) { | |
2970 | char buf[SID_LEN + 1]; | |
2971 | VLOG_ERR("this server is leader but server %s claims to be", | |
2972 | raft_get_nickname(raft, sid, buf, sizeof buf)); | |
2973 | return false; | |
2974 | } else if (!uuid_equals(sid, &raft->leader_sid)) { | |
2975 | if (!uuid_is_zero(&raft->leader_sid)) { | |
2976 | char buf1[SID_LEN + 1]; | |
2977 | char buf2[SID_LEN + 1]; | |
2978 | VLOG_ERR("leader for term %"PRIu64" changed from %s to %s", | |
2979 | raft->term, | |
2980 | raft_get_nickname(raft, &raft->leader_sid, | |
2981 | buf1, sizeof buf1), | |
2982 | raft_get_nickname(raft, sid, buf2, sizeof buf2)); | |
2983 | } else { | |
2984 | char buf[SID_LEN + 1]; | |
2985 | VLOG_INFO("server %s is leader for term %"PRIu64, | |
2986 | raft_get_nickname(raft, sid, buf, sizeof buf), | |
2987 | raft->term); | |
2988 | } | |
923f01ca | 2989 | raft_set_leader(raft, sid); |
1b1d2e6d BP |
2990 | |
2991 | /* Record the leader to the log. This is not used by the algorithm | |
2992 | * (although it could be, for quick restart), but it is used for | |
2993 | * offline analysis to check for conformance with the properties | |
2994 | * that Raft guarantees. */ | |
2995 | struct raft_record r = { | |
2996 | .type = RAFT_REC_LEADER, | |
2997 | .term = raft->term, | |
2998 | .sid = *sid, | |
2999 | }; | |
3000 | ignore(ovsdb_log_write_and_free(raft->log, raft_record_to_json(&r))); | |
3001 | } | |
93ee4209 HZ |
3002 | if (raft->role == RAFT_CANDIDATE) { |
3003 | /* Section 3.4: While waiting for votes, a candidate may | |
3004 | * receive an AppendEntries RPC from another server claiming to | |
3005 | * be leader. If the leader’s term (included in its RPC) is at | |
3006 | * least as large as the candidate’s current term, then the | |
3007 | * candidate recognizes the leader as legitimate and returns to | |
3008 | * follower state. */ | |
3009 | raft->role = RAFT_FOLLOWER; | |
3010 | } | |
1b1d2e6d BP |
3011 | return true; |
3012 | } | |
3013 | ||
3014 | static void | |
3015 | raft_handle_append_request(struct raft *raft, | |
3016 | const struct raft_append_request *rq) | |
3017 | { | |
3018 | /* We do not check whether the server that sent the request is part of the | |
3019 | * cluster. As section 4.1 says, "A server accepts AppendEntries requests | |
3020 | * from a leader that is not part of the server’s latest configuration. | |
3021 | * Otherwise, a new server could never be added to the cluster (it would | |
3022 | * never accept any log entries preceding the configuration entry that adds | |
3023 | * the server)." */ | |
3024 | if (!raft_update_leader(raft, &rq->common.sid)) { | |
3025 | raft_send_append_reply(raft, rq, RAFT_APPEND_INCONSISTENCY, | |
3026 | "usurped leadership"); | |
3027 | return; | |
3028 | } | |
0f954f32 | 3029 | raft_reset_election_timer(raft); |
1b1d2e6d BP |
3030 | |
3031 | /* First check for the common case, where the AppendEntries request is | |
3032 | * entirely for indexes covered by 'log_start' ... 'log_end - 1', something | |
3033 | * like this: | |
3034 | * | |
3035 | * rq->prev_log_index | |
3036 | * | first_entry_index | |
3037 | * | | nth_entry_index | |
3038 | * | | | | |
3039 | * v v v | |
3040 | * +---+---+---+---+ | |
3041 | * T | T | T | T | T | | |
3042 | * +---+-------+---+ | |
3043 | * +---+---+---+---+ | |
3044 | * T | T | T | T | T | | |
3045 | * +---+---+---+---+ | |
3046 | * ^ ^ | |
3047 | * | | | |
3048 | * log_start log_end | |
3049 | * */ | |
3050 | uint64_t first_entry_index = rq->prev_log_index + 1; | |
3051 | uint64_t nth_entry_index = rq->prev_log_index + rq->n_entries; | |
3052 | if (OVS_LIKELY(first_entry_index >= raft->log_start)) { | |
3053 | raft_handle_append_entries(raft, rq, | |
3054 | rq->prev_log_index, rq->prev_log_term, | |
3055 | rq->entries, rq->n_entries); | |
3056 | return; | |
3057 | } | |
3058 | ||
3059 | /* Now a series of checks for odd cases, where the AppendEntries request | |
3060 | * extends earlier than the beginning of our log, into the log entries | |
3061 | * discarded by the most recent snapshot. */ | |
3062 | ||
3063 | /* | |
3064 | * Handle the case where the indexes covered by rq->entries[] are entirely | |
3065 | * disjoint with 'log_start - 1' ... 'log_end - 1', as shown below. So, | |
3066 | * everything in the AppendEntries request must already have been | |
3067 | * committed, and we might as well return true. | |
3068 | * | |
3069 | * rq->prev_log_index | |
3070 | * | first_entry_index | |
3071 | * | | nth_entry_index | |
3072 | * | | | | |
3073 | * v v v | |
3074 | * +---+---+---+---+ | |
3075 | * T | T | T | T | T | | |
3076 | * +---+-------+---+ | |
3077 | * +---+---+---+---+ | |
3078 | * T | T | T | T | T | | |
3079 | * +---+---+---+---+ | |
3080 | * ^ ^ | |
3081 | * | | | |
3082 | * log_start log_end | |
3083 | */ | |
3084 | if (nth_entry_index < raft->log_start - 1) { | |
3085 | raft_send_append_reply(raft, rq, RAFT_APPEND_OK, | |
3086 | "append before log start"); | |
3087 | return; | |
3088 | } | |
3089 | ||
3090 | /* | |
3091 | * Handle the case where the last entry in rq->entries[] has the same index | |
3092 | * as 'log_start - 1', so we can compare their terms: | |
3093 | * | |
3094 | * rq->prev_log_index | |
3095 | * | first_entry_index | |
3096 | * | | nth_entry_index | |
3097 | * | | | | |
3098 | * v v v | |
3099 | * +---+---+---+---+ | |
3100 | * T | T | T | T | T | | |
3101 | * +---+-------+---+ | |
3102 | * +---+---+---+---+ | |
3103 | * T | T | T | T | T | | |
3104 | * +---+---+---+---+ | |
3105 | * ^ ^ | |
3106 | * | | | |
3107 | * log_start log_end | |
3108 | * | |
3109 | * There's actually a sub-case where rq->n_entries == 0, in which we | |
3110 | * compare rq->prev_term: | |
3111 | * | |
3112 | * rq->prev_log_index | |
3113 | * | | |
3114 | * | | |
3115 | * | | |
3116 | * v | |
3117 | * T | |
3118 | * | |
3119 | * +---+---+---+---+ | |
3120 | * T | T | T | T | T | | |
3121 | * +---+---+---+---+ | |
3122 | * ^ ^ | |
3123 | * | | | |
3124 | * log_start log_end | |
3125 | */ | |
3126 | if (nth_entry_index == raft->log_start - 1) { | |
3127 | if (rq->n_entries | |
3128 | ? raft->snap.term == rq->entries[rq->n_entries - 1].term | |
3129 | : raft->snap.term == rq->prev_log_term) { | |
3130 | raft_send_append_reply(raft, rq, RAFT_APPEND_OK, "no change"); | |
3131 | } else { | |
3132 | raft_send_append_reply(raft, rq, RAFT_APPEND_INCONSISTENCY, | |
3133 | "term mismatch"); | |
3134 | } | |
3135 | return; | |
3136 | } | |
3137 | ||
3138 | /* | |
3139 | * We now know that the data in rq->entries[] overlaps the data in | |
3140 | * raft->entries[], as shown below, with some positive 'ofs': | |
3141 | * | |
3142 | * rq->prev_log_index | |
3143 | * | first_entry_index | |
3144 | * | | nth_entry_index | |
3145 | * | | | | |
3146 | * v v v | |
3147 | * +---+---+---+---+---+ | |
3148 | * T | T | T | T | T | T | | |
3149 | * +---+-------+---+---+ | |
3150 | * +---+---+---+---+ | |
3151 | * T | T | T | T | T | | |
3152 | * +---+---+---+---+ | |
3153 | * ^ ^ | |
3154 | * | | | |
3155 | * log_start log_end | |
3156 | * | |
3157 | * |<-- ofs -->| | |
3158 | * | |
3159 | * We transform this into the following by trimming the first 'ofs' | |
3160 | * elements off of rq->entries[], ending up with the following. Notice how | |
3161 | * we retain the term but not the data for rq->entries[ofs - 1]: | |
3162 | * | |
3163 | * first_entry_index + ofs - 1 | |
3164 | * | first_entry_index + ofs | |
3165 | * | | nth_entry_index + ofs | |
3166 | * | | | | |
3167 | * v v v | |
3168 | * +---+---+ | |
3169 | * T | T | T | | |
3170 | * +---+---+ | |
3171 | * +---+---+---+---+ | |
3172 | * T | T | T | T | T | | |
3173 | * +---+---+---+---+ | |
3174 | * ^ ^ | |
3175 | * | | | |
3176 | * log_start log_end | |
3177 | */ | |
3178 | uint64_t ofs = raft->log_start - first_entry_index; | |
3179 | raft_handle_append_entries(raft, rq, | |
3180 | raft->log_start - 1, rq->entries[ofs - 1].term, | |
3181 | &rq->entries[ofs], rq->n_entries - ofs); | |
3182 | } | |
3183 | ||
3184 | /* Returns true if 'raft' has another log entry or snapshot to read. */ | |
3185 | bool | |
3186 | raft_has_next_entry(const struct raft *raft_) | |
3187 | { | |
3188 | struct raft *raft = CONST_CAST(struct raft *, raft_); | |
3189 | struct uuid eid; | |
3190 | return raft_peek_next_entry(raft, &eid) != NULL; | |
3191 | } | |
3192 | ||
3193 | /* Returns the next log entry or snapshot from 'raft', or NULL if there are | |
3194 | * none left to read. Stores the entry ID of the log entry in '*eid'. Stores | |
3195 | * true in '*is_snapshot' if the returned data is a snapshot, false if it is a | |
3196 | * log entry. */ | |
3197 | const struct json * | |
3198 | raft_next_entry(struct raft *raft, struct uuid *eid, bool *is_snapshot) | |
3199 | { | |
3200 | const struct json *data = raft_get_next_entry(raft, eid); | |
3201 | *is_snapshot = data == raft->snap.data; | |
3202 | return data; | |
3203 | } | |
3204 | ||
3205 | /* Returns the log index of the last-read snapshot or log entry. */ | |
3206 | uint64_t | |
3207 | raft_get_applied_index(const struct raft *raft) | |
3208 | { | |
3209 | return raft->last_applied; | |
3210 | } | |
3211 | ||
3212 | /* Returns the log index of the last snapshot or log entry that is available to | |
3213 | * be read. */ | |
3214 | uint64_t | |
3215 | raft_get_commit_index(const struct raft *raft) | |
3216 | { | |
3217 | return raft->commit_index; | |
3218 | } | |
3219 | ||
3220 | static struct raft_server * | |
3221 | raft_find_peer(struct raft *raft, const struct uuid *uuid) | |
3222 | { | |
3223 | struct raft_server *s = raft_find_server(raft, uuid); | |
3224 | return s && !uuid_equals(&raft->sid, &s->sid) ? s : NULL; | |
3225 | } | |
3226 | ||
3227 | static struct raft_server * | |
3228 | raft_find_new_server(struct raft *raft, const struct uuid *uuid) | |
3229 | { | |
3230 | return raft_server_find(&raft->add_servers, uuid); | |
3231 | } | |
3232 | ||
3233 | /* Figure 3.1: "If there exists an N such that N > commitIndex, a | |
3234 | * majority of matchIndex[i] >= N, and log[N].term == currentTerm, set | |
3235 | * commitIndex = N (sections 3.5 and 3.6)." */ | |
3236 | static void | |
3237 | raft_consider_updating_commit_index(struct raft *raft) | |
3238 | { | |
3239 | /* This loop cannot just bail out when it comes across a log entry that | |
3240 | * does not match the criteria. For example, Figure 3.7(d2) shows a | |
3241 | * case where the log entry for term 2 cannot be committed directly | |
3242 | * (because it is not for the current term) but it can be committed as | |
3243 | * a side effect of commit the entry for term 4 (the current term). | |
3244 | * XXX Is there a more efficient way to do this? */ | |
3245 | ovs_assert(raft->role == RAFT_LEADER); | |
3246 | ||
3247 | uint64_t new_commit_index = raft->commit_index; | |
3248 | for (uint64_t idx = MAX(raft->commit_index + 1, raft->log_start); | |
3249 | idx < raft->log_end; idx++) { | |
3250 | if (raft->entries[idx - raft->log_start].term == raft->term) { | |
3251 | size_t count = 0; | |
3252 | struct raft_server *s2; | |
3253 | HMAP_FOR_EACH (s2, hmap_node, &raft->servers) { | |
3254 | if (s2->match_index >= idx) { | |
3255 | count++; | |
3256 | } | |
3257 | } | |
3258 | if (count > hmap_count(&raft->servers) / 2) { | |
3259 | VLOG_DBG("index %"PRIu64" committed to %"PRIuSIZE" servers, " | |
3260 | "applying", idx, count); | |
3261 | new_commit_index = idx; | |
3262 | } | |
3263 | } | |
3264 | } | |
0f954f32 HZ |
3265 | if (raft_update_commit_index(raft, new_commit_index)) { |
3266 | raft_send_heartbeats(raft); | |
3267 | } | |
1b1d2e6d BP |
3268 | } |
3269 | ||
3270 | static void | |
3271 | raft_update_match_index(struct raft *raft, struct raft_server *s, | |
3272 | uint64_t min_index) | |
3273 | { | |
3274 | ovs_assert(raft->role == RAFT_LEADER); | |
3275 | if (min_index > s->match_index) { | |
3276 | s->match_index = min_index; | |
3277 | raft_consider_updating_commit_index(raft); | |
3278 | } | |
3279 | } | |
3280 | ||
3281 | static void | |
3282 | raft_update_our_match_index(struct raft *raft, uint64_t min_index) | |
3283 | { | |
e8208c66 BP |
3284 | struct raft_server *server = raft_find_server(raft, &raft->sid); |
3285 | if (server) { | |
3286 | raft_update_match_index(raft, server, min_index); | |
3287 | } | |
1b1d2e6d BP |
3288 | } |
3289 | ||
3290 | static void | |
3291 | raft_send_install_snapshot_request(struct raft *raft, | |
3292 | const struct raft_server *s, | |
3293 | const char *comment) | |
3294 | { | |
3295 | union raft_rpc rpc = { | |
3296 | .install_snapshot_request = { | |
3297 | .common = { | |
3298 | .type = RAFT_RPC_INSTALL_SNAPSHOT_REQUEST, | |
3299 | .sid = s->sid, | |
3300 | .comment = CONST_CAST(char *, comment), | |
3301 | }, | |
3302 | .term = raft->term, | |
3303 | .last_index = raft->log_start - 1, | |
3304 | .last_term = raft->snap.term, | |
3305 | .last_servers = raft->snap.servers, | |
3306 | .last_eid = raft->snap.eid, | |
3307 | .data = raft->snap.data, | |
9bfb280a | 3308 | .election_timer = raft->election_timer, /* use latest value */ |
1b1d2e6d BP |
3309 | } |
3310 | }; | |
8c2c503b IM |
3311 | |
3312 | if (s->last_install_snapshot_request) { | |
3313 | struct raft_install_snapshot_request *old, *new; | |
3314 | ||
3315 | old = s->last_install_snapshot_request; | |
3316 | new = &rpc.install_snapshot_request; | |
3317 | if ( old->term == new->term | |
3318 | && old->last_index == new->last_index | |
3319 | && old->last_term == new->last_term | |
3320 | && old->last_servers == new->last_servers | |
3321 | && old->data == new->data | |
3322 | && old->election_timer == new->election_timer | |
3323 | && uuid_equals(&old->last_eid, &new->last_eid)) { | |
3324 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
3325 | ||
3326 | VLOG_WARN_RL(&rl, "not sending exact same install_snapshot_request" | |
3327 | " to server %s again", s->nickname); | |
3328 | return; | |
3329 | } | |
3330 | } | |
3331 | free(s->last_install_snapshot_request); | |
3332 | CONST_CAST(struct raft_server *, s)->last_install_snapshot_request | |
3333 | = xmemdup(&rpc.install_snapshot_request, | |
3334 | sizeof rpc.install_snapshot_request); | |
3335 | ||
1b1d2e6d BP |
3336 | raft_send(raft, &rpc); |
3337 | } | |
3338 | ||
3339 | static void | |
3340 | raft_handle_append_reply(struct raft *raft, | |
3341 | const struct raft_append_reply *rpy) | |
3342 | { | |
3343 | if (raft->role != RAFT_LEADER) { | |
3344 | VLOG_INFO("rejected append_reply (not leader)"); | |
3345 | return; | |
3346 | } | |
3347 | ||
3348 | /* Most commonly we'd be getting an AppendEntries reply from a configured | |
3349 | * server (e.g. a peer), but we can also get them from servers in the | |
3350 | * process of being added. */ | |
3351 | struct raft_server *s = raft_find_peer(raft, &rpy->common.sid); | |
3352 | if (!s) { | |
3353 | s = raft_find_new_server(raft, &rpy->common.sid); | |
3354 | if (!s) { | |
3355 | VLOG_INFO("rejected append_reply from unknown server "SID_FMT, | |
3356 | SID_ARGS(&rpy->common.sid)); | |
3357 | return; | |
3358 | } | |
3359 | } | |
3360 | ||
89771c1e | 3361 | s->replied = true; |
1b1d2e6d BP |
3362 | if (rpy->result == RAFT_APPEND_OK) { |
3363 | /* Figure 3.1: "If successful, update nextIndex and matchIndex for | |
3364 | * follower (section 3.5)." */ | |
3365 | uint64_t min_index = rpy->prev_log_index + rpy->n_entries + 1; | |
3366 | if (s->next_index < min_index) { | |
3367 | s->next_index = min_index; | |
3368 | } | |
3369 | raft_update_match_index(raft, s, min_index - 1); | |
3370 | } else { | |
3371 | /* Figure 3.1: "If AppendEntries fails because of log inconsistency, | |
3372 | * decrement nextIndex and retry (section 3.5)." | |
3373 | * | |
3374 | * We also implement the optimization suggested in section 4.2.1: | |
3375 | * "Various approaches can make nextIndex converge to its correct value | |
3376 | * more quickly, including those described in Chapter 3. The simplest | |
3377 | * approach to solving this particular problem of adding a new server, | |
3378 | * however, is to have followers return the length of their logs in the | |
3379 | * AppendEntries response; this allows the leader to cap the follower’s | |
3380 | * nextIndex accordingly." */ | |
3381 | s->next_index = (s->next_index > 0 | |
3382 | ? MIN(s->next_index - 1, rpy->log_end) | |
3383 | : 0); | |
3384 | ||
3385 | if (rpy->result == RAFT_APPEND_IO_ERROR) { | |
3386 | /* Append failed but not because of a log inconsistency. Because | |
3387 | * of the I/O error, there's no point in re-sending the append | |
3388 | * immediately. */ | |
3389 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
3390 | VLOG_INFO_RL(&rl, "%s reported I/O error", s->nickname); | |
3391 | return; | |
3392 | } | |
3393 | } | |
3394 | ||
3395 | /* | |
3396 | * Our behavior here must depend on the value of next_index relative to | |
3397 | * log_start and log_end. There are three cases: | |
3398 | * | |
3399 | * Case 1 | Case 2 | Case 3 | |
3400 | * <---------------->|<------------->|<------------------> | |
3401 | * | | | |
3402 | * | |
3403 | * +---+---+---+---+ | |
3404 | * T | T | T | T | T | | |
3405 | * +---+---+---+---+ | |
3406 | * ^ ^ | |
3407 | * | | | |
3408 | * log_start log_end | |
3409 | */ | |
3410 | if (s->next_index < raft->log_start) { | |
3411 | /* Case 1. */ | |
3412 | raft_send_install_snapshot_request(raft, s, NULL); | |
3413 | } else if (s->next_index < raft->log_end) { | |
3414 | /* Case 2. */ | |
99c2dc8d | 3415 | raft_send_append_request(raft, s, raft->log_end - s->next_index, NULL); |
1b1d2e6d BP |
3416 | } else { |
3417 | /* Case 3. */ | |
3418 | if (s->phase == RAFT_PHASE_CATCHUP) { | |
3419 | s->phase = RAFT_PHASE_CAUGHT_UP; | |
3420 | raft_run_reconfigure(raft); | |
3421 | } | |
3422 | } | |
3423 | } | |
3424 | ||
3425 | static bool | |
3426 | raft_should_suppress_disruptive_server(struct raft *raft, | |
3427 | const union raft_rpc *rpc) | |
3428 | { | |
3429 | if (rpc->type != RAFT_RPC_VOTE_REQUEST) { | |
3430 | return false; | |
3431 | } | |
3432 | ||
3433 | /* Section 4.2.3 "Disruptive Servers" says: | |
3434 | * | |
3435 | * ...if a server receives a RequestVote request within the minimum | |
3436 | * election timeout of hearing from a current leader, it does not update | |
3437 | * its term or grant its vote... | |
3438 | * | |
3439 | * ...This change conflicts with the leadership transfer mechanism as | |
3440 | * described in Chapter 3, in which a server legitimately starts an | |
3441 | * election without waiting an election timeout. In that case, | |
3442 | * RequestVote messages should be processed by other servers even when | |
3443 | * they believe a current cluster leader exists. Those RequestVote | |
3444 | * requests can include a special flag to indicate this behavior (“I | |
3445 | * have permission to disrupt the leader--it told me to!”). | |
3446 | * | |
3447 | * This clearly describes how the followers should act, but not the leader. | |
3448 | * We just ignore vote requests that arrive at a current leader. This | |
3449 | * seems to be fairly safe, since a majority other than the current leader | |
3450 | * can still elect a new leader and the first AppendEntries from that new | |
3451 | * leader will depose the current leader. */ | |
3452 | const struct raft_vote_request *rq = raft_vote_request_cast(rpc); | |
3453 | if (rq->leadership_transfer) { | |
3454 | return false; | |
3455 | } | |
3456 | ||
3457 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
3458 | long long int now = time_msec(); | |
3459 | switch (raft->role) { | |
3460 | case RAFT_LEADER: | |
3461 | VLOG_WARN_RL(&rl, "ignoring vote request received as leader"); | |
3462 | return true; | |
3463 | ||
3464 | case RAFT_FOLLOWER: | |
8e354614 | 3465 | if (now < raft->election_base + raft->election_timer) { |
1b1d2e6d | 3466 | VLOG_WARN_RL(&rl, "ignoring vote request received after only " |
8e354614 HZ |
3467 | "%lld ms (minimum election time is %"PRIu64" ms)", |
3468 | now - raft->election_base, raft->election_timer); | |
1b1d2e6d BP |
3469 | return true; |
3470 | } | |
3471 | return false; | |
3472 | ||
3473 | case RAFT_CANDIDATE: | |
3474 | return false; | |
3475 | ||
3476 | default: | |
3477 | OVS_NOT_REACHED(); | |
3478 | } | |
3479 | } | |
3480 | ||
3481 | /* Returns true if a reply should be sent. */ | |
3482 | static bool | |
3483 | raft_handle_vote_request__(struct raft *raft, | |
3484 | const struct raft_vote_request *rq) | |
3485 | { | |
3486 | /* Figure 3.1: "If votedFor is null or candidateId, and candidate's vote is | |
3487 | * at least as up-to-date as receiver's log, grant vote (sections 3.4, | |
3488 | * 3.6)." */ | |
3489 | if (uuid_equals(&raft->vote, &rq->common.sid)) { | |
3490 | /* Already voted for this candidate in this term. Resend vote. */ | |
3491 | return true; | |
3492 | } else if (!uuid_is_zero(&raft->vote)) { | |
3493 | /* Already voted for different candidate in this term. Send a reply | |
3494 | * saying what candidate we did vote for. This isn't a necessary part | |
3495 | * of the Raft protocol but it can make debugging easier. */ | |
3496 | return true; | |
3497 | } | |
3498 | ||
3499 | /* Section 3.6.1: "The RequestVote RPC implements this restriction: the RPC | |
3500 | * includes information about the candidate’s log, and the voter denies its | |
3501 | * vote if its own log is more up-to-date than that of the candidate. Raft | |
3502 | * determines which of two logs is more up-to-date by comparing the index | |
3503 | * and term of the last entries in the logs. If the logs have last entries | |
3504 | * with different terms, then the log with the later term is more | |
3505 | * up-to-date. If the logs end with the same term, then whichever log is | |
3506 | * longer is more up-to-date." */ | |
3507 | uint64_t last_term = (raft->log_end > raft->log_start | |
3508 | ? raft->entries[raft->log_end - 1 | |
3509 | - raft->log_start].term | |
3510 | : raft->snap.term); | |
3511 | if (last_term > rq->last_log_term | |
3512 | || (last_term == rq->last_log_term | |
3513 | && raft->log_end - 1 > rq->last_log_index)) { | |
3514 | /* Our log is more up-to-date than the peer's. Withhold vote. */ | |
3515 | return false; | |
3516 | } | |
3517 | ||
3518 | /* Record a vote for the peer. */ | |
3519 | if (!raft_set_term(raft, raft->term, &rq->common.sid)) { | |
3520 | return false; | |
3521 | } | |
3522 | ||
0f954f32 | 3523 | raft_reset_election_timer(raft); |
1b1d2e6d BP |
3524 | |
3525 | return true; | |
3526 | } | |
3527 | ||
3528 | static void | |
3529 | raft_send_vote_reply(struct raft *raft, const struct uuid *dst, | |
3530 | const struct uuid *vote) | |
3531 | { | |
3532 | union raft_rpc rpy = { | |
3533 | .vote_reply = { | |
3534 | .common = { | |
3535 | .type = RAFT_RPC_VOTE_REPLY, | |
3536 | .sid = *dst, | |
3537 | }, | |
3538 | .term = raft->term, | |
3539 | .vote = *vote, | |
3540 | }, | |
3541 | }; | |
3542 | raft_send(raft, &rpy); | |
3543 | } | |
3544 | ||
3545 | static void | |
3546 | raft_handle_vote_request(struct raft *raft, | |
3547 | const struct raft_vote_request *rq) | |
3548 | { | |
3549 | if (raft_handle_vote_request__(raft, rq)) { | |
3550 | raft_send_vote_reply(raft, &rq->common.sid, &raft->vote); | |
3551 | } | |
3552 | } | |
3553 | ||
3554 | static void | |
3555 | raft_handle_vote_reply(struct raft *raft, | |
3556 | const struct raft_vote_reply *rpy) | |
3557 | { | |
3558 | if (!raft_receive_term__(raft, &rpy->common, rpy->term)) { | |
3559 | return; | |
3560 | } | |
3561 | ||
3562 | if (raft->role != RAFT_CANDIDATE) { | |
3563 | return; | |
3564 | } | |
3565 | ||
3566 | struct raft_server *s = raft_find_peer(raft, &rpy->common.sid); | |
3567 | if (s) { | |
3568 | raft_accept_vote(raft, s, &rpy->vote); | |
3569 | } | |
3570 | } | |
3571 | ||
3572 | /* Returns true if 'raft''s log contains reconfiguration entries that have not | |
3573 | * yet been committed. */ | |
3574 | static bool | |
3575 | raft_has_uncommitted_configuration(const struct raft *raft) | |
3576 | { | |
3577 | for (uint64_t i = raft->commit_index + 1; i < raft->log_end; i++) { | |
3578 | ovs_assert(i >= raft->log_start); | |
3579 | const struct raft_entry *e = &raft->entries[i - raft->log_start]; | |
3580 | if (e->servers) { | |
3581 | return true; | |
3582 | } | |
3583 | } | |
3584 | return false; | |
3585 | } | |
3586 | ||
3587 | static void | |
3588 | raft_log_reconfiguration(struct raft *raft) | |
3589 | { | |
3590 | struct json *servers_json = raft_servers_to_json(&raft->servers); | |
3591 | raft_command_unref(raft_command_execute__( | |
8e354614 | 3592 | raft, NULL, servers_json, 0, NULL, NULL)); |
1b1d2e6d BP |
3593 | json_destroy(servers_json); |
3594 | } | |
3595 | ||
3596 | static void | |
3597 | raft_run_reconfigure(struct raft *raft) | |
3598 | { | |
3599 | ovs_assert(raft->role == RAFT_LEADER); | |
3600 | ||
3601 | /* Reconfiguration only progresses when configuration changes commit. */ | |
3602 | if (raft_has_uncommitted_configuration(raft)) { | |
3603 | return; | |
3604 | } | |
3605 | ||
3606 | /* If we were waiting for a configuration change to commit, it's done. */ | |
3607 | struct raft_server *s; | |
3608 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
3609 | if (s->phase == RAFT_PHASE_COMMITTING) { | |
3610 | raft_send_add_server_reply__(raft, &s->sid, s->address, | |
3611 | true, RAFT_SERVER_COMPLETED); | |
3612 | s->phase = RAFT_PHASE_STABLE; | |
3613 | } | |
3614 | } | |
3615 | if (raft->remove_server) { | |
3616 | raft_send_remove_server_reply__(raft, &raft->remove_server->sid, | |
3617 | &raft->remove_server->requester_sid, | |
3618 | raft->remove_server->requester_conn, | |
3619 | true, RAFT_SERVER_COMPLETED); | |
3620 | raft_server_destroy(raft->remove_server); | |
3621 | raft->remove_server = NULL; | |
3622 | } | |
3623 | ||
3624 | /* If a new server is caught up, add it to the configuration. */ | |
3625 | HMAP_FOR_EACH (s, hmap_node, &raft->add_servers) { | |
3626 | if (s->phase == RAFT_PHASE_CAUGHT_UP) { | |
3627 | /* Move 's' from 'raft->add_servers' to 'raft->servers'. */ | |
3628 | hmap_remove(&raft->add_servers, &s->hmap_node); | |
3629 | hmap_insert(&raft->servers, &s->hmap_node, uuid_hash(&s->sid)); | |
3630 | ||
3631 | /* Mark 's' as waiting for commit. */ | |
3632 | s->phase = RAFT_PHASE_COMMITTING; | |
3633 | ||
3634 | raft_log_reconfiguration(raft); | |
3635 | ||
3636 | /* When commit completes we'll transition to RAFT_PHASE_STABLE and | |
3637 | * send a RAFT_SERVER_OK reply. */ | |
3638 | ||
3639 | return; | |
3640 | } | |
3641 | } | |
3642 | ||
3643 | /* Remove a server, if one is scheduled for removal. */ | |
3644 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
3645 | if (s->phase == RAFT_PHASE_REMOVE) { | |
3646 | hmap_remove(&raft->servers, &s->hmap_node); | |
3647 | raft->remove_server = s; | |
3648 | ||
3649 | raft_log_reconfiguration(raft); | |
3650 | ||
3651 | return; | |
3652 | } | |
3653 | } | |
3654 | } | |
3655 | ||
3656 | static void | |
3657 | raft_handle_add_server_request(struct raft *raft, | |
3658 | const struct raft_add_server_request *rq) | |
3659 | { | |
3660 | /* Figure 4.1: "1. Reply NOT_LEADER if not leader (section 6.2)." */ | |
3661 | if (raft->role != RAFT_LEADER) { | |
3662 | raft_send_add_server_reply(raft, rq, false, RAFT_SERVER_NOT_LEADER); | |
3663 | return; | |
3664 | } | |
3665 | ||
3666 | /* Check for an existing server. */ | |
3667 | struct raft_server *s = raft_find_server(raft, &rq->common.sid); | |
3668 | if (s) { | |
3669 | /* If the server is scheduled to be removed, cancel it. */ | |
3670 | if (s->phase == RAFT_PHASE_REMOVE) { | |
3671 | s->phase = RAFT_PHASE_STABLE; | |
3672 | raft_send_add_server_reply(raft, rq, false, RAFT_SERVER_CANCELED); | |
3673 | return; | |
3674 | } | |
3675 | ||
3676 | /* If the server is being added, then it's in progress. */ | |
3677 | if (s->phase != RAFT_PHASE_STABLE) { | |
3678 | raft_send_add_server_reply(raft, rq, | |
3679 | false, RAFT_SERVER_IN_PROGRESS); | |
3680 | } | |
3681 | ||
3682 | /* Nothing to do--server is already part of the configuration. */ | |
3683 | raft_send_add_server_reply(raft, rq, | |
3684 | true, RAFT_SERVER_ALREADY_PRESENT); | |
3685 | return; | |
3686 | } | |
3687 | ||
3688 | /* Check for a server being removed. */ | |
3689 | if (raft->remove_server | |
3690 | && uuid_equals(&rq->common.sid, &raft->remove_server->sid)) { | |
3691 | raft_send_add_server_reply(raft, rq, false, RAFT_SERVER_COMMITTING); | |
3692 | return; | |
3693 | } | |
3694 | ||
3695 | /* Check for a server already being added. */ | |
3696 | if (raft_find_new_server(raft, &rq->common.sid)) { | |
3697 | raft_send_add_server_reply(raft, rq, false, RAFT_SERVER_IN_PROGRESS); | |
3698 | return; | |
3699 | } | |
3700 | ||
3701 | /* Add server to 'add_servers'. */ | |
3702 | s = raft_server_add(&raft->add_servers, &rq->common.sid, rq->address); | |
3703 | raft_server_init_leader(raft, s); | |
3704 | s->requester_sid = rq->common.sid; | |
3705 | s->requester_conn = NULL; | |
3706 | s->phase = RAFT_PHASE_CATCHUP; | |
3707 | ||
3708 | /* Start sending the log. If this is the first time we've tried to add | |
3709 | * this server, then this will quickly degenerate into an InstallSnapshot | |
3710 | * followed by a series of AddEntries, but if it's a retry of an earlier | |
3711 | * AddRequest that was interrupted (e.g. by a timeout or a loss of | |
3712 | * leadership) then it will gracefully resume populating the log. | |
3713 | * | |
3714 | * See the last few paragraphs of section 4.2.1 for further insight. */ | |
3715 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); | |
3716 | VLOG_INFO_RL(&rl, | |
3717 | "starting to add server %s ("SID_FMT" at %s) " | |
3718 | "to cluster "CID_FMT, s->nickname, SID_ARGS(&s->sid), | |
3719 | rq->address, CID_ARGS(&raft->cid)); | |
3720 | raft_send_append_request(raft, s, 0, "initialize new server"); | |
3721 | } | |
3722 | ||
3723 | static void | |
3724 | raft_handle_add_server_reply(struct raft *raft, | |
3725 | const struct raft_add_server_reply *rpy) | |
3726 | { | |
3727 | if (!raft->joining) { | |
3728 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
3729 | VLOG_WARN_RL(&rl, "received add_server_reply even though we're " | |
3730 | "already part of the cluster"); | |
3731 | return; | |
3732 | } | |
3733 | ||
3734 | if (rpy->success) { | |
3735 | raft->joining = false; | |
3736 | ||
3737 | /* It is tempting, at this point, to check that this server is part of | |
3738 | * the current configuration. However, this is not necessarily the | |
3739 | * case, because the log entry that added this server to the cluster | |
3740 | * might have been committed by a majority of the cluster that does not | |
3741 | * include this one. This actually happens in testing. */ | |
3742 | } else { | |
3743 | const char *address; | |
3744 | SSET_FOR_EACH (address, &rpy->remote_addresses) { | |
3745 | if (sset_add(&raft->remote_addresses, address)) { | |
3746 | VLOG_INFO("%s: learned new server address for joining cluster", | |
3747 | address); | |
3748 | } | |
3749 | } | |
3750 | } | |
3751 | } | |
3752 | ||
3753 | /* This is called by raft_unixctl_kick() as well as via RPC. */ | |
3754 | static void | |
3755 | raft_handle_remove_server_request(struct raft *raft, | |
3756 | const struct raft_remove_server_request *rq) | |
3757 | { | |
3758 | /* Figure 4.1: "1. Reply NOT_LEADER if not leader (section 6.2)." */ | |
3759 | if (raft->role != RAFT_LEADER) { | |
3760 | raft_send_remove_server_reply(raft, rq, false, RAFT_SERVER_NOT_LEADER); | |
3761 | return; | |
3762 | } | |
3763 | ||
3764 | /* If the server to remove is currently waiting to be added, cancel it. */ | |
3765 | struct raft_server *target = raft_find_new_server(raft, &rq->sid); | |
3766 | if (target) { | |
3767 | raft_send_add_server_reply__(raft, &target->sid, target->address, | |
3768 | false, RAFT_SERVER_CANCELED); | |
3769 | hmap_remove(&raft->add_servers, &target->hmap_node); | |
3770 | raft_server_destroy(target); | |
3771 | return; | |
3772 | } | |
3773 | ||
3774 | /* If the server isn't configured, report that. */ | |
3775 | target = raft_find_server(raft, &rq->sid); | |
3776 | if (!target) { | |
3777 | raft_send_remove_server_reply(raft, rq, | |
3778 | true, RAFT_SERVER_ALREADY_GONE); | |
3779 | return; | |
3780 | } | |
3781 | ||
3782 | /* Check whether we're waiting for the addition of the server to commit. */ | |
3783 | if (target->phase == RAFT_PHASE_COMMITTING) { | |
3784 | raft_send_remove_server_reply(raft, rq, false, RAFT_SERVER_COMMITTING); | |
3785 | return; | |
3786 | } | |
3787 | ||
3788 | /* Check whether the server is already scheduled for removal. */ | |
3789 | if (target->phase == RAFT_PHASE_REMOVE) { | |
3790 | raft_send_remove_server_reply(raft, rq, | |
3791 | false, RAFT_SERVER_IN_PROGRESS); | |
3792 | return; | |
3793 | } | |
3794 | ||
3795 | /* Make sure that if we remove this server then that at least one other | |
3796 | * server will be left. We don't count servers currently being added (in | |
3797 | * 'add_servers') since those could fail. */ | |
3798 | struct raft_server *s; | |
3799 | int n = 0; | |
3800 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
3801 | if (s != target && s->phase != RAFT_PHASE_REMOVE) { | |
3802 | n++; | |
3803 | } | |
3804 | } | |
3805 | if (!n) { | |
3806 | raft_send_remove_server_reply(raft, rq, false, RAFT_SERVER_EMPTY); | |
3807 | return; | |
3808 | } | |
3809 | ||
3810 | /* Mark the server for removal. */ | |
3811 | target->phase = RAFT_PHASE_REMOVE; | |
3812 | if (rq->requester_conn) { | |
3813 | target->requester_sid = UUID_ZERO; | |
3814 | unixctl_command_reply(rq->requester_conn, "started removal"); | |
3815 | } else { | |
3816 | target->requester_sid = rq->common.sid; | |
3817 | target->requester_conn = NULL; | |
3818 | } | |
3819 | ||
3820 | raft_run_reconfigure(raft); | |
3821 | /* Operation in progress, reply will be sent later. */ | |
3822 | } | |
3823 | ||
3824 | static void | |
17bd4149 | 3825 | raft_finished_leaving_cluster(struct raft *raft) |
1b1d2e6d | 3826 | { |
17bd4149 BP |
3827 | VLOG_INFO(SID_FMT": finished leaving cluster "CID_FMT, |
3828 | SID_ARGS(&raft->sid), CID_ARGS(&raft->cid)); | |
1b1d2e6d | 3829 | |
17bd4149 BP |
3830 | raft_record_note(raft, "left", "this server left the cluster"); |
3831 | ||
3832 | raft->leaving = false; | |
3833 | raft->left = true; | |
3834 | } | |
1b1d2e6d | 3835 | |
17bd4149 BP |
3836 | static void |
3837 | raft_handle_remove_server_reply(struct raft *raft, | |
3838 | const struct raft_remove_server_reply *rpc) | |
3839 | { | |
3840 | if (rpc->success | |
3841 | && (uuid_is_zero(&rpc->target_sid) | |
3842 | || uuid_equals(&rpc->target_sid, &raft->sid))) { | |
3843 | raft_finished_leaving_cluster(raft); | |
1b1d2e6d BP |
3844 | } |
3845 | } | |
3846 | ||
3847 | static bool | |
3848 | raft_handle_write_error(struct raft *raft, struct ovsdb_error *error) | |
3849 | { | |
3850 | if (error && !raft->failed) { | |
3851 | raft->failed = true; | |
3852 | ||
3853 | char *s = ovsdb_error_to_string_free(error); | |
3854 | VLOG_WARN("%s: entering failure mode due to I/O error (%s)", | |
3855 | raft->name, s); | |
3856 | free(s); | |
3857 | } | |
3858 | return !raft->failed; | |
3859 | } | |
3860 | ||
3861 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
3862 | raft_write_snapshot(struct raft *raft, struct ovsdb_log *log, | |
3863 | uint64_t new_log_start, | |
3864 | const struct raft_entry *new_snapshot) | |
3865 | { | |
3866 | struct raft_header h = { | |
3867 | .sid = raft->sid, | |
3868 | .cid = raft->cid, | |
3869 | .name = raft->name, | |
3870 | .local_address = raft->local_address, | |
3871 | .snap_index = new_log_start - 1, | |
3872 | .snap = *new_snapshot, | |
3873 | }; | |
3874 | struct ovsdb_error *error = ovsdb_log_write_and_free( | |
3875 | log, raft_header_to_json(&h)); | |
3876 | if (error) { | |
3877 | return error; | |
3878 | } | |
3879 | ovsdb_log_mark_base(raft->log); | |
3880 | ||
3881 | /* Write log records. */ | |
3882 | for (uint64_t index = new_log_start; index < raft->log_end; index++) { | |
3883 | const struct raft_entry *e = &raft->entries[index - raft->log_start]; | |
3884 | struct raft_record r = { | |
3885 | .type = RAFT_REC_ENTRY, | |
3886 | .term = e->term, | |
3887 | .entry = { | |
3888 | .index = index, | |
3889 | .data = e->data, | |
3890 | .servers = e->servers, | |
8e354614 | 3891 | .election_timer = e->election_timer, |
1b1d2e6d BP |
3892 | .eid = e->eid, |
3893 | }, | |
3894 | }; | |
3895 | error = ovsdb_log_write_and_free(log, raft_record_to_json(&r)); | |
3896 | if (error) { | |
3897 | return error; | |
3898 | } | |
3899 | } | |
3900 | ||
3901 | /* Write term and vote (if any). | |
3902 | * | |
3903 | * The term is redundant if we wrote a log record for that term above. The | |
3904 | * vote, if any, is never redundant. | |
3905 | */ | |
3906 | error = raft_write_state(log, raft->term, &raft->vote); | |
3907 | if (error) { | |
3908 | return error; | |
3909 | } | |
3910 | ||
3911 | /* Write commit_index if it's beyond the new start of the log. */ | |
3912 | if (raft->commit_index >= new_log_start) { | |
3913 | struct raft_record r = { | |
3914 | .type = RAFT_REC_COMMIT_INDEX, | |
3915 | .commit_index = raft->commit_index, | |
3916 | }; | |
3917 | return ovsdb_log_write_and_free(log, raft_record_to_json(&r)); | |
3918 | } | |
3919 | return NULL; | |
3920 | } | |
3921 | ||
3922 | static struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
3923 | raft_save_snapshot(struct raft *raft, | |
3924 | uint64_t new_start, const struct raft_entry *new_snapshot) | |
3925 | ||
3926 | { | |
3927 | struct ovsdb_log *new_log; | |
3928 | struct ovsdb_error *error; | |
3929 | error = ovsdb_log_replace_start(raft->log, &new_log); | |
3930 | if (error) { | |
3931 | return error; | |
3932 | } | |
3933 | ||
3934 | error = raft_write_snapshot(raft, new_log, new_start, new_snapshot); | |
3935 | if (error) { | |
3936 | ovsdb_log_replace_abort(new_log); | |
3937 | return error; | |
3938 | } | |
3939 | ||
3940 | return ovsdb_log_replace_commit(raft->log, new_log); | |
3941 | } | |
3942 | ||
3943 | static bool | |
3944 | raft_handle_install_snapshot_request__( | |
3945 | struct raft *raft, const struct raft_install_snapshot_request *rq) | |
3946 | { | |
0f954f32 | 3947 | raft_reset_election_timer(raft); |
1b1d2e6d BP |
3948 | |
3949 | /* | |
3950 | * Our behavior here depend on new_log_start in the snapshot compared to | |
3951 | * log_start and log_end. There are three cases: | |
3952 | * | |
3953 | * Case 1 | Case 2 | Case 3 | |
3954 | * <---------------->|<------------->|<------------------> | |
3955 | * | | | |
3956 | * | |
3957 | * +---+---+---+---+ | |
3958 | * T | T | T | T | T | | |
3959 | * +---+---+---+---+ | |
3960 | * ^ ^ | |
3961 | * | | | |
3962 | * log_start log_end | |
3963 | */ | |
3964 | uint64_t new_log_start = rq->last_index + 1; | |
3965 | if (new_log_start < raft->log_start) { | |
3966 | /* Case 1: The new snapshot covers less than our current one. Nothing | |
3967 | * to do. */ | |
3968 | return true; | |
3969 | } else if (new_log_start < raft->log_end) { | |
3970 | /* Case 2: The new snapshot starts in the middle of our log. We could | |
3971 | * discard the first 'new_log_start - raft->log_start' entries in the | |
3972 | * log. But there's not much value in that, since snapshotting is | |
3973 | * supposed to be a local decision. Just skip it. */ | |
3974 | return true; | |
3975 | } | |
3976 | ||
3977 | /* Case 3: The new snapshot starts past the end of our current log, so | |
3978 | * discard all of our current log. */ | |
3979 | const struct raft_entry new_snapshot = { | |
3980 | .term = rq->last_term, | |
3981 | .data = rq->data, | |
3982 | .eid = rq->last_eid, | |
3983 | .servers = rq->last_servers, | |
8e354614 | 3984 | .election_timer = rq->election_timer, |
1b1d2e6d BP |
3985 | }; |
3986 | struct ovsdb_error *error = raft_save_snapshot(raft, new_log_start, | |
3987 | &new_snapshot); | |
3988 | if (error) { | |
3989 | char *error_s = ovsdb_error_to_string(error); | |
3990 | VLOG_WARN("could not save snapshot: %s", error_s); | |
3991 | free(error_s); | |
3992 | return false; | |
3993 | } | |
3994 | ||
3995 | for (size_t i = 0; i < raft->log_end - raft->log_start; i++) { | |
3996 | raft_entry_uninit(&raft->entries[i]); | |
3997 | } | |
3998 | raft->log_start = raft->log_end = new_log_start; | |
3999 | raft->log_synced = raft->log_end - 1; | |
4000 | raft->commit_index = raft->log_start - 1; | |
4001 | if (raft->last_applied < raft->commit_index) { | |
4002 | raft->last_applied = raft->log_start - 2; | |
4003 | } | |
4004 | ||
4005 | raft_entry_uninit(&raft->snap); | |
4006 | raft_entry_clone(&raft->snap, &new_snapshot); | |
4007 | ||
4008 | raft_get_servers_from_log(raft, VLL_INFO); | |
8e354614 | 4009 | raft_get_election_timer_from_log(raft); |
1b1d2e6d BP |
4010 | |
4011 | return true; | |
4012 | } | |
4013 | ||
4014 | static void | |
4015 | raft_handle_install_snapshot_request( | |
4016 | struct raft *raft, const struct raft_install_snapshot_request *rq) | |
4017 | { | |
4018 | if (raft_handle_install_snapshot_request__(raft, rq)) { | |
4019 | union raft_rpc rpy = { | |
4020 | .install_snapshot_reply = { | |
4021 | .common = { | |
4022 | .type = RAFT_RPC_INSTALL_SNAPSHOT_REPLY, | |
4023 | .sid = rq->common.sid, | |
4024 | }, | |
4025 | .term = raft->term, | |
4026 | .last_index = rq->last_index, | |
4027 | .last_term = rq->last_term, | |
4028 | }, | |
4029 | }; | |
4030 | raft_send(raft, &rpy); | |
4031 | } | |
4032 | } | |
4033 | ||
4034 | static void | |
4035 | raft_handle_install_snapshot_reply( | |
4036 | struct raft *raft, const struct raft_install_snapshot_reply *rpy) | |
4037 | { | |
4038 | /* We might get an InstallSnapshot reply from a configured server (e.g. a | |
4039 | * peer) or a server in the process of being added. */ | |
4040 | struct raft_server *s = raft_find_peer(raft, &rpy->common.sid); | |
4041 | if (!s) { | |
4042 | s = raft_find_new_server(raft, &rpy->common.sid); | |
4043 | if (!s) { | |
4044 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
4045 | VLOG_INFO_RL(&rl, "cluster "CID_FMT": received %s from " | |
4046 | "unknown server "SID_FMT, CID_ARGS(&raft->cid), | |
4047 | raft_rpc_type_to_string(rpy->common.type), | |
4048 | SID_ARGS(&rpy->common.sid)); | |
4049 | return; | |
4050 | } | |
4051 | } | |
4052 | ||
4053 | if (rpy->last_index != raft->log_start - 1 || | |
4054 | rpy->last_term != raft->snap.term) { | |
4055 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
4056 | VLOG_INFO_RL(&rl, "cluster "CID_FMT": server %s installed " | |
4057 | "out-of-date snapshot, starting over", | |
4058 | CID_ARGS(&raft->cid), s->nickname); | |
4059 | raft_send_install_snapshot_request(raft, s, | |
4060 | "installed obsolete snapshot"); | |
4061 | return; | |
4062 | } | |
4063 | ||
4064 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(10, 10); | |
4065 | VLOG_INFO_RL(&rl, "cluster "CID_FMT": installed snapshot on server %s " | |
4066 | " up to %"PRIu64":%"PRIu64, CID_ARGS(&raft->cid), | |
4067 | s->nickname, rpy->last_term, rpy->last_index); | |
315e88cb HZ |
4068 | s->next_index = raft->log_start; |
4069 | raft_send_append_request(raft, s, raft->log_end - s->next_index, | |
4070 | "snapshot installed"); | |
1b1d2e6d BP |
4071 | } |
4072 | ||
4073 | /* Returns true if 'raft' has grown enough since the last snapshot that | |
4074 | * reducing the log to a snapshot would be valuable, false otherwise. */ | |
4075 | bool | |
4076 | raft_grew_lots(const struct raft *raft) | |
4077 | { | |
4078 | return ovsdb_log_grew_lots(raft->log); | |
4079 | } | |
4080 | ||
4081 | /* Returns the number of log entries that could be trimmed off the on-disk log | |
4082 | * by snapshotting. */ | |
4083 | uint64_t | |
4084 | raft_get_log_length(const struct raft *raft) | |
4085 | { | |
4086 | return (raft->last_applied < raft->log_start | |
4087 | ? 0 | |
4088 | : raft->last_applied - raft->log_start + 1); | |
4089 | } | |
4090 | ||
4091 | /* Returns true if taking a snapshot of 'raft', with raft_store_snapshot(), is | |
4092 | * possible. */ | |
4093 | bool | |
4094 | raft_may_snapshot(const struct raft *raft) | |
4095 | { | |
4096 | return (!raft->joining | |
4097 | && !raft->leaving | |
4098 | && !raft->left | |
4099 | && !raft->failed | |
4100 | && raft->last_applied >= raft->log_start); | |
4101 | } | |
4102 | ||
4103 | /* Replaces the log for 'raft', up to the last log entry read, by | |
4104 | * 'new_snapshot_data'. Returns NULL if successful, otherwise an error that | |
4105 | * the caller must eventually free. | |
4106 | * | |
4107 | * This function can only succeed if raft_may_snapshot() returns true. It is | |
4108 | * only valuable to call it if raft_get_log_length() is significant and | |
4109 | * especially if raft_grew_lots() returns true. */ | |
4110 | struct ovsdb_error * OVS_WARN_UNUSED_RESULT | |
4111 | raft_store_snapshot(struct raft *raft, const struct json *new_snapshot_data) | |
4112 | { | |
4113 | if (raft->joining) { | |
4114 | return ovsdb_error(NULL, | |
4115 | "cannot store a snapshot while joining cluster"); | |
4116 | } else if (raft->leaving) { | |
4117 | return ovsdb_error(NULL, | |
4118 | "cannot store a snapshot while leaving cluster"); | |
4119 | } else if (raft->left) { | |
4120 | return ovsdb_error(NULL, | |
4121 | "cannot store a snapshot after leaving cluster"); | |
4122 | } else if (raft->failed) { | |
4123 | return ovsdb_error(NULL, | |
4124 | "cannot store a snapshot following failure"); | |
4125 | } | |
4126 | ||
4127 | if (raft->last_applied < raft->log_start) { | |
4128 | return ovsdb_error(NULL, "not storing a duplicate snapshot"); | |
4129 | } | |
4130 | ||
4131 | uint64_t new_log_start = raft->last_applied + 1; | |
a521491b | 4132 | struct raft_entry new_snapshot = { |
1b1d2e6d | 4133 | .term = raft_get_term(raft, new_log_start - 1), |
a521491b | 4134 | .data = json_clone(new_snapshot_data), |
1b1d2e6d | 4135 | .eid = *raft_get_eid(raft, new_log_start - 1), |
a521491b | 4136 | .servers = json_clone(raft_servers_for_index(raft, new_log_start - 1)), |
8e354614 | 4137 | .election_timer = raft->election_timer, |
1b1d2e6d BP |
4138 | }; |
4139 | struct ovsdb_error *error = raft_save_snapshot(raft, new_log_start, | |
4140 | &new_snapshot); | |
4141 | if (error) { | |
a521491b | 4142 | raft_entry_uninit(&new_snapshot); |
1b1d2e6d BP |
4143 | return error; |
4144 | } | |
4145 | ||
4146 | raft->log_synced = raft->log_end - 1; | |
4147 | raft_entry_uninit(&raft->snap); | |
a521491b | 4148 | raft->snap = new_snapshot; |
1b1d2e6d BP |
4149 | for (size_t i = 0; i < new_log_start - raft->log_start; i++) { |
4150 | raft_entry_uninit(&raft->entries[i]); | |
4151 | } | |
4152 | memmove(&raft->entries[0], &raft->entries[new_log_start - raft->log_start], | |
4153 | (raft->log_end - new_log_start) * sizeof *raft->entries); | |
4154 | raft->log_start = new_log_start; | |
4155 | return NULL; | |
4156 | } | |
4157 | ||
4158 | static void | |
4159 | raft_handle_become_leader(struct raft *raft, | |
4160 | const struct raft_become_leader *rq) | |
4161 | { | |
4162 | if (raft->role == RAFT_FOLLOWER) { | |
4163 | char buf[SID_LEN + 1]; | |
4164 | VLOG_INFO("received leadership transfer from %s in term %"PRIu64, | |
4165 | raft_get_nickname(raft, &rq->common.sid, buf, sizeof buf), | |
4166 | rq->term); | |
4167 | raft_start_election(raft, true); | |
4168 | } | |
4169 | } | |
4170 | ||
4171 | static void | |
4172 | raft_send_execute_command_reply(struct raft *raft, | |
4173 | const struct uuid *sid, | |
4174 | const struct uuid *eid, | |
4175 | enum raft_command_status status, | |
4176 | uint64_t commit_index) | |
4177 | { | |
67dba070 HZ |
4178 | if (failure_test == FT_CRASH_BEFORE_SEND_EXEC_REP) { |
4179 | ovs_fatal(0, "Raft test: crash before sending execute_command_reply"); | |
4180 | } | |
1b1d2e6d BP |
4181 | union raft_rpc rpc = { |
4182 | .execute_command_reply = { | |
4183 | .common = { | |
4184 | .type = RAFT_RPC_EXECUTE_COMMAND_REPLY, | |
4185 | .sid = *sid, | |
4186 | }, | |
4187 | .result = *eid, | |
4188 | .status = status, | |
4189 | .commit_index = commit_index, | |
4190 | }, | |
4191 | }; | |
4192 | raft_send(raft, &rpc); | |
67dba070 HZ |
4193 | if (failure_test == FT_CRASH_AFTER_SEND_EXEC_REP) { |
4194 | ovs_fatal(0, "Raft test: crash after sending execute_command_reply."); | |
4195 | } | |
1b1d2e6d BP |
4196 | } |
4197 | ||
4198 | static enum raft_command_status | |
4199 | raft_handle_execute_command_request__( | |
4200 | struct raft *raft, const struct raft_execute_command_request *rq) | |
4201 | { | |
4202 | if (raft->role != RAFT_LEADER) { | |
4203 | return RAFT_CMD_NOT_LEADER; | |
4204 | } | |
4205 | ||
4206 | const struct uuid *current_eid = raft_current_eid(raft); | |
4207 | if (!uuid_equals(&rq->prereq, current_eid)) { | |
4208 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
4209 | VLOG_INFO_RL(&rl, "current entry eid "UUID_FMT" does not match " | |
4210 | "prerequisite "UUID_FMT" in execute_command_request", | |
4211 | UUID_ARGS(current_eid), UUID_ARGS(&rq->prereq)); | |
4212 | return RAFT_CMD_BAD_PREREQ; | |
4213 | } | |
4214 | ||
4215 | struct raft_command *cmd = raft_command_initiate(raft, rq->data, | |
8e354614 | 4216 | NULL, 0, &rq->result); |
1b1d2e6d BP |
4217 | cmd->sid = rq->common.sid; |
4218 | ||
4219 | enum raft_command_status status = cmd->status; | |
48b1c764 | 4220 | raft_command_unref(cmd); |
1b1d2e6d BP |
4221 | return status; |
4222 | } | |
4223 | ||
4224 | static void | |
4225 | raft_handle_execute_command_request( | |
4226 | struct raft *raft, const struct raft_execute_command_request *rq) | |
4227 | { | |
4228 | enum raft_command_status status | |
4229 | = raft_handle_execute_command_request__(raft, rq); | |
4230 | if (status != RAFT_CMD_INCOMPLETE) { | |
4231 | raft_send_execute_command_reply(raft, &rq->common.sid, &rq->result, | |
4232 | status, 0); | |
4233 | } | |
4234 | } | |
4235 | ||
4236 | static void | |
4237 | raft_handle_execute_command_reply( | |
4238 | struct raft *raft, const struct raft_execute_command_reply *rpy) | |
4239 | { | |
4240 | struct raft_command *cmd = raft_find_command_by_eid(raft, &rpy->result); | |
4241 | if (!cmd) { | |
4242 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 5); | |
4243 | char buf[SID_LEN + 1]; | |
4244 | VLOG_INFO_RL(&rl, | |
4245 | "%s received \"%s\" reply from %s for unknown command", | |
4246 | raft->local_nickname, | |
4247 | raft_command_status_to_string(rpy->status), | |
4248 | raft_get_nickname(raft, &rpy->common.sid, | |
4249 | buf, sizeof buf)); | |
4250 | return; | |
4251 | } | |
4252 | ||
4253 | if (rpy->status == RAFT_CMD_INCOMPLETE) { | |
4254 | cmd->timestamp = time_msec(); | |
4255 | } else { | |
4256 | cmd->index = rpy->commit_index; | |
4257 | raft_command_complete(raft, cmd, rpy->status); | |
4258 | } | |
4259 | } | |
4260 | ||
4261 | static void | |
4262 | raft_handle_rpc(struct raft *raft, const union raft_rpc *rpc) | |
4263 | { | |
4264 | uint64_t term = raft_rpc_get_term(rpc); | |
4265 | if (term | |
4266 | && !raft_should_suppress_disruptive_server(raft, rpc) | |
4267 | && !raft_receive_term__(raft, &rpc->common, term)) { | |
4268 | if (rpc->type == RAFT_RPC_APPEND_REQUEST) { | |
4269 | /* Section 3.3: "If a server receives a request with a stale term | |
4270 | * number, it rejects the request." */ | |
4271 | raft_send_append_reply(raft, raft_append_request_cast(rpc), | |
4272 | RAFT_APPEND_INCONSISTENCY, "stale term"); | |
4273 | } | |
4274 | return; | |
4275 | } | |
4276 | ||
4277 | switch (rpc->type) { | |
4278 | #define RAFT_RPC(ENUM, NAME) \ | |
4279 | case ENUM: \ | |
4280 | raft_handle_##NAME(raft, &rpc->NAME); \ | |
4281 | break; | |
4282 | RAFT_RPC_TYPES | |
4283 | #undef RAFT_RPC | |
4284 | default: | |
4285 | OVS_NOT_REACHED(); | |
4286 | } | |
4287 | } | |
4288 | \f | |
4289 | static bool | |
4290 | raft_rpc_is_heartbeat(const union raft_rpc *rpc) | |
4291 | { | |
4292 | return ((rpc->type == RAFT_RPC_APPEND_REQUEST | |
4293 | || rpc->type == RAFT_RPC_APPEND_REPLY) | |
4294 | && rpc->common.comment | |
4295 | && !strcmp(rpc->common.comment, "heartbeat")); | |
4296 | } | |
4297 | ||
4298 | \f | |
4299 | static bool | |
02acb41a BP |
4300 | raft_send_to_conn_at(struct raft *raft, const union raft_rpc *rpc, |
4301 | struct raft_conn *conn, int line_number) | |
1b1d2e6d | 4302 | { |
02acb41a | 4303 | log_rpc(rpc, "-->", conn, line_number); |
1b1d2e6d BP |
4304 | return !jsonrpc_session_send( |
4305 | conn->js, raft_rpc_to_jsonrpc(&raft->cid, &raft->sid, rpc)); | |
4306 | } | |
4307 | ||
4308 | static bool | |
4309 | raft_is_rpc_synced(const struct raft *raft, const union raft_rpc *rpc) | |
4310 | { | |
4311 | uint64_t term = raft_rpc_get_term(rpc); | |
4312 | uint64_t index = raft_rpc_get_min_sync_index(rpc); | |
4313 | const struct uuid *vote = raft_rpc_get_vote(rpc); | |
4314 | ||
4315 | return (term <= raft->synced_term | |
4316 | && index <= raft->log_synced | |
4317 | && (!vote || uuid_equals(vote, &raft->synced_vote))); | |
4318 | } | |
4319 | ||
4320 | static bool | |
02acb41a | 4321 | raft_send_at(struct raft *raft, const union raft_rpc *rpc, int line_number) |
1b1d2e6d BP |
4322 | { |
4323 | const struct uuid *dst = &rpc->common.sid; | |
4324 | if (uuid_equals(dst, &raft->sid)) { | |
4325 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
02acb41a BP |
4326 | VLOG_WARN_RL(&rl, "attempted to send RPC to self from raft.c:%d", |
4327 | line_number); | |
1b1d2e6d BP |
4328 | return false; |
4329 | } | |
4330 | ||
4331 | struct raft_conn *conn = raft_find_conn_by_sid(raft, dst); | |
4332 | if (!conn) { | |
4333 | static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 1); | |
4334 | char buf[SID_LEN + 1]; | |
02acb41a BP |
4335 | VLOG_DBG_RL(&rl, "%s: no connection to %s, cannot send RPC " |
4336 | "from raft.c:%d", raft->local_nickname, | |
4337 | raft_get_nickname(raft, dst, buf, sizeof buf), | |
4338 | line_number); | |
1b1d2e6d BP |
4339 | return false; |
4340 | } | |
4341 | ||
4342 | if (!raft_is_rpc_synced(raft, rpc)) { | |
4343 | raft_waiter_create(raft, RAFT_W_RPC, false)->rpc = raft_rpc_clone(rpc); | |
4344 | return true; | |
4345 | } | |
4346 | ||
02acb41a | 4347 | return raft_send_to_conn_at(raft, rpc, conn, line_number); |
1b1d2e6d BP |
4348 | } |
4349 | \f | |
4350 | static struct raft * | |
4351 | raft_lookup_by_name(const char *name) | |
4352 | { | |
4353 | struct raft *raft; | |
4354 | ||
4355 | HMAP_FOR_EACH_WITH_HASH (raft, hmap_node, hash_string(name, 0), | |
4356 | &all_rafts) { | |
4357 | if (!strcmp(raft->name, name)) { | |
4358 | return raft; | |
4359 | } | |
4360 | } | |
4361 | return NULL; | |
4362 | } | |
4363 | ||
4364 | static void | |
4365 | raft_unixctl_cid(struct unixctl_conn *conn, | |
4366 | int argc OVS_UNUSED, const char *argv[], | |
4367 | void *aux OVS_UNUSED) | |
4368 | { | |
4369 | struct raft *raft = raft_lookup_by_name(argv[1]); | |
4370 | if (!raft) { | |
4371 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4372 | } else if (uuid_is_zero(&raft->cid)) { | |
4373 | unixctl_command_reply_error(conn, "cluster id not yet known"); | |
4374 | } else { | |
4375 | char *uuid = xasprintf(UUID_FMT, UUID_ARGS(&raft->cid)); | |
4376 | unixctl_command_reply(conn, uuid); | |
4377 | free(uuid); | |
4378 | } | |
4379 | } | |
4380 | ||
4381 | static void | |
4382 | raft_unixctl_sid(struct unixctl_conn *conn, | |
4383 | int argc OVS_UNUSED, const char *argv[], | |
4384 | void *aux OVS_UNUSED) | |
4385 | { | |
4386 | struct raft *raft = raft_lookup_by_name(argv[1]); | |
4387 | if (!raft) { | |
4388 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4389 | } else { | |
4390 | char *uuid = xasprintf(UUID_FMT, UUID_ARGS(&raft->sid)); | |
4391 | unixctl_command_reply(conn, uuid); | |
4392 | free(uuid); | |
4393 | } | |
4394 | } | |
4395 | ||
4396 | static void | |
4397 | raft_put_sid(const char *title, const struct uuid *sid, | |
4398 | const struct raft *raft, struct ds *s) | |
4399 | { | |
4400 | ds_put_format(s, "%s: ", title); | |
4401 | if (uuid_equals(sid, &raft->sid)) { | |
4402 | ds_put_cstr(s, "self"); | |
4403 | } else if (uuid_is_zero(sid)) { | |
4404 | ds_put_cstr(s, "unknown"); | |
4405 | } else { | |
4406 | char buf[SID_LEN + 1]; | |
4407 | ds_put_cstr(s, raft_get_nickname(raft, sid, buf, sizeof buf)); | |
4408 | } | |
4409 | ds_put_char(s, '\n'); | |
4410 | } | |
4411 | ||
4412 | static void | |
4413 | raft_unixctl_status(struct unixctl_conn *conn, | |
4414 | int argc OVS_UNUSED, const char *argv[], | |
4415 | void *aux OVS_UNUSED) | |
4416 | { | |
4417 | struct raft *raft = raft_lookup_by_name(argv[1]); | |
4418 | if (!raft) { | |
4419 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4420 | return; | |
4421 | } | |
4422 | ||
4423 | struct ds s = DS_EMPTY_INITIALIZER; | |
4424 | ds_put_format(&s, "%s\n", raft->local_nickname); | |
4425 | ds_put_format(&s, "Name: %s\n", raft->name); | |
4426 | ds_put_format(&s, "Cluster ID: "); | |
4427 | if (!uuid_is_zero(&raft->cid)) { | |
4428 | ds_put_format(&s, CID_FMT" ("UUID_FMT")\n", | |
4429 | CID_ARGS(&raft->cid), UUID_ARGS(&raft->cid)); | |
4430 | } else { | |
4431 | ds_put_format(&s, "not yet known\n"); | |
4432 | } | |
4433 | ds_put_format(&s, "Server ID: "SID_FMT" ("UUID_FMT")\n", | |
4434 | SID_ARGS(&raft->sid), UUID_ARGS(&raft->sid)); | |
4435 | ds_put_format(&s, "Address: %s\n", raft->local_address); | |
4436 | ds_put_format(&s, "Status: %s\n", | |
4437 | raft->joining ? "joining cluster" | |
4438 | : raft->leaving ? "leaving cluster" | |
4439 | : raft->left ? "left cluster" | |
4440 | : raft->failed ? "failed" | |
4441 | : "cluster member"); | |
4442 | if (raft->joining) { | |
4443 | ds_put_format(&s, "Remotes for joining:"); | |
4444 | const char *address; | |
4445 | SSET_FOR_EACH (address, &raft->remote_addresses) { | |
4446 | ds_put_format(&s, " %s", address); | |
4447 | } | |
4448 | ds_put_char(&s, '\n'); | |
4449 | } | |
4450 | if (raft->role == RAFT_LEADER) { | |
4451 | struct raft_server *as; | |
4452 | HMAP_FOR_EACH (as, hmap_node, &raft->add_servers) { | |
4453 | ds_put_format(&s, "Adding server %s ("SID_FMT" at %s) (%s)\n", | |
4454 | as->nickname, SID_ARGS(&as->sid), as->address, | |
4455 | raft_server_phase_to_string(as->phase)); | |
4456 | } | |
4457 | ||
4458 | struct raft_server *rs = raft->remove_server; | |
4459 | if (rs) { | |
4460 | ds_put_format(&s, "Removing server %s ("SID_FMT" at %s) (%s)\n", | |
4461 | rs->nickname, SID_ARGS(&rs->sid), rs->address, | |
4462 | raft_server_phase_to_string(rs->phase)); | |
4463 | } | |
4464 | } | |
4465 | ||
4466 | ds_put_format(&s, "Role: %s\n", | |
4467 | raft->role == RAFT_LEADER ? "leader" | |
4468 | : raft->role == RAFT_CANDIDATE ? "candidate" | |
4469 | : raft->role == RAFT_FOLLOWER ? "follower" | |
4470 | : "<error>"); | |
4471 | ds_put_format(&s, "Term: %"PRIu64"\n", raft->term); | |
4472 | raft_put_sid("Leader", &raft->leader_sid, raft, &s); | |
4473 | raft_put_sid("Vote", &raft->vote, raft, &s); | |
4474 | ds_put_char(&s, '\n'); | |
4475 | ||
8e354614 HZ |
4476 | ds_put_format(&s, "Election timer: %"PRIu64, raft->election_timer); |
4477 | if (raft->role == RAFT_LEADER && raft->election_timer_new) { | |
4478 | ds_put_format(&s, " (changing to %"PRIu64")", | |
4479 | raft->election_timer_new); | |
4480 | } | |
4481 | ds_put_char(&s, '\n'); | |
4482 | ||
1b1d2e6d BP |
4483 | ds_put_format(&s, "Log: [%"PRIu64", %"PRIu64"]\n", |
4484 | raft->log_start, raft->log_end); | |
4485 | ||
4486 | uint64_t n_uncommitted = raft->log_end - raft->commit_index - 1; | |
4487 | ds_put_format(&s, "Entries not yet committed: %"PRIu64"\n", n_uncommitted); | |
4488 | ||
4489 | uint64_t n_unapplied = raft->log_end - raft->last_applied - 1; | |
4490 | ds_put_format(&s, "Entries not yet applied: %"PRIu64"\n", n_unapplied); | |
4491 | ||
4492 | const struct raft_conn *c; | |
4493 | ds_put_cstr(&s, "Connections:"); | |
4494 | LIST_FOR_EACH (c, list_node, &raft->conns) { | |
4495 | bool connected = jsonrpc_session_is_connected(c->js); | |
4496 | ds_put_format(&s, " %s%s%s%s", | |
4497 | connected ? "" : "(", | |
4498 | c->incoming ? "<-" : "->", c->nickname, | |
4499 | connected ? "" : ")"); | |
4500 | } | |
4501 | ds_put_char(&s, '\n'); | |
4502 | ||
4503 | ds_put_cstr(&s, "Servers:\n"); | |
4504 | struct raft_server *server; | |
4505 | HMAP_FOR_EACH (server, hmap_node, &raft->servers) { | |
4506 | ds_put_format(&s, " %s ("SID_FMT" at %s)", | |
4507 | server->nickname, | |
4508 | SID_ARGS(&server->sid), server->address); | |
4509 | if (uuid_equals(&server->sid, &raft->sid)) { | |
4510 | ds_put_cstr(&s, " (self)"); | |
4511 | } | |
4512 | if (server->phase != RAFT_PHASE_STABLE) { | |
4513 | ds_put_format (&s, " (%s)", | |
4514 | raft_server_phase_to_string(server->phase)); | |
4515 | } | |
4516 | if (raft->role == RAFT_CANDIDATE) { | |
4517 | if (!uuid_is_zero(&server->vote)) { | |
4518 | char buf[SID_LEN + 1]; | |
4519 | ds_put_format(&s, " (voted for %s)", | |
4520 | raft_get_nickname(raft, &server->vote, | |
4521 | buf, sizeof buf)); | |
4522 | } | |
4523 | } else if (raft->role == RAFT_LEADER) { | |
4524 | ds_put_format(&s, " next_index=%"PRIu64" match_index=%"PRIu64, | |
4525 | server->next_index, server->match_index); | |
4526 | } | |
4527 | ds_put_char(&s, '\n'); | |
4528 | } | |
4529 | ||
4530 | unixctl_command_reply(conn, ds_cstr(&s)); | |
4531 | ds_destroy(&s); | |
4532 | } | |
4533 | ||
4534 | static void | |
4535 | raft_unixctl_leave__(struct unixctl_conn *conn, struct raft *raft) | |
4536 | { | |
4537 | if (raft_is_leaving(raft)) { | |
4538 | unixctl_command_reply_error(conn, | |
4539 | "already in progress leaving cluster"); | |
4540 | } else if (raft_is_joining(raft)) { | |
4541 | unixctl_command_reply_error(conn, | |
4542 | "can't leave while join in progress"); | |
4543 | } else if (raft_failed(raft)) { | |
4544 | unixctl_command_reply_error(conn, | |
4545 | "can't leave after failure"); | |
4546 | } else { | |
4547 | raft_leave(raft); | |
4548 | unixctl_command_reply(conn, NULL); | |
4549 | } | |
4550 | } | |
4551 | ||
4552 | static void | |
4553 | raft_unixctl_leave(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
4554 | const char *argv[], void *aux OVS_UNUSED) | |
4555 | { | |
4556 | struct raft *raft = raft_lookup_by_name(argv[1]); | |
4557 | if (!raft) { | |
4558 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4559 | return; | |
4560 | } | |
4561 | ||
4562 | raft_unixctl_leave__(conn, raft); | |
4563 | } | |
4564 | ||
4565 | static struct raft_server * | |
4566 | raft_lookup_server_best_match(struct raft *raft, const char *id) | |
4567 | { | |
4568 | struct raft_server *best = NULL; | |
4569 | int best_score = -1; | |
4570 | int n_best = 0; | |
4571 | ||
4572 | struct raft_server *s; | |
4573 | HMAP_FOR_EACH (s, hmap_node, &raft->servers) { | |
4574 | int score = (!strcmp(id, s->address) | |
4575 | ? INT_MAX | |
4576 | : uuid_is_partial_match(&s->sid, id)); | |
4577 | if (score > best_score) { | |
4578 | best = s; | |
4579 | best_score = score; | |
4580 | n_best = 1; | |
4581 | } else if (score == best_score) { | |
4582 | n_best++; | |
4583 | } | |
4584 | } | |
4585 | return n_best == 1 ? best : NULL; | |
4586 | } | |
4587 | ||
4588 | static void | |
4589 | raft_unixctl_kick(struct unixctl_conn *conn, int argc OVS_UNUSED, | |
4590 | const char *argv[], void *aux OVS_UNUSED) | |
4591 | { | |
4592 | const char *cluster_name = argv[1]; | |
4593 | const char *server_name = argv[2]; | |
4594 | ||
4595 | struct raft *raft = raft_lookup_by_name(cluster_name); | |
4596 | if (!raft) { | |
4597 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4598 | return; | |
4599 | } | |
4600 | ||
4601 | struct raft_server *server = raft_lookup_server_best_match(raft, | |
4602 | server_name); | |
4603 | if (!server) { | |
4604 | unixctl_command_reply_error(conn, "unknown server"); | |
4605 | return; | |
4606 | } | |
4607 | ||
4608 | if (uuid_equals(&server->sid, &raft->sid)) { | |
4609 | raft_unixctl_leave__(conn, raft); | |
4610 | } else if (raft->role == RAFT_LEADER) { | |
4611 | const struct raft_remove_server_request rq = { | |
4612 | .sid = server->sid, | |
4613 | .requester_conn = conn, | |
4614 | }; | |
4615 | raft_handle_remove_server_request(raft, &rq); | |
4616 | } else { | |
4617 | const union raft_rpc rpc = { | |
4618 | .remove_server_request = { | |
4619 | .common = { | |
4620 | .type = RAFT_RPC_REMOVE_SERVER_REQUEST, | |
4621 | .sid = raft->leader_sid, | |
4622 | .comment = "via unixctl" | |
4623 | }, | |
4624 | .sid = server->sid, | |
4625 | } | |
4626 | }; | |
4627 | if (raft_send(raft, &rpc)) { | |
4628 | unixctl_command_reply(conn, "sent removal request to leader"); | |
4629 | } else { | |
4630 | unixctl_command_reply_error(conn, | |
4631 | "failed to send removal request"); | |
4632 | } | |
4633 | } | |
4634 | } | |
4635 | ||
8e354614 HZ |
4636 | static void |
4637 | raft_get_election_timer_from_log(struct raft *raft) | |
4638 | { | |
4639 | if (raft->snap.election_timer) { | |
4640 | raft->election_timer = raft->snap.election_timer; | |
4641 | } | |
4642 | for (uint64_t index = raft->commit_index; index >= raft->log_start; | |
4643 | index--) { | |
4644 | struct raft_entry *e = &raft->entries[index - raft->log_start]; | |
4645 | if (e->election_timer) { | |
4646 | raft->election_timer = e->election_timer; | |
4647 | break; | |
4648 | } | |
4649 | } | |
4650 | } | |
4651 | ||
4652 | static void | |
4653 | raft_log_election_timer(struct raft *raft) | |
4654 | { | |
4655 | raft_command_unref(raft_command_execute__(raft, NULL, NULL, | |
4656 | raft->election_timer_new, NULL, | |
4657 | NULL)); | |
4658 | } | |
4659 | ||
4660 | static void | |
4661 | raft_unixctl_change_election_timer(struct unixctl_conn *conn, | |
4662 | int argc OVS_UNUSED, const char *argv[], | |
4663 | void *aux OVS_UNUSED) | |
4664 | { | |
4665 | const char *cluster_name = argv[1]; | |
4666 | const char *election_timer_str = argv[2]; | |
4667 | ||
4668 | struct raft *raft = raft_lookup_by_name(cluster_name); | |
4669 | if (!raft) { | |
4670 | unixctl_command_reply_error(conn, "unknown cluster"); | |
4671 | return; | |
4672 | } | |
4673 | ||
4674 | if (raft->role != RAFT_LEADER) { | |
4675 | unixctl_command_reply_error(conn, "election timer must be changed" | |
4676 | " through leader."); | |
4677 | return; | |
4678 | } | |
4679 | ||
4680 | /* If there are pending changes for election timer, reject it. */ | |
4681 | if (raft->election_timer_new) { | |
4682 | unixctl_command_reply_error(conn, "election timer change pending."); | |
4683 | return; | |
4684 | } | |
4685 | ||
4686 | uint64_t election_timer = atoll(election_timer_str); | |
4687 | if (election_timer == raft->election_timer) { | |
4688 | unixctl_command_reply(conn, "change election timer to current value."); | |
4689 | return; | |
4690 | } | |
4691 | ||
4692 | /* Election timer smaller than 100ms or bigger than 10min doesn't make | |
4693 | * sense. */ | |
4694 | if (election_timer < 100 || election_timer > 600000) { | |
4695 | unixctl_command_reply_error(conn, "election timer must be between " | |
4696 | "100 and 600000, in msec."); | |
4697 | return; | |
4698 | } | |
4699 | ||
4700 | /* If election timer is to be enlarged, it should be done gradually so that | |
4701 | * it won't cause timeout when new value is applied on leader but not yet | |
4702 | * applied on some of the followers. */ | |
4703 | if (election_timer > raft->election_timer * 2) { | |
4704 | unixctl_command_reply_error(conn, "election timer increase should not " | |
4705 | "exceed the current value x 2."); | |
4706 | return; | |
4707 | } | |
4708 | ||
4709 | raft->election_timer_new = election_timer; | |
4710 | raft_log_election_timer(raft); | |
4711 | unixctl_command_reply(conn, "change of election timer initiated."); | |
4712 | } | |
4713 | ||
67dba070 HZ |
4714 | static void |
4715 | raft_unixctl_failure_test(struct unixctl_conn *conn OVS_UNUSED, | |
4716 | int argc OVS_UNUSED, const char *argv[], | |
4717 | void *aux OVS_UNUSED) | |
4718 | { | |
4719 | const char *test = argv[1]; | |
4720 | if (!strcmp(test, "crash-before-sending-append-request")) { | |
4721 | failure_test = FT_CRASH_BEFORE_SEND_APPEND_REQ; | |
4722 | } else if (!strcmp(test, "crash-after-sending-append-request")) { | |
4723 | failure_test = FT_CRASH_AFTER_SEND_APPEND_REQ; | |
4724 | } else if (!strcmp(test, "crash-before-sending-execute-command-reply")) { | |
4725 | failure_test = FT_CRASH_BEFORE_SEND_EXEC_REP; | |
4726 | } else if (!strcmp(test, "crash-after-sending-execute-command-reply")) { | |
4727 | failure_test = FT_CRASH_AFTER_SEND_EXEC_REP; | |
4728 | } else if (!strcmp(test, "crash-before-sending-execute-command-request")) { | |
4729 | failure_test = FT_CRASH_BEFORE_SEND_EXEC_REQ; | |
4730 | } else if (!strcmp(test, "crash-after-sending-execute-command-request")) { | |
4731 | failure_test = FT_CRASH_AFTER_SEND_EXEC_REQ; | |
4732 | } else if (!strcmp(test, "crash-after-receiving-append-request-update")) { | |
4733 | failure_test = FT_CRASH_AFTER_RECV_APPEND_REQ_UPDATE; | |
4734 | } else if (!strcmp(test, "delay-election")) { | |
4735 | failure_test = FT_DELAY_ELECTION; | |
4736 | struct raft *raft; | |
4737 | HMAP_FOR_EACH (raft, hmap_node, &all_rafts) { | |
4738 | if (raft->role == RAFT_FOLLOWER) { | |
4739 | raft_reset_election_timer(raft); | |
4740 | } | |
4741 | } | |
93ee4209 HZ |
4742 | } else if (!strcmp(test, "dont-send-vote-request")) { |
4743 | failure_test = FT_DONT_SEND_VOTE_REQUEST; | |
67dba070 HZ |
4744 | } else if (!strcmp(test, "clear")) { |
4745 | failure_test = FT_NO_TEST; | |
4746 | unixctl_command_reply(conn, "test dismissed"); | |
4747 | return; | |
4748 | } else { | |
4749 | unixctl_command_reply_error(conn, "unknown test scenario"); | |
4750 | return; | |
4751 | } | |
4752 | unixctl_command_reply(conn, "test engaged"); | |
4753 | } | |
4754 | ||
1b1d2e6d BP |
4755 | static void |
4756 | raft_init(void) | |
4757 | { | |
4758 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
4759 | if (!ovsthread_once_start(&once)) { | |
4760 | return; | |
4761 | } | |
4762 | unixctl_command_register("cluster/cid", "DB", 1, 1, | |
4763 | raft_unixctl_cid, NULL); | |
4764 | unixctl_command_register("cluster/sid", "DB", 1, 1, | |
4765 | raft_unixctl_sid, NULL); | |
4766 | unixctl_command_register("cluster/status", "DB", 1, 1, | |
4767 | raft_unixctl_status, NULL); | |
4768 | unixctl_command_register("cluster/leave", "DB", 1, 1, | |
4769 | raft_unixctl_leave, NULL); | |
4770 | unixctl_command_register("cluster/kick", "DB SERVER", 2, 2, | |
4771 | raft_unixctl_kick, NULL); | |
8e354614 HZ |
4772 | unixctl_command_register("cluster/change-election-timer", "DB TIME", 2, 2, |
4773 | raft_unixctl_change_election_timer, NULL); | |
67dba070 HZ |
4774 | unixctl_command_register("cluster/failure-test", "FAILURE SCENARIO", 1, 1, |
4775 | raft_unixctl_failure_test, NULL); | |
1b1d2e6d BP |
4776 | ovsthread_once_done(&once); |
4777 | } |