patches/0006-CPG-callback-merging.patch

   1 From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
   2 From: =?UTF-8?q?Fabian=20Gr=C3=BCnbichler?= <f.gruenbichler@proxmox.com>
   3 Date: Wed, 22 May 2019 12:27:19 +0200
   4 Subject: [PATCH] CPG callback merging
   5 MIME-Version: 1.0
   6 Content-Type: text/plain; charset=UTF-8
   7 Content-Transfer-Encoding: 8bit
   8
   9 cherry-picked from upstream PR 468.
  10
  11 this series aims to minimize callback calls in case of cluster cold
  12 starts or re-joining after network partitions, which cause a huge spike
  13 in network load on bigger clusters running pmxcfs.
  14
  15 Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
  16 ---
  17  ...more-comments-to-notify_lib_joinlist.patch | 193 +++++++++++++++++
  18  ...illing-of-member_list-to-subfunction.patch | 130 ++++++++++++
  19  ...ify_lib_joinlist-drop-conn-parameter.patch | 112 ++++++++++
  20  ...-confchg-event-per-group-on-joinlist.patch | 199 ++++++++++++++++++
  21  debian/patches/series                         |   4 +
  22  5 files changed, 638 insertions(+)
  23  create mode 100644 debian/patches/cpg-Add-more-comments-to-notify_lib_joinlist.patch
  24  create mode 100644 debian/patches/cpg-Move-filling-of-member_list-to-subfunction.patch
  25  create mode 100644 debian/patches/cpg-notify_lib_joinlist-drop-conn-parameter.patch
  26  create mode 100644 debian/patches/cpg-send-single-confchg-event-per-group-on-joinlist.patch
  27
  28 diff --git a/debian/patches/cpg-Add-more-comments-to-notify_lib_joinlist.patch b/debian/patches/cpg-Add-more-comments-to-notify_lib_joinlist.patch
  29 new file mode 100644
  30 index 00000000..e32102fa
  31 --- /dev/null
  32 +++ b/debian/patches/cpg-Add-more-comments-to-notify_lib_joinlist.patch
  33 @@ -0,0 +1,193 @@
  34 +From: Jan Friesse <jfriesse@redhat.com>
  35 +Date: Wed, 15 May 2019 17:39:13 +0200
  36 +Subject: cpg: Add more comments to notify_lib_joinlist
  37 +
  38 +And make handling of left_list more generic. Also free skiplist
  39 +allocated by joinlist_inform_clients function. Last (but not least)
  40 +remove czechlish founded (should have been pp of "find").
  41 +
  42 +Signed-off-by: Jan Friesse <jfriesse@redhat.com>
  43 +(cherry picked from commit 10702c7e6c0cf8164ad8f05acb4e334461555973)
  44 +---
  45 + exec/cpg.c | 91 +++++++++++++++++++++++++++++++++++++++++---------------------
  46 + 1 file changed, 60 insertions(+), 31 deletions(-)
  47 +
  48 +diff --git a/exec/cpg.c b/exec/cpg.c
  49 +index e04ba57..2ecdc6b 100644
  50 +--- a/exec/cpg.c
  51 ++++ b/exec/cpg.c
  52 +@@ -663,31 +663,37 @@ static int notify_lib_joinlist(
  53 +       int size;
  54 +       char *buf;
  55 +       struct qb_list_head *iter;
  56 +-      int count;
  57 ++      int member_list_entries;
  58 +       struct res_lib_cpg_confchg_callback *res;
  59 +       mar_cpg_address_t *retgi;
  60 ++      int i;
  61 +
  62 +-      count = 0;
  63 ++      /*
  64 ++       * Find size of member_list (use process_info_list but remove items in left_list)
  65 ++       */
  66 ++      member_list_entries = 0;
  67 +
  68 +       qb_list_for_each(iter, &process_info_list_head) {
  69 +               struct process_info *pi = qb_list_entry (iter, struct process_info, list);
  70 ++
  71 +               if (mar_name_compare (&pi->group, group_name) == 0) {
  72 +-                      int i;
  73 +-                      int founded = 0;
  74 ++                      int in_left_list = 0;
  75 +
  76 +                       for (i = 0; i < left_list_entries; i++) {
  77 +                               if (left_list[i].nodeid == pi->nodeid && left_list[i].pid == pi->pid) {
  78 +-                                      founded++;
  79 ++                                      in_left_list = 1;
  80 ++                                      break ;
  81 +                               }
  82 +                       }
  83 +
  84 +-                      if (!founded)
  85 +-                              count++;
  86 ++                      if (!in_left_list) {
  87 ++                              member_list_entries++;
  88 ++                      }
  89 +               }
  90 +       }
  91 +
  92 +       size = sizeof(struct res_lib_cpg_confchg_callback) +
  93 +-              sizeof(mar_cpg_address_t) * (count + left_list_entries + joined_list_entries);
  94 ++              sizeof(mar_cpg_address_t) * (member_list_entries + left_list_entries + joined_list_entries);
  95 +       buf = alloca(size);
  96 +       if (!buf)
  97 +               return CS_ERR_LIBRARY;
  98 +@@ -695,27 +701,30 @@ static int notify_lib_joinlist(
  99 +       res = (struct res_lib_cpg_confchg_callback *)buf;
 100 +       res->joined_list_entries = joined_list_entries;
 101 +       res->left_list_entries = left_list_entries;
 102 +-      res->member_list_entries = count;
 103 ++      res->member_list_entries = member_list_entries;
 104 +       retgi = res->member_list;
 105 +       res->header.size = size;
 106 +       res->header.id = id;
 107 +       res->header.error = CS_OK;
 108 +       memcpy(&res->group_name, group_name, sizeof(mar_cpg_name_t));
 109 +
 110 ++      /*
 111 ++       * Fill res->memberlist. Use process_info_list but remove items in left_list.
 112 ++       */
 113 +       qb_list_for_each(iter, &process_info_list_head) {
 114 +-              struct process_info *pi=qb_list_entry (iter, struct process_info, list);
 115 ++              struct process_info *pi = qb_list_entry (iter, struct process_info, list);
 116 +
 117 +               if (mar_name_compare (&pi->group, group_name) == 0) {
 118 +-                      int i;
 119 +-                      int founded = 0;
 120 ++                      int in_left_list = 0;
 121 +
 122 +-                      for (i = 0;i < left_list_entries; i++) {
 123 ++                      for (i = 0; i < left_list_entries; i++) {
 124 +                               if (left_list[i].nodeid == pi->nodeid && left_list[i].pid == pi->pid) {
 125 +-                                      founded++;
 126 ++                                      in_left_list = 1;
 127 ++                                      break ;
 128 +                               }
 129 +                       }
 130 +
 131 +-                      if (!founded) {
 132 ++                      if (!in_left_list) {
 133 +                               retgi->nodeid = pi->nodeid;
 134 +                               retgi->pid = pi->pid;
 135 +                               retgi->reason = CPG_REASON_UNDEFINED;
 136 +@@ -724,23 +733,30 @@ static int notify_lib_joinlist(
 137 +               }
 138 +       }
 139 +
 140 ++      /*
 141 ++       * Fill res->left_list
 142 ++       */
 143 +       if (left_list_entries) {
 144 +               memcpy (retgi, left_list, left_list_entries * sizeof(mar_cpg_address_t));
 145 +               retgi += left_list_entries;
 146 +       }
 147 +
 148 +       if (joined_list_entries) {
 149 +-              int i;
 150 +-
 151 ++              /*
 152 ++               * Fill res->joined_list
 153 ++               */
 154 +               memcpy (retgi, joined_list, joined_list_entries * sizeof(mar_cpg_address_t));
 155 +               retgi += joined_list_entries;
 156 +
 157 +-              for (i=0; i < joined_list_entries; i++) {
 158 ++              /*
 159 ++               * Update cpd_state for all local joined processes in group
 160 ++               */
 161 ++              for (i = 0; i < joined_list_entries; i++) {
 162 +                       if (joined_list[i].nodeid == api->totem_nodeid_get()) {
 163 +-                      qb_list_for_each(iter, &cpg_pd_list_head) {
 164 ++                              qb_list_for_each(iter, &cpg_pd_list_head) {
 165 +                                       struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 166 +-                                      if (mar_name_compare (&cpd->group_name, group_name) == 0 &&
 167 +-                                          joined_list[i].pid == cpd->pid) {
 168 ++                                      if (joined_list[i].pid == cpd->pid &&
 169 ++                                          mar_name_compare (&cpd->group_name, group_name) == 0) {
 170 +                                               cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
 171 +                                       }
 172 +                               }
 173 +@@ -748,6 +764,9 @@ static int notify_lib_joinlist(
 174 +               }
 175 +       }
 176 +
 177 ++      /*
 178 ++       * Send notification to all ipc clients joined in group_name
 179 ++       */
 180 +       qb_list_for_each(iter, &cpg_pd_list_head) {
 181 +               struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 182 +               if (mar_name_compare (&cpd->group_name, group_name) == 0) {
 183 +@@ -760,16 +779,25 @@ static int notify_lib_joinlist(
 184 +               }
 185 +       }
 186 +
 187 +-      if (left_list_entries &&
 188 +-          left_list[0].nodeid == api->totem_nodeid_get() &&
 189 +-          left_list[0].reason == CONFCHG_CPG_REASON_LEAVE) {
 190 +-      qb_list_for_each(iter, &cpg_pd_list_head) {
 191 +-                      struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 192 +-                      if (mar_name_compare (&cpd->group_name, group_name) == 0 &&
 193 +-                          left_list[0].pid == cpd->pid) {
 194 +-                              cpd->pid = 0;
 195 +-                              memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 196 +-                              cpd->cpd_state = CPD_STATE_UNJOINED;
 197 ++      if (left_list_entries) {
 198 ++              /*
 199 ++               * Zero internal cpd state for all local processes leaving group
 200 ++               * (this loop is not strictly needed because left_list always either
 201 ++               *  contains exactly one process running on local node or more items
 202 ++               *  but none of them is running on local node)
 203 ++               */
 204 ++              for (i = 0; i < joined_list_entries; i++) {
 205 ++                      if (left_list[i].nodeid == api->totem_nodeid_get() &&
 206 ++                          left_list[i].reason == CONFCHG_CPG_REASON_LEAVE) {
 207 ++                              qb_list_for_each(iter, &cpg_pd_list_head) {
 208 ++                                      struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 209 ++                                      if (left_list[i].pid == cpd->pid &&
 210 ++                                          mar_name_compare (&cpd->group_name, group_name) == 0) {
 211 ++                                              cpd->pid = 0;
 212 ++                                              memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 213 ++                                              cpd->cpd_state = CPD_STATE_UNJOINED;
 214 ++                                      }
 215 ++                              }
 216 +                       }
 217 +               }
 218 +       }
 219 +@@ -966,6 +994,7 @@ static void joinlist_inform_clients (void)
 220 +               free(jld);
 221 +       }
 222 +       qb_map_iter_free(miter);
 223 ++      qb_map_destroy(group_notify_map);
 224 +
 225 +       joinlist_remove_zombie_pi_entries ();
 226 + }
 227 diff --git a/debian/patches/cpg-Move-filling-of-member_list-to-subfunction.patch b/debian/patches/cpg-Move-filling-of-member_list-to-subfunction.patch
 228 new file mode 100644
 229 index 00000000..9ec65353
 230 --- /dev/null
 231 +++ b/debian/patches/cpg-Move-filling-of-member_list-to-subfunction.patch
 232 @@ -0,0 +1,130 @@
 233 +From: Jan Friesse <jfriesse@redhat.com>
 234 +Date: Thu, 16 May 2019 14:08:25 +0200
 235 +Subject: cpg: Move filling of member_list to subfunction
 236 +
 237 +Signed-off-by: Jan Friesse <jfriesse@redhat.com>
 238 +(cherry picked from commit 0c1de94461368b4a924b516b77691e6ec471dcb0)
 239 +---
 240 + exec/cpg.c | 86 ++++++++++++++++++++++++++++++++++++--------------------------
 241 + 1 file changed, 50 insertions(+), 36 deletions(-)
 242 +
 243 +diff --git a/exec/cpg.c b/exec/cpg.c
 244 +index 2ecdc6b..98a16a6 100644
 245 +--- a/exec/cpg.c
 246 ++++ b/exec/cpg.c
 247 +@@ -652,26 +652,27 @@ static int notify_lib_totem_membership (
 248 +       return CS_OK;
 249 + }
 250 +
 251 +-static int notify_lib_joinlist(
 252 ++/*
 253 ++ * Helper function for notify_lib_joinlist which prepares member_list using
 254 ++ * process_info_list with removed left_list items.
 255 ++ * member_list_entries - When not NULL it contains number of member_list entries
 256 ++ * member_list - When not NULL it is used as pointer to start of preallocated
 257 ++ *               array of members. Pointer is adjusted to the end of array on
 258 ++ *               exit.
 259 ++ */
 260 ++static void notify_lib_joinlist_fill_member_list(
 261 +       const mar_cpg_name_t *group_name,
 262 +-      int joined_list_entries,
 263 +-      mar_cpg_address_t *joined_list,
 264 +       int left_list_entries,
 265 +-      mar_cpg_address_t *left_list,
 266 +-      int id)
 267 ++      const mar_cpg_address_t *left_list,
 268 ++      int *member_list_entries,
 269 ++      mar_cpg_address_t **member_list)
 270 + {
 271 +-      int size;
 272 +-      char *buf;
 273 +       struct qb_list_head *iter;
 274 +-      int member_list_entries;
 275 +-      struct res_lib_cpg_confchg_callback *res;
 276 +-      mar_cpg_address_t *retgi;
 277 +       int i;
 278 +
 279 +-      /*
 280 +-       * Find size of member_list (use process_info_list but remove items in left_list)
 281 +-       */
 282 +-      member_list_entries = 0;
 283 ++      if (member_list_entries != NULL) {
 284 ++              *member_list_entries = 0;
 285 ++      }
 286 +
 287 +       qb_list_for_each(iter, &process_info_list_head) {
 288 +               struct process_info *pi = qb_list_entry (iter, struct process_info, list);
 289 +@@ -687,10 +688,42 @@ static int notify_lib_joinlist(
 290 +                       }
 291 +
 292 +                       if (!in_left_list) {
 293 +-                              member_list_entries++;
 294 ++                              if (member_list_entries != NULL) {
 295 ++                                      (*member_list_entries)++;
 296 ++                              }
 297 ++
 298 ++                              if (member_list != NULL) {
 299 ++                                      (*member_list)->nodeid = pi->nodeid;
 300 ++                                      (*member_list)->pid = pi->pid;
 301 ++                                      (*member_list)->reason = CPG_REASON_UNDEFINED;
 302 ++                                      (*member_list)++;
 303 ++                              }
 304 +                       }
 305 +               }
 306 +       }
 307 ++}
 308 ++
 309 ++static int notify_lib_joinlist(
 310 ++      const mar_cpg_name_t *group_name,
 311 ++      int joined_list_entries,
 312 ++      mar_cpg_address_t *joined_list,
 313 ++      int left_list_entries,
 314 ++      mar_cpg_address_t *left_list,
 315 ++      int id)
 316 ++{
 317 ++      int size;
 318 ++      char *buf;
 319 ++      struct qb_list_head *iter;
 320 ++      int member_list_entries;
 321 ++      struct res_lib_cpg_confchg_callback *res;
 322 ++      mar_cpg_address_t *retgi;
 323 ++      int i;
 324 ++
 325 ++      /*
 326 ++       * Find size of member_list (use process_info_list but remove items in left_list)
 327 ++       */
 328 ++      notify_lib_joinlist_fill_member_list(group_name, left_list_entries, left_list,
 329 ++          &member_list_entries, NULL);
 330 +
 331 +       size = sizeof(struct res_lib_cpg_confchg_callback) +
 332 +               sizeof(mar_cpg_address_t) * (member_list_entries + left_list_entries + joined_list_entries);
 333 +@@ -711,27 +744,8 @@ static int notify_lib_joinlist(
 334 +       /*
 335 +        * Fill res->memberlist. Use process_info_list but remove items in left_list.
 336 +        */
 337 +-      qb_list_for_each(iter, &process_info_list_head) {
 338 +-              struct process_info *pi = qb_list_entry (iter, struct process_info, list);
 339 +-
 340 +-              if (mar_name_compare (&pi->group, group_name) == 0) {
 341 +-                      int in_left_list = 0;
 342 +-
 343 +-                      for (i = 0; i < left_list_entries; i++) {
 344 +-                              if (left_list[i].nodeid == pi->nodeid && left_list[i].pid == pi->pid) {
 345 +-                                      in_left_list = 1;
 346 +-                                      break ;
 347 +-                              }
 348 +-                      }
 349 +-
 350 +-                      if (!in_left_list) {
 351 +-                              retgi->nodeid = pi->nodeid;
 352 +-                              retgi->pid = pi->pid;
 353 +-                              retgi->reason = CPG_REASON_UNDEFINED;
 354 +-                              retgi++;
 355 +-                      }
 356 +-              }
 357 +-      }
 358 ++      notify_lib_joinlist_fill_member_list(group_name, left_list_entries, left_list,
 359 ++          NULL, &retgi);
 360 +
 361 +       /*
 362 +        * Fill res->left_list
 363 diff --git a/debian/patches/cpg-notify_lib_joinlist-drop-conn-parameter.patch b/debian/patches/cpg-notify_lib_joinlist-drop-conn-parameter.patch
 364 new file mode 100644
 365 index 00000000..e1742bf9
 366 --- /dev/null
 367 +++ b/debian/patches/cpg-notify_lib_joinlist-drop-conn-parameter.patch
 368 @@ -0,0 +1,112 @@
 369 +From: =?utf-8?q?Fabian_Gr=C3=BCnbichler?= <f.gruenbichler@proxmox.com>
 370 +Date: Wed, 15 May 2019 13:45:13 +0200
 371 +Subject: cpg: notify_lib_joinlist: drop conn parameter
 372 +
 373 +since it is always set to NULL.
 374 +
 375 +(cherry picked from commit d5e9723695b2ae63162b238300013d785e875ea7)
 376 +---
 377 + exec/cpg.c | 54 ++++++++++++++++++++++++------------------------------
 378 + 1 file changed, 24 insertions(+), 30 deletions(-)
 379 +
 380 +diff --git a/exec/cpg.c b/exec/cpg.c
 381 +index e39ca34..b752cc1 100644
 382 +--- a/exec/cpg.c
 383 ++++ b/exec/cpg.c
 384 +@@ -647,7 +647,6 @@ static int notify_lib_totem_membership (
 385 +
 386 + static int notify_lib_joinlist(
 387 +       const mar_cpg_name_t *group_name,
 388 +-      void *conn,
 389 +       int joined_list_entries,
 390 +       mar_cpg_address_t *joined_list,
 391 +       int left_list_entries,
 392 +@@ -728,40 +727,35 @@ static int notify_lib_joinlist(
 393 +               retgi += joined_list_entries;
 394 +       }
 395 +
 396 +-      if (conn) {
 397 +-              api->ipc_dispatch_send (conn, buf, size);
 398 +-      } else {
 399 +       qb_list_for_each(iter, &cpg_pd_list_head) {
 400 +-                      struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 401 +-                      if (mar_name_compare (&cpd->group_name, group_name) == 0) {
 402 +-                              assert (joined_list_entries <= 1);
 403 +-                              if (joined_list_entries) {
 404 +-                                      if (joined_list[0].pid == cpd->pid &&
 405 +-                                              joined_list[0].nodeid == api->totem_nodeid_get()) {
 406 +-                                              cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
 407 +-                                      }
 408 ++              struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 409 ++              if (mar_name_compare (&cpd->group_name, group_name) == 0) {
 410 ++                      assert (joined_list_entries <= 1);
 411 ++                      if (joined_list_entries) {
 412 ++                              if (joined_list[0].pid == cpd->pid &&
 413 ++                                      joined_list[0].nodeid == api->totem_nodeid_get()) {
 414 ++                                      cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
 415 +                               }
 416 +-                              if (cpd->cpd_state == CPD_STATE_JOIN_COMPLETED ||
 417 +-                                      cpd->cpd_state == CPD_STATE_LEAVE_STARTED) {
 418 ++                      }
 419 ++                      if (cpd->cpd_state == CPD_STATE_JOIN_COMPLETED ||
 420 ++                              cpd->cpd_state == CPD_STATE_LEAVE_STARTED) {
 421 +
 422 +-                                      api->ipc_dispatch_send (cpd->conn, buf, size);
 423 +-                                      cpd->transition_counter++;
 424 +-                              }
 425 +-                              if (left_list_entries) {
 426 +-                                      if (left_list[0].pid == cpd->pid &&
 427 +-                                              left_list[0].nodeid == api->totem_nodeid_get() &&
 428 +-                                              left_list[0].reason == CONFCHG_CPG_REASON_LEAVE) {
 429 +-
 430 +-                                              cpd->pid = 0;
 431 +-                                              memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 432 +-                                              cpd->cpd_state = CPD_STATE_UNJOINED;
 433 +-                                      }
 434 ++                              api->ipc_dispatch_send (cpd->conn, buf, size);
 435 ++                              cpd->transition_counter++;
 436 ++                      }
 437 ++                      if (left_list_entries) {
 438 ++                              if (left_list[0].pid == cpd->pid &&
 439 ++                                      left_list[0].nodeid == api->totem_nodeid_get() &&
 440 ++                                      left_list[0].reason == CONFCHG_CPG_REASON_LEAVE) {
 441 ++
 442 ++                                      cpd->pid = 0;
 443 ++                                      memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 444 ++                                      cpd->cpd_state = CPD_STATE_UNJOINED;
 445 +                               }
 446 +                       }
 447 +               }
 448 +       }
 449 +
 450 +-
 451 +       /*
 452 +        * Traverse thru cpds and send totem membership for cpd, where it is not send yet
 453 +        */
 454 +@@ -858,7 +852,7 @@ static void downlist_inform_clients (void)
 455 +               }
 456 +
 457 +               /* send confchg event */
 458 +-              notify_lib_joinlist(&group, NULL,
 459 ++              notify_lib_joinlist(&group,
 460 +                       0, NULL,
 461 +                       pcd->left_list_entries,
 462 +                       pcd->left_list,
 463 +@@ -1157,7 +1151,7 @@ static void do_proc_join(
 464 +       notify_info.nodeid = nodeid;
 465 +       notify_info.reason = reason;
 466 +
 467 +-      notify_lib_joinlist(&pi->group, NULL,
 468 ++      notify_lib_joinlist(&pi->group,
 469 +                           1, &notify_info,
 470 +                           0, NULL,
 471 +                           MESSAGE_RES_CPG_CONFCHG_CALLBACK);
 472 +@@ -1177,7 +1171,7 @@ static void do_proc_leave(
 473 +       notify_info.nodeid = nodeid;
 474 +       notify_info.reason = reason;
 475 +
 476 +-      notify_lib_joinlist(name, NULL,
 477 ++      notify_lib_joinlist(name,
 478 +               0, NULL,
 479 +               1, &notify_info,
 480 +               MESSAGE_RES_CPG_CONFCHG_CALLBACK);
 481 diff --git a/debian/patches/cpg-send-single-confchg-event-per-group-on-joinlist.patch b/debian/patches/cpg-send-single-confchg-event-per-group-on-joinlist.patch
 482 new file mode 100644
 483 index 00000000..013baaa0
 484 --- /dev/null
 485 +++ b/debian/patches/cpg-send-single-confchg-event-per-group-on-joinlist.patch
 486 @@ -0,0 +1,199 @@
 487 +From: =?utf-8?q?Fabian_Gr=C3=BCnbichler?= <f.gruenbichler@proxmox.com>
 488 +Date: Wed, 8 May 2019 16:31:15 +0200
 489 +Subject: cpg: send single confchg event per group on joinlist
 490 +MIME-Version: 1.0
 491 +Content-Type: text/plain; charset="utf-8"
 492 +Content-Transfer-Encoding: 8bit
 493 +
 494 +using a similar approach to
 495 +
 496 +43bead364514e8ae2ba00bcf07c460e31d0b1765 "Send one confchg event per CPG group to CPG client"
 497 +
 498 +which did the same for leave events on a network partition.
 499 +
 500 +Fixes: #351
 501 +
 502 +Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
 503 +(cherry picked from commit 9c60289b831ef0b138142c3fc39c9a9da8a2ada2)
 504 +---
 505 + exec/cpg.c | 97 ++++++++++++++++++++++++++++++++++++++++++++++----------------
 506 + 1 file changed, 73 insertions(+), 24 deletions(-)
 507 +
 508 +diff --git a/exec/cpg.c b/exec/cpg.c
 509 +index b752cc1..e04ba57 100644
 510 +--- a/exec/cpg.c
 511 ++++ b/exec/cpg.c
 512 +@@ -194,6 +194,12 @@ struct join_list_entry {
 513 +       mar_cpg_name_t group_name;
 514 + };
 515 +
 516 ++struct join_list_confchg_data {
 517 ++      mar_cpg_name_t cpg_group;
 518 ++      mar_cpg_address_t join_list[CPG_MEMBERS_MAX];
 519 ++      int join_list_entries;
 520 ++};
 521 ++
 522 + /*
 523 +  * Service Interfaces required by service_message_handler struct
 524 +  */
 525 +@@ -312,7 +318,8 @@ static void do_proc_join(
 526 +       const mar_cpg_name_t *name,
 527 +       uint32_t pid,
 528 +       unsigned int nodeid,
 529 +-      int reason);
 530 ++      int reason,
 531 ++      qb_map_t *group_notify_map);
 532 +
 533 + static void do_proc_leave(
 534 +       const mar_cpg_name_t *name,
 535 +@@ -723,35 +730,46 @@ static int notify_lib_joinlist(
 536 +       }
 537 +
 538 +       if (joined_list_entries) {
 539 ++              int i;
 540 ++
 541 +               memcpy (retgi, joined_list, joined_list_entries * sizeof(mar_cpg_address_t));
 542 +               retgi += joined_list_entries;
 543 ++
 544 ++              for (i=0; i < joined_list_entries; i++) {
 545 ++                      if (joined_list[i].nodeid == api->totem_nodeid_get()) {
 546 ++                      qb_list_for_each(iter, &cpg_pd_list_head) {
 547 ++                                      struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 548 ++                                      if (mar_name_compare (&cpd->group_name, group_name) == 0 &&
 549 ++                                          joined_list[i].pid == cpd->pid) {
 550 ++                                              cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
 551 ++                                      }
 552 ++                              }
 553 ++                      }
 554 ++              }
 555 +       }
 556 +
 557 +       qb_list_for_each(iter, &cpg_pd_list_head) {
 558 +               struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 559 +               if (mar_name_compare (&cpd->group_name, group_name) == 0) {
 560 +-                      assert (joined_list_entries <= 1);
 561 +-                      if (joined_list_entries) {
 562 +-                              if (joined_list[0].pid == cpd->pid &&
 563 +-                                      joined_list[0].nodeid == api->totem_nodeid_get()) {
 564 +-                                      cpd->cpd_state = CPD_STATE_JOIN_COMPLETED;
 565 +-                              }
 566 +-                      }
 567 +                       if (cpd->cpd_state == CPD_STATE_JOIN_COMPLETED ||
 568 +                               cpd->cpd_state == CPD_STATE_LEAVE_STARTED) {
 569 +
 570 +                               api->ipc_dispatch_send (cpd->conn, buf, size);
 571 +                               cpd->transition_counter++;
 572 +                       }
 573 +-                      if (left_list_entries) {
 574 +-                              if (left_list[0].pid == cpd->pid &&
 575 +-                                      left_list[0].nodeid == api->totem_nodeid_get() &&
 576 +-                                      left_list[0].reason == CONFCHG_CPG_REASON_LEAVE) {
 577 +-
 578 +-                                      cpd->pid = 0;
 579 +-                                      memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 580 +-                                      cpd->cpd_state = CPD_STATE_UNJOINED;
 581 +-                              }
 582 ++              }
 583 ++      }
 584 ++
 585 ++      if (left_list_entries &&
 586 ++          left_list[0].nodeid == api->totem_nodeid_get() &&
 587 ++          left_list[0].reason == CONFCHG_CPG_REASON_LEAVE) {
 588 ++      qb_list_for_each(iter, &cpg_pd_list_head) {
 589 ++                      struct cpg_pd *cpd = qb_list_entry (iter, struct cpg_pd, list);
 590 ++                      if (mar_name_compare (&cpd->group_name, group_name) == 0 &&
 591 ++                          left_list[0].pid == cpd->pid) {
 592 ++                              cpd->pid = 0;
 593 ++                              memset (&cpd->group_name, 0, sizeof(cpd->group_name));
 594 ++                              cpd->cpd_state = CPD_STATE_UNJOINED;
 595 +                       }
 596 +               }
 597 +       }
 598 +@@ -915,6 +933,11 @@ static void joinlist_inform_clients (void)
 599 +       struct joinlist_msg *stored_msg;
 600 +       struct qb_list_head *iter;
 601 +       unsigned int i;
 602 ++      qb_map_t *group_notify_map;
 603 ++      qb_map_iter_t *miter;
 604 ++      struct join_list_confchg_data *jld;
 605 ++
 606 ++      group_notify_map = qb_skiplist_create();
 607 +
 608 +       i = 0;
 609 +       qb_list_for_each(iter, &joinlist_messages_head) {
 610 +@@ -931,9 +954,19 @@ static void joinlist_inform_clients (void)
 611 +               }
 612 +
 613 +               do_proc_join (&stored_msg->group_name, stored_msg->pid, stored_msg->sender_nodeid,
 614 +-                      CONFCHG_CPG_REASON_NODEUP);
 615 ++                      CONFCHG_CPG_REASON_NODEUP, group_notify_map);
 616 +       }
 617 +
 618 ++      miter = qb_map_iter_create(group_notify_map);
 619 ++      while (qb_map_iter_next(miter, (void **)&jld)) {
 620 ++              notify_lib_joinlist(&jld->cpg_group,
 621 ++                                  jld->join_list_entries, jld->join_list,
 622 ++                                  0, NULL,
 623 ++                                  MESSAGE_RES_CPG_CONFCHG_CALLBACK);
 624 ++              free(jld);
 625 ++      }
 626 ++      qb_map_iter_free(miter);
 627 ++
 628 +       joinlist_remove_zombie_pi_entries ();
 629 + }
 630 +
 631 +@@ -1111,13 +1144,15 @@ static void do_proc_join(
 632 +       const mar_cpg_name_t *name,
 633 +       uint32_t pid,
 634 +       unsigned int nodeid,
 635 +-      int reason)
 636 ++      int reason,
 637 ++      qb_map_t *group_notify_map)
 638 + {
 639 +       struct process_info *pi;
 640 +       struct process_info *pi_entry;
 641 +       mar_cpg_address_t notify_info;
 642 +       struct qb_list_head *list;
 643 +       struct qb_list_head *list_to_add = NULL;
 644 ++      int size;
 645 +
 646 +       if (process_info_find (name, pid, nodeid) != NULL) {
 647 +               return ;
 648 +@@ -1151,10 +1186,24 @@ static void do_proc_join(
 649 +       notify_info.nodeid = nodeid;
 650 +       notify_info.reason = reason;
 651 +
 652 +-      notify_lib_joinlist(&pi->group,
 653 +-                          1, &notify_info,
 654 +-                          0, NULL,
 655 +-                          MESSAGE_RES_CPG_CONFCHG_CALLBACK);
 656 ++      if (group_notify_map == NULL) {
 657 ++              notify_lib_joinlist(&pi->group,
 658 ++                                  1, &notify_info,
 659 ++                                  0, NULL,
 660 ++                                  MESSAGE_RES_CPG_CONFCHG_CALLBACK);
 661 ++      } else {
 662 ++              struct join_list_confchg_data *jld = qb_map_get(group_notify_map, pi->group.value);
 663 ++              if (jld == NULL) {
 664 ++                      jld = (struct join_list_confchg_data *)calloc(1, sizeof(struct join_list_confchg_data));
 665 ++                      memcpy(&jld->cpg_group, &pi->group, sizeof(mar_cpg_name_t));
 666 ++                      qb_map_put(group_notify_map, jld->cpg_group.value, jld);
 667 ++              }
 668 ++              size = jld->join_list_entries;
 669 ++              jld->join_list[size].nodeid = notify_info.nodeid;
 670 ++              jld->join_list[size].pid = notify_info.pid;
 671 ++              jld->join_list[size].reason = notify_info.reason;
 672 ++              jld->join_list_entries++;
 673 ++      }
 674 + }
 675 +
 676 + static void do_proc_leave(
 677 +@@ -1219,7 +1268,7 @@ static void message_handler_req_exec_cpg_procjoin (
 678 +
 679 +       do_proc_join (&req_exec_cpg_procjoin->group_name,
 680 +               req_exec_cpg_procjoin->pid, nodeid,
 681 +-              CONFCHG_CPG_REASON_JOIN);
 682 ++              CONFCHG_CPG_REASON_JOIN, NULL);
 683 + }
 684 +
 685 + static void message_handler_req_exec_cpg_procleave (
 686 diff --git a/debian/patches/series b/debian/patches/series
 687 index 803038e4..fec11c03 100644
 688 --- a/debian/patches/series
 689 +++ b/debian/patches/series
 690 @@ -52,3 +52,7 @@ vqsim-Check-length-of-copied-optarg.patch
 691  Fix-various-typos-in-ChangeLog.patch
 692  Enable-PrivateTmp-in-the-systemd-service-files.patch
 693  pve__only-start-corosync.service-if-conf-exists.patch
 694 +cpg-notify_lib_joinlist-drop-conn-parameter.patch
 695 +cpg-send-single-confchg-event-per-group-on-joinlist.patch
 696 +cpg-Add-more-comments-to-notify_lib_joinlist.patch
 697 +cpg-Move-filling-of-member_list-to-subfunction.patch