]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | ====================== |
2 | Peering | |
3 | ====================== | |
4 | ||
5 | Concepts | |
6 | -------- | |
7 | ||
8 | *Peering* | |
9 | the process of bringing all of the OSDs that store | |
10 | a Placement Group (PG) into agreement about the state | |
11 | of all of the objects (and their metadata) in that PG. | |
12 | Note that agreeing on the state does not mean that | |
13 | they all have the latest contents. | |
14 | ||
15 | *Acting set* | |
16 | the ordered list of OSDs who are (or were as of some epoch) | |
17 | responsible for a particular PG. | |
18 | ||
19 | *Up set* | |
20 | the ordered list of OSDs responsible for a particular PG for | |
21 | a particular epoch according to CRUSH. Normally this | |
22 | is the same as the *acting set*, except when the *acting set* has been | |
23 | explicitly overridden via *PG temp* in the OSDMap. | |
24 | ||
25 | *PG temp* | |
26 | a temporary placement group acting set used while backfilling the | |
27 | primary osd. Let say acting is [0,1,2] and we are | |
28 | active+clean. Something happens and acting is now [3,1,2]. osd 3 is | |
29 | empty and can't serve reads although it is the primary. osd.3 will | |
30 | see that and request a *PG temp* of [1,2,3] to the monitors using a | |
31 | MOSDPGTemp message so that osd.1 temporarily becomes the | |
32 | primary. It will select osd.3 as a backfill peer and continue to | |
33 | serve reads and writes while osd.3 is backfilled. When backfilling | |
34 | is complete, *PG temp* is discarded and the acting set changes back | |
35 | to [3,1,2] and osd.3 becomes the primary. | |
36 | ||
37 | *current interval* or *past interval* | |
38 | a sequence of OSD map epochs during which the *acting set* and *up | |
39 | set* for particular PG do not change | |
40 | ||
41 | *primary* | |
42 | the (by convention first) member of the *acting set*, | |
43 | who is responsible for coordination peering, and is | |
44 | the only OSD that will accept client initiated | |
45 | writes to objects in a placement group. | |
46 | ||
47 | *replica* | |
48 | a non-primary OSD in the *acting set* for a placement group | |
49 | (and who has been recognized as such and *activated* by the primary). | |
50 | ||
51 | *stray* | |
52 | an OSD who is not a member of the current *acting set*, but | |
53 | has not yet been told that it can delete its copies of a | |
54 | particular placement group. | |
55 | ||
56 | *recovery* | |
57 | ensuring that copies of all of the objects in a PG | |
58 | are on all of the OSDs in the *acting set*. Once | |
59 | *peering* has been performed, the primary can start | |
60 | accepting write operations, and *recovery* can proceed | |
61 | in the background. | |
62 | ||
63 | *PG info* basic metadata about the PG's creation epoch, the version | |
64 | for the most recent write to the PG, *last epoch started*, *last | |
65 | epoch clean*, and the beginning of the *current interval*. Any | |
66 | inter-OSD communication about PGs includes the *PG info*, such that | |
67 | any OSD that knows a PG exists (or once existed) also has a lower | |
68 | bound on *last epoch clean* or *last epoch started*. | |
69 | ||
70 | *PG log* | |
71 | a list of recent updates made to objects in a PG. | |
72 | Note that these logs can be truncated after all OSDs | |
73 | in the *acting set* have acknowledged up to a certain | |
74 | point. | |
75 | ||
76 | *missing set* | |
77 | Each OSD notes update log entries and if they imply updates to | |
78 | the contents of an object, adds that object to a list of needed | |
79 | updates. This list is called the *missing set* for that <OSD,PG>. | |
80 | ||
81 | *Authoritative History* | |
82 | a complete, and fully ordered set of operations that, if | |
83 | performed, would bring an OSD's copy of a Placement Group | |
84 | up to date. | |
85 | ||
86 | *epoch* | |
87 | a (monotonically increasing) OSD map version number | |
88 | ||
89 | *last epoch start* | |
90 | the last epoch at which all nodes in the *acting set* | |
91 | for a particular placement group agreed on an | |
92 | *authoritative history*. At this point, *peering* is | |
93 | deemed to have been successful. | |
94 | ||
95 | *up_thru* | |
96 | before a primary can successfully complete the *peering* process, | |
97 | it must inform a monitor that is alive through the current | |
98 | OSD map epoch by having the monitor set its *up_thru* in the osd | |
99 | map. This helps peering ignore previous *acting sets* for which | |
100 | peering never completed after certain sequences of failures, such as | |
101 | the second interval below: | |
102 | ||
103 | - *acting set* = [A,B] | |
104 | - *acting set* = [A] | |
105 | - *acting set* = [] very shortly after (e.g., simultaneous failure, but staggered detection) | |
106 | - *acting set* = [B] (B restarts, A does not) | |
107 | ||
108 | *last epoch clean* | |
109 | the last epoch at which all nodes in the *acting set* | |
110 | for a particular placement group were completely | |
111 | up to date (both PG logs and object contents). | |
112 | At this point, *recovery* is deemed to have been | |
113 | completed. | |
114 | ||
115 | Description of the Peering Process | |
116 | ---------------------------------- | |
117 | ||
118 | The *Golden Rule* is that no write operation to any PG | |
119 | is acknowledged to a client until it has been persisted | |
120 | by all members of the *acting set* for that PG. This means | |
121 | that if we can communicate with at least one member of | |
122 | each *acting set* since the last successful *peering*, someone | |
123 | will have a record of every (acknowledged) operation | |
124 | since the last successful *peering*. | |
125 | This means that it should be possible for the current | |
126 | primary to construct and disseminate a new *authoritative history*. | |
127 | ||
128 | It is also important to appreciate the role of the OSD map | |
129 | (list of all known OSDs and their states, as well as some | |
130 | information about the placement groups) in the *peering* | |
131 | process: | |
132 | ||
133 | When OSDs go up or down (or get added or removed) | |
134 | this has the potential to affect the *active sets* | |
135 | of many placement groups. | |
136 | ||
137 | Before a primary successfully completes the *peering* | |
138 | process, the OSD map must reflect that the OSD was alive | |
139 | and well as of the first epoch in the *current interval*. | |
140 | ||
141 | Changes can only be made after successful *peering*. | |
142 | ||
143 | Thus, a new primary can use the latest OSD map along with a recent | |
144 | history of past maps to generate a set of *past intervals* to | |
145 | determine which OSDs must be consulted before we can successfully | |
146 | *peer*. The set of past intervals is bounded by *last epoch started*, | |
147 | the most recent *past interval* for which we know *peering* completed. | |
148 | The process by which an OSD discovers a PG exists in the first place is | |
149 | by exchanging *PG info* messages, so the OSD always has some lower | |
150 | bound on *last epoch started*. | |
151 | ||
152 | The high level process is for the current PG primary to: | |
153 | ||
154 | 1. get a recent OSD map (to identify the members of the all | |
155 | interesting *acting sets*, and confirm that we are still the | |
156 | primary). | |
157 | ||
158 | #. generate a list of *past intervals* since *last epoch started*. | |
159 | Consider the subset of those for which *up_thru* was greater than | |
160 | the first interval epoch by the last interval epoch's OSD map; that is, | |
161 | the subset for which *peering* could have completed before the *acting | |
162 | set* changed to another set of OSDs. | |
163 | ||
164 | Successful *peering* will require that we be able to contact at | |
165 | least one OSD from each of *past interval*'s *acting set*. | |
166 | ||
167 | #. ask every node in that list for its *PG info*, which includes the most | |
168 | recent write made to the PG, and a value for *last epoch started*. If | |
169 | we learn about a *last epoch started* that is newer than our own, we can | |
170 | prune older *past intervals* and reduce the peer OSDs we need to contact. | |
171 | ||
172 | #. if anyone else has (in its PG log) operations that I do not have, | |
173 | instruct them to send me the missing log entries so that the primary's | |
174 | *PG log* is up to date (includes the newest write).. | |
175 | ||
176 | #. for each member of the current *acting set*: | |
177 | ||
178 | a. ask it for copies of all PG log entries since *last epoch start* | |
179 | so that I can verify that they agree with mine (or know what | |
180 | objects I will be telling it to delete). | |
181 | ||
182 | If the cluster failed before an operation was persisted by all | |
183 | members of the *acting set*, and the subsequent *peering* did not | |
184 | remember that operation, and a node that did remember that | |
185 | operation later rejoined, its logs would record a different | |
186 | (divergent) history than the *authoritative history* that was | |
187 | reconstructed in the *peering* after the failure. | |
188 | ||
189 | Since the *divergent* events were not recorded in other logs | |
190 | from that *acting set*, they were not acknowledged to the client, | |
191 | and there is no harm in discarding them (so that all OSDs agree | |
192 | on the *authoritative history*). But, we will have to instruct | |
193 | any OSD that stores data from a divergent update to delete the | |
194 | affected (and now deemed to be apocryphal) objects. | |
195 | ||
196 | #. ask it for its *missing set* (object updates recorded | |
197 | in its PG log, but for which it does not have the new data). | |
198 | This is the list of objects that must be fully replicated | |
199 | before we can accept writes. | |
200 | ||
201 | #. at this point, the primary's PG log contains an *authoritative history* of | |
202 | the placement group, and the OSD now has sufficient | |
203 | information to bring any other OSD in the *acting set* up to date. | |
204 | ||
205 | #. if the primary's *up_thru* value in the current OSD map is not greater than | |
206 | or equal to the first epoch in the *current interval*, send a request to the | |
207 | monitor to update it, and wait until receive an updated OSD map that reflects | |
208 | the change. | |
209 | ||
210 | #. for each member of the current *acting set*: | |
211 | ||
212 | a. send them log updates to bring their PG logs into agreement with | |
213 | my own (*authoritative history*) ... which may involve deciding | |
214 | to delete divergent objects. | |
215 | ||
216 | #. await acknowledgment that they have persisted the PG log entries. | |
217 | ||
218 | #. at this point all OSDs in the *acting set* agree on all of the meta-data, | |
219 | and would (in any future *peering*) return identical accounts of all | |
220 | updates. | |
221 | ||
222 | a. start accepting client write operations (because we have unanimous | |
223 | agreement on the state of the objects into which those updates are | |
224 | being accepted). Note, however, that if a client tries to write to an | |
225 | object it will be promoted to the front of the recovery queue, and the | |
226 | write willy be applied after it is fully replicated to the current *acting set*. | |
227 | ||
228 | #. update the *last epoch started* value in our local *PG info*, and instruct | |
229 | other *active set* OSDs to do the same. | |
230 | ||
231 | #. start pulling object data updates that other OSDs have, but I do not. We may | |
232 | need to query OSDs from additional *past intervals* prior to *last epoch started* | |
233 | (the last time *peering* completed) and following *last epoch clean* (the last epoch that | |
234 | recovery completed) in order to find copies of all objects. | |
235 | ||
236 | #. start pushing object data updates to other OSDs that do not yet have them. | |
237 | ||
238 | We push these updates from the primary (rather than having the replicas | |
239 | pull them) because this allows the primary to ensure that a replica has | |
240 | the current contents before sending it an update write. It also makes | |
241 | it possible for a single read (from the primary) to be used to write | |
242 | the data to multiple replicas. If each replica did its own pulls, | |
243 | the data might have to be read multiple times. | |
244 | ||
245 | #. once all replicas store the all copies of all objects (that | |
246 | existed prior to the start of this epoch) we can update *last | |
247 | epoch clean* in the *PG info*, and we can dismiss all of the | |
248 | *stray* replicas, allowing them to delete their copies of objects | |
249 | for which they are no longer in the *acting set*. | |
250 | ||
251 | We could not dismiss the *strays* prior to this because it was possible | |
252 | that one of those *strays* might hold the sole surviving copy of an | |
253 | old object (all of whose copies disappeared before they could be | |
254 | replicated on members of the current *acting set*). | |
255 | ||
256 | State Model | |
257 | ----------- | |
258 | ||
259 | .. graphviz:: peering_graph.generated.dot |