daemons/controld/controld_join

/* */
This source file includes following definitions.
controld_destroy_failed_sync_table
controld_remove_failed_sync_node
record_failed_sync_node
lookup_failed_sync_node
crm_update_peer_join
start_join_round
create_dc_message
join_make_offer
do_dc_join_offer_all
do_dc_join_offer_one
compare_int_fields
do_dc_join_filter_offer
do_dc_join_finalize
free_max_generation
finalize_sync_callback
join_node_state_commit_callback
do_dc_join_ack
finalize_join_for
check_join_state
do_dc_join_final
crmd_join_phase_count
crmd_join_phase_log
   1 /*
   2  * Copyright 2004-2024 the Pacemaker project contributors
   3  *
   4  * The version control history for this file may have further details.
   5  *
   6  * This source code is licensed under the GNU General Public License version 2
   7  * or later (GPLv2+) WITHOUT ANY WARRANTY.
   8  */
   9 
  10 #include <crm_internal.h>
  11 
  12 #include <crm/crm.h>
  13 
  14 #include <crm/common/xml.h>
  15 #include <crm/cluster.h>
  16 
  17 #include <pacemaker-controld.h>
  18 
  19 static char *max_generation_from = NULL;
  20 static xmlNodePtr max_generation_xml = NULL;
  21 
  22 /*!
  23  * \internal
  24  * \brief Nodes from which a CIB sync has failed since the peer joined
  25  *
  26  * This table is of the form (<tt>node_name -> join_id</tt>). \p node_name is
  27  * the name of a client node from which a CIB \p sync_from() call has failed in
  28  * \p do_dc_join_finalize() since the client joined the cluster as a peer.
  29  * \p join_id is the ID of the join round in which the \p sync_from() failed,
  30  * and is intended for use in nack log messages.
  31  */
  32 static GHashTable *failed_sync_nodes = NULL;
  33 
  34 void finalize_join_for(gpointer key, gpointer value, gpointer user_data);
  35 void finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data);
  36 gboolean check_join_state(enum crmd_fsa_state cur_state, const char *source);
  37 
  38 /* Numeric counter used to identify join rounds (an unsigned int would be
  39  * appropriate, except we get and set it in XML as int)
  40  */
  41 static int current_join_id = 0;
  42 
  43 /*!
  44  * \internal
  45  * \brief Destroy the hash table containing failed sync nodes
  46  */
  47 void
  48 controld_destroy_failed_sync_table(void)
     /*  */
  49 {
  50     if (failed_sync_nodes != NULL) {
  51         g_hash_table_destroy(failed_sync_nodes);
  52         failed_sync_nodes = NULL;
  53     }
  54 }
  55 
  56 /*!
  57  * \internal
  58  * \brief Remove a node from the failed sync nodes table if present
  59  *
  60  * \param[in] node_name  Node name to remove
  61  */
  62 void
  63 controld_remove_failed_sync_node(const char *node_name)
     /*  */
  64 {
  65     if (failed_sync_nodes != NULL) {
  66         g_hash_table_remove(failed_sync_nodes, (gchar *) node_name);
  67     }
  68 }
  69 
  70 /*!
  71  * \internal
  72  * \brief Add to a hash table a node whose CIB failed to sync
  73  *
  74  * \param[in] node_name  Name of node whose CIB failed to sync
  75  * \param[in] join_id    Join round when the failure occurred
  76  */
  77 static void
  78 record_failed_sync_node(const char *node_name, gint join_id)
     /*  */
  79 {
  80     if (failed_sync_nodes == NULL) {
  81         failed_sync_nodes = pcmk__strikey_table(g_free, NULL);
  82     }
  83 
  84     /* If the node is already in the table then we failed to nack it during the
  85      * filter offer step
  86      */
  87     CRM_LOG_ASSERT(g_hash_table_insert(failed_sync_nodes, g_strdup(node_name),
  88                                        GINT_TO_POINTER(join_id)));
  89 }
  90 
  91 /*!
  92  * \internal
  93  * \brief Look up a node name in the failed sync table
  94  *
  95  * \param[in]  node_name  Name of node to look up
  96  * \param[out] join_id    Where to store the join ID of when the sync failed
  97  *
  98  * \return Standard Pacemaker return code. Specifically, \p pcmk_rc_ok if the
  99  *         node name was found, or \p pcmk_rc_node_unknown otherwise.
 100  * \note \p *join_id is set to -1 if the node is not found.
 101  */
 102 static int
 103 lookup_failed_sync_node(const char *node_name, gint *join_id)
     /*  */
 104 {
 105     *join_id = -1;
 106 
 107     if (failed_sync_nodes != NULL) {
 108         gpointer result = g_hash_table_lookup(failed_sync_nodes,
 109                                               (gchar *) node_name);
 110         if (result != NULL) {
 111             *join_id = GPOINTER_TO_INT(result);
 112             return pcmk_rc_ok;
 113         }
 114     }
 115     return pcmk_rc_node_unknown;
 116 }
 117 
 118 void
 119 crm_update_peer_join(const char *source, crm_node_t * node, enum crm_join_phase phase)
     /*  */
 120 {
 121     enum crm_join_phase last = 0;
 122 
 123     CRM_CHECK(node != NULL, return);
 124 
 125     /* Remote nodes do not participate in joins */
 126     if (pcmk_is_set(node->flags, crm_remote_node)) {
 127         return;
 128     }
 129 
 130     last = node->join;
 131 
 132     if(phase == last) {
 133         crm_trace("Node %s join-%d phase is still %s "
 134                   CRM_XS " nodeid=%u source=%s",
 135                   node->uname, current_join_id, crm_join_phase_str(last),
 136                   node->id, source);
 137 
 138     } else if ((phase <= crm_join_none) || (phase == (last + 1))) {
 139         node->join = phase;
 140         crm_trace("Node %s join-%d phase is now %s (was %s) "
 141                   CRM_XS " nodeid=%u source=%s",
 142                  node->uname, current_join_id, crm_join_phase_str(phase),
 143                  crm_join_phase_str(last), node->id, source);
 144 
 145     } else {
 146         crm_warn("Rejecting join-%d phase update for node %s because "
 147                  "can't go from %s to %s " CRM_XS " nodeid=%u source=%s",
 148                  current_join_id, node->uname, crm_join_phase_str(last),
 149                  crm_join_phase_str(phase), node->id, source);
 150     }
 151 }
 152 
 153 static void
 154 start_join_round(void)
     /*  */
 155 {
 156     GHashTableIter iter;
 157     crm_node_t *peer = NULL;
 158 
 159     crm_debug("Starting new join round join-%d", current_join_id);
 160 
 161     g_hash_table_iter_init(&iter, crm_peer_cache);
 162     while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
 163         crm_update_peer_join(__func__, peer, crm_join_none);
 164     }
 165     if (max_generation_from != NULL) {
 166         free(max_generation_from);
 167         max_generation_from = NULL;
 168     }
 169     if (max_generation_xml != NULL) {
 170         free_xml(max_generation_xml);
 171         max_generation_xml = NULL;
 172     }
 173     controld_clear_fsa_input_flags(R_HAVE_CIB);
 174 }
 175 
 176 /*!
 177  * \internal
 178  * \brief Create a join message from the DC
 179  *
 180  * \param[in] join_op  Join operation name
 181  * \param[in] host_to  Recipient of message
 182  */
 183 static xmlNode *
 184 create_dc_message(const char *join_op, const char *host_to)
     /*  */
 185 {
 186     xmlNode *msg = create_request(join_op, NULL, host_to, CRM_SYSTEM_CRMD,
 187                                   CRM_SYSTEM_DC, NULL);
 188 
 189     /* Identify which election this is a part of */
 190     crm_xml_add_int(msg, PCMK__XA_JOIN_ID, current_join_id);
 191 
 192     /* Add a field specifying whether the DC is shutting down. This keeps the
 193      * joining node from fencing the old DC if it becomes the new DC.
 194      */
 195     pcmk__xe_set_bool_attr(msg, PCMK__XA_DC_LEAVING,
 196                            pcmk_is_set(controld_globals.fsa_input_register,
 197                                        R_SHUTDOWN));
 198     return msg;
 199 }
 200 
 201 static void
 202 join_make_offer(gpointer key, gpointer value, gpointer user_data)
     /*  */
 203 {
 204     xmlNode *offer = NULL;
 205     crm_node_t *member = (crm_node_t *)value;
 206 
 207     CRM_ASSERT(member != NULL);
 208     if (!pcmk__cluster_is_node_active(member)) {
 209         crm_info("Not making join-%d offer to inactive node %s",
 210                  current_join_id,
 211                  (member->uname? member->uname : "with unknown name"));
 212         if(member->expected == NULL && pcmk__str_eq(member->state, CRM_NODE_LOST, pcmk__str_casei)) {
 213             /* You would think this unsafe, but in fact this plus an
 214              * active resource is what causes it to be fenced.
 215              *
 216              * Yes, this does mean that any node that dies at the same
 217              * time as the old DC and is not running resource (still)
 218              * won't be fenced.
 219              *
 220              * I'm not happy about this either.
 221              */
 222             pcmk__update_peer_expected(__func__, member, CRMD_JOINSTATE_DOWN);
 223         }
 224         return;
 225     }
 226 
 227     if (member->uname == NULL) {
 228         crm_info("Not making join-%d offer to node uuid %s with unknown name",
 229                  current_join_id, member->uuid);
 230         return;
 231     }
 232 
 233     if (controld_globals.membership_id != crm_peer_seq) {
 234         controld_globals.membership_id = crm_peer_seq;
 235         crm_info("Making join-%d offers based on membership event %llu",
 236                  current_join_id, crm_peer_seq);
 237     }
 238 
 239     if(user_data && member->join > crm_join_none) {
 240         crm_info("Not making join-%d offer to already known node %s (%s)",
 241                  current_join_id, member->uname,
 242                  crm_join_phase_str(member->join));
 243         return;
 244     }
 245 
 246     crm_update_peer_join(__func__, (crm_node_t*)member, crm_join_none);
 247 
 248     offer = create_dc_message(CRM_OP_JOIN_OFFER, member->uname);
 249 
 250     // Advertise our feature set so the joining node can bail if not compatible
 251     crm_xml_add(offer, PCMK_XA_CRM_FEATURE_SET, CRM_FEATURE_SET);
 252 
 253     crm_info("Sending join-%d offer to %s", current_join_id, member->uname);
 254     pcmk__cluster_send_message(member, crm_msg_crmd, offer);
 255     free_xml(offer);
 256 
 257     crm_update_peer_join(__func__, member, crm_join_welcomed);
 258 }
 259 
 260 /*       A_DC_JOIN_OFFER_ALL    */
 261 void
 262 do_dc_join_offer_all(long long action,
     /*  */
 263                      enum crmd_fsa_cause cause,
 264                      enum crmd_fsa_state cur_state,
 265                      enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 266 {
 267     int count;
 268 
 269     /* Reset everyone's status back to down or in_ccm in the CIB.
 270      * Any nodes that are active in the CIB but not in the cluster membership
 271      * will be seen as offline by the scheduler anyway.
 272      */
 273     current_join_id++;
 274     start_join_round();
 275 
 276     update_dc(NULL);
 277     if (cause == C_HA_MESSAGE && current_input == I_NODE_JOIN) {
 278         crm_info("A new node joined the cluster");
 279     }
 280     g_hash_table_foreach(crm_peer_cache, join_make_offer, NULL);
 281 
 282     count = crmd_join_phase_count(crm_join_welcomed);
 283     crm_info("Waiting on join-%d requests from %d outstanding node%s",
 284              current_join_id, count, pcmk__plural_s(count));
 285 
 286     // Don't waste time by invoking the scheduler yet
 287 }
 288 
 289 /*       A_DC_JOIN_OFFER_ONE    */
 290 void
 291 do_dc_join_offer_one(long long action,
     /*  */
 292                      enum crmd_fsa_cause cause,
 293                      enum crmd_fsa_state cur_state,
 294                      enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 295 {
 296     crm_node_t *member;
 297     ha_msg_input_t *welcome = NULL;
 298     int count;
 299     const char *join_to = NULL;
 300 
 301     if (msg_data->data == NULL) {
 302         crm_info("Making join-%d offers to any unconfirmed nodes "
 303                  "because an unknown node joined", current_join_id);
 304         g_hash_table_foreach(crm_peer_cache, join_make_offer, &member);
 305         check_join_state(cur_state, __func__);
 306         return;
 307     }
 308 
 309     welcome = fsa_typed_data(fsa_dt_ha_msg);
 310     if (welcome == NULL) {
 311         // fsa_typed_data() already logged an error
 312         return;
 313     }
 314 
 315     join_to = crm_element_value(welcome->msg, PCMK__XA_SRC);
 316     if (join_to == NULL) {
 317         crm_err("Can't make join-%d offer to unknown node", current_join_id);
 318         return;
 319     }
 320     member = pcmk__get_node(0, join_to, NULL, pcmk__node_search_cluster_member);
 321 
 322     /* It is possible that a node will have been sick or starting up when the
 323      * original offer was made. However, it will either re-announce itself in
 324      * due course, or we can re-store the original offer on the client.
 325      */
 326 
 327     crm_update_peer_join(__func__, member, crm_join_none);
 328     join_make_offer(NULL, member, NULL);
 329 
 330     /* If the offer isn't to the local node, make an offer to the local node as
 331      * well, to ensure the correct value for max_generation_from.
 332      */
 333     if (strcasecmp(join_to, controld_globals.our_nodename) != 0) {
 334         member = pcmk__get_node(0, controld_globals.our_nodename, NULL,
 335                                 pcmk__node_search_cluster_member);
 336         join_make_offer(NULL, member, NULL);
 337     }
 338 
 339     /* This was a genuine join request; cancel any existing transition and
 340      * invoke the scheduler.
 341      */
 342     abort_transition(PCMK_SCORE_INFINITY, pcmk__graph_restart, "Node join",
 343                      NULL);
 344 
 345     count = crmd_join_phase_count(crm_join_welcomed);
 346     crm_info("Waiting on join-%d requests from %d outstanding node%s",
 347              current_join_id, count, pcmk__plural_s(count));
 348 
 349     // Don't waste time by invoking the scheduler yet
 350 }
 351 
 352 static int
 353 compare_int_fields(xmlNode * left, xmlNode * right, const char *field)
     /*  */
 354 {
 355     const char *elem_l = crm_element_value(left, field);
 356     const char *elem_r = crm_element_value(right, field);
 357 
 358     long long int_elem_l;
 359     long long int_elem_r;
 360 
 361     pcmk__scan_ll(elem_l, &int_elem_l, -1LL);
 362     pcmk__scan_ll(elem_r, &int_elem_r, -1LL);
 363 
 364     if (int_elem_l < int_elem_r) {
 365         return -1;
 366 
 367     } else if (int_elem_l > int_elem_r) {
 368         return 1;
 369     }
 370 
 371     return 0;
 372 }
 373 
 374 /*       A_DC_JOIN_PROCESS_REQ  */
 375 void
 376 do_dc_join_filter_offer(long long action,
     /*  */
 377                         enum crmd_fsa_cause cause,
 378                         enum crmd_fsa_state cur_state,
 379                         enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 380 {
 381     xmlNode *generation = NULL;
 382 
 383     int cmp = 0;
 384     int join_id = -1;
 385     int count = 0;
 386     gint value = 0;
 387     gboolean ack_nack_bool = TRUE;
 388     ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
 389 
 390     const char *join_from = crm_element_value(join_ack->msg, PCMK__XA_SRC);
 391     const char *ref = crm_element_value(join_ack->msg, PCMK_XA_REFERENCE);
 392     const char *join_version = crm_element_value(join_ack->msg,
 393                                                  PCMK_XA_CRM_FEATURE_SET);
 394     crm_node_t *join_node = NULL;
 395 
 396     if (join_from == NULL) {
 397         crm_err("Ignoring invalid join request without node name");
 398         return;
 399     }
 400     join_node = pcmk__get_node(0, join_from, NULL,
 401                                pcmk__node_search_cluster_member);
 402 
 403     crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id);
 404     if (join_id != current_join_id) {
 405         crm_debug("Ignoring join-%d request from %s because we are on join-%d",
 406                   join_id, join_from, current_join_id);
 407         check_join_state(cur_state, __func__);
 408         return;
 409     }
 410 
 411     generation = join_ack->xml;
 412     if (max_generation_xml != NULL && generation != NULL) {
 413         int lpc = 0;
 414 
 415         const char *attributes[] = {
 416             PCMK_XA_ADMIN_EPOCH,
 417             PCMK_XA_EPOCH,
 418             PCMK_XA_NUM_UPDATES,
 419         };
 420 
 421         /* It's not obvious that join_ack->xml is the PCMK__XE_GENERATION_TUPLE
 422          * element from the join client. The "if" guard is for clarity.
 423          */
 424         if (pcmk__xe_is(generation, PCMK__XE_GENERATION_TUPLE)) {
 425             for (lpc = 0; cmp == 0 && lpc < PCMK__NELEM(attributes); lpc++) {
 426                 cmp = compare_int_fields(max_generation_xml, generation,
 427                                          attributes[lpc]);
 428             }
 429 
 430         } else {    // Should always be PCMK__XE_GENERATION_TUPLE
 431             CRM_LOG_ASSERT(false);
 432         }
 433     }
 434 
 435     if (ref == NULL) {
 436         ref = "none"; // for logging only
 437     }
 438 
 439     if (lookup_failed_sync_node(join_from, &value) == pcmk_rc_ok) {
 440         crm_err("Rejecting join-%d request from node %s because we failed to "
 441                 "sync its CIB in join-%d " CRM_XS " ref=%s",
 442                 join_id, join_from, value, ref);
 443         ack_nack_bool = FALSE;
 444 
 445     } else if (!pcmk__cluster_is_node_active(join_node)) {
 446         if (match_down_event(join_from) != NULL) {
 447             /* The join request was received after the node was fenced or
 448              * otherwise shutdown in a way that we're aware of. No need to log
 449              * an error in this rare occurrence; we know the client was recently
 450              * shut down, and receiving a lingering in-flight request is not
 451              * cause for alarm.
 452              */
 453             crm_debug("Rejecting join-%d request from inactive node %s "
 454                       CRM_XS " ref=%s", join_id, join_from, ref);
 455         } else {
 456             crm_err("Rejecting join-%d request from inactive node %s "
 457                     CRM_XS " ref=%s", join_id, join_from, ref);
 458         }
 459         ack_nack_bool = FALSE;
 460 
 461     } else if (generation == NULL) {
 462         crm_err("Rejecting invalid join-%d request from node %s "
 463                 "missing CIB generation " CRM_XS " ref=%s",
 464                 join_id, join_from, ref);
 465         ack_nack_bool = FALSE;
 466 
 467     } else if ((join_version == NULL)
 468                || !feature_set_compatible(CRM_FEATURE_SET, join_version)) {
 469         crm_err("Rejecting join-%d request from node %s because feature set %s"
 470                 " is incompatible with ours (%s) " CRM_XS " ref=%s",
 471                 join_id, join_from, (join_version? join_version : "pre-3.1.0"),
 472                 CRM_FEATURE_SET, ref);
 473         ack_nack_bool = FALSE;
 474 
 475     } else if (max_generation_xml == NULL) {
 476         const char *validation = crm_element_value(generation,
 477                                                    PCMK_XA_VALIDATE_WITH);
 478 
 479         if (pcmk__get_schema(validation) == NULL) {
 480             crm_err("Rejecting join-%d request from %s (with first CIB "
 481                     "generation) due to unknown schema version %s "
 482                     CRM_XS " ref=%s",
 483                     join_id, join_from, pcmk__s(validation, "(missing)"), ref);
 484             ack_nack_bool = FALSE;
 485 
 486         } else {
 487             crm_debug("Accepting join-%d request from %s (with first CIB "
 488                       "generation) " CRM_XS " ref=%s",
 489                       join_id, join_from, ref);
 490             max_generation_xml = pcmk__xml_copy(NULL, generation);
 491             pcmk__str_update(&max_generation_from, join_from);
 492         }
 493 
 494     } else if ((cmp < 0)
 495                || ((cmp == 0)
 496                    && pcmk__str_eq(join_from, controld_globals.our_nodename,
 497                                    pcmk__str_casei))) {
 498         const char *validation = crm_element_value(generation,
 499                                                    PCMK_XA_VALIDATE_WITH);
 500 
 501         if (pcmk__get_schema(validation) == NULL) {
 502             crm_err("Rejecting join-%d request from %s (with better CIB "
 503                     "generation than current best from %s) due to unknown "
 504                     "schema version %s " CRM_XS " ref=%s",
 505                     join_id, join_from, max_generation_from,
 506                     pcmk__s(validation, "(missing)"), ref);
 507             ack_nack_bool = FALSE;
 508 
 509         } else {
 510             crm_debug("Accepting join-%d request from %s (with better CIB "
 511                       "generation than current best from %s) " CRM_XS " ref=%s",
 512                       join_id, join_from, max_generation_from, ref);
 513             crm_log_xml_debug(max_generation_xml, "Old max generation");
 514             crm_log_xml_debug(generation, "New max generation");
 515 
 516             free_xml(max_generation_xml);
 517             max_generation_xml = pcmk__xml_copy(NULL, join_ack->xml);
 518             pcmk__str_update(&max_generation_from, join_from);
 519         }
 520 
 521     } else {
 522         crm_debug("Accepting join-%d request from %s " CRM_XS " ref=%s",
 523                   join_id, join_from, ref);
 524     }
 525 
 526     if (!ack_nack_bool) {
 527         if (compare_version(join_version, "3.17.0") < 0) {
 528             /* Clients with CRM_FEATURE_SET < 3.17.0 may respawn infinitely
 529              * after a nack message, don't send one
 530              */
 531             crm_update_peer_join(__func__, join_node, crm_join_nack_quiet);
 532         } else {
 533             crm_update_peer_join(__func__, join_node, crm_join_nack);
 534         }
 535         pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_NACK);
 536 
 537     } else {
 538         crm_update_peer_join(__func__, join_node, crm_join_integrated);
 539         pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER);
 540     }
 541 
 542     count = crmd_join_phase_count(crm_join_integrated);
 543     crm_debug("%d node%s currently integrated in join-%d",
 544               count, pcmk__plural_s(count), join_id);
 545 
 546     if (check_join_state(cur_state, __func__) == FALSE) {
 547         // Don't waste time by invoking the scheduler yet
 548         count = crmd_join_phase_count(crm_join_welcomed);
 549         crm_debug("Waiting on join-%d requests from %d outstanding node%s",
 550                   join_id, count, pcmk__plural_s(count));
 551     }
 552 }
 553 
 554 /*      A_DC_JOIN_FINALIZE      */
 555 void
 556 do_dc_join_finalize(long long action,
     /*  */
 557                     enum crmd_fsa_cause cause,
 558                     enum crmd_fsa_state cur_state,
 559                     enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 560 {
 561     char *sync_from = NULL;
 562     int rc = pcmk_ok;
 563     int count_welcomed = crmd_join_phase_count(crm_join_welcomed);
 564     int count_finalizable = crmd_join_phase_count(crm_join_integrated)
 565                             + crmd_join_phase_count(crm_join_nack)
 566                             + crmd_join_phase_count(crm_join_nack_quiet);
 567 
 568     /* This we can do straight away and avoid clients timing us out
 569      *  while we compute the latest CIB
 570      */
 571     if (count_welcomed != 0) {
 572         crm_debug("Waiting on join-%d requests from %d outstanding node%s "
 573                   "before finalizing join", current_join_id, count_welcomed,
 574                   pcmk__plural_s(count_welcomed));
 575         crmd_join_phase_log(LOG_DEBUG);
 576         /* crmd_fsa_stall(FALSE); Needed? */
 577         return;
 578 
 579     } else if (count_finalizable == 0) {
 580         crm_debug("Finalization not needed for join-%d at the current time",
 581                   current_join_id);
 582         crmd_join_phase_log(LOG_DEBUG);
 583         check_join_state(controld_globals.fsa_state, __func__);
 584         return;
 585     }
 586 
 587     controld_clear_fsa_input_flags(R_HAVE_CIB);
 588     if (pcmk__str_eq(max_generation_from, controld_globals.our_nodename,
 589                      pcmk__str_null_matches|pcmk__str_casei)) {
 590         controld_set_fsa_input_flags(R_HAVE_CIB);
 591     }
 592 
 593     if (!controld_globals.transition_graph->complete) {
 594         crm_warn("Delaying join-%d finalization while transition in progress",
 595                  current_join_id);
 596         crmd_join_phase_log(LOG_DEBUG);
 597         crmd_fsa_stall(FALSE);
 598         return;
 599     }
 600 
 601     if (pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) {
 602         // Send our CIB out to everyone
 603         sync_from = pcmk__str_copy(controld_globals.our_nodename);
 604         crm_debug("Finalizing join-%d for %d node%s (sync'ing from local CIB)",
 605                   current_join_id, count_finalizable,
 606                   pcmk__plural_s(count_finalizable));
 607         crm_log_xml_debug(max_generation_xml, "Requested CIB version");
 608 
 609     } else {
 610         // Ask for the agreed best CIB
 611         sync_from = pcmk__str_copy(max_generation_from);
 612         crm_notice("Finalizing join-%d for %d node%s (sync'ing CIB from %s)",
 613                    current_join_id, count_finalizable,
 614                    pcmk__plural_s(count_finalizable), sync_from);
 615         crm_log_xml_notice(max_generation_xml, "Requested CIB version");
 616     }
 617     crmd_join_phase_log(LOG_DEBUG);
 618 
 619     rc = controld_globals.cib_conn->cmds->sync_from(controld_globals.cib_conn,
 620                                                     sync_from, NULL, cib_none);
 621     fsa_register_cib_callback(rc, sync_from, finalize_sync_callback);
 622 }
 623 
 624 void
 625 free_max_generation(void)
     /*  */
 626 {
 627     free(max_generation_from);
 628     max_generation_from = NULL;
 629 
 630     free_xml(max_generation_xml);
 631     max_generation_xml = NULL;
 632 }
 633 
 634 void
 635 finalize_sync_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
     /*  */
 636 {
 637     CRM_LOG_ASSERT(-EPERM != rc);
 638 
 639     if (rc != pcmk_ok) {
 640         const char *sync_from = (const char *) user_data;
 641 
 642         do_crm_log(((rc == -pcmk_err_old_data)? LOG_WARNING : LOG_ERR),
 643                    "Could not sync CIB from %s in join-%d: %s",
 644                    sync_from, current_join_id, pcmk_strerror(rc));
 645 
 646         if (rc != -pcmk_err_old_data) {
 647             record_failed_sync_node(sync_from, current_join_id);
 648         }
 649 
 650         /* restart the whole join process */
 651         register_fsa_error_adv(C_FSA_INTERNAL, I_ELECTION_DC, NULL, NULL,
 652                                __func__);
 653 
 654     } else if (!AM_I_DC) {
 655         crm_debug("Sync'ed CIB for join-%d but no longer DC", current_join_id);
 656 
 657     } else if (controld_globals.fsa_state != S_FINALIZE_JOIN) {
 658         crm_debug("Sync'ed CIB for join-%d but no longer in S_FINALIZE_JOIN "
 659                   "(%s)", current_join_id,
 660                   fsa_state2string(controld_globals.fsa_state));
 661 
 662     } else {
 663         controld_set_fsa_input_flags(R_HAVE_CIB);
 664 
 665         /* make sure dc_uuid is re-set to us */
 666         if (!check_join_state(controld_globals.fsa_state, __func__)) {
 667             int count_finalizable = 0;
 668 
 669             count_finalizable = crmd_join_phase_count(crm_join_integrated)
 670                                 + crmd_join_phase_count(crm_join_nack)
 671                                 + crmd_join_phase_count(crm_join_nack_quiet);
 672 
 673             crm_debug("Notifying %d node%s of join-%d results",
 674                       count_finalizable, pcmk__plural_s(count_finalizable),
 675                       current_join_id);
 676             g_hash_table_foreach(crm_peer_cache, finalize_join_for, NULL);
 677         }
 678     }
 679 }
 680 
 681 static void
 682 join_node_state_commit_callback(xmlNode *msg, int call_id, int rc,
     /*  */
 683                                 xmlNode *output, void *user_data)
 684 {
 685     const char *node = user_data;
 686 
 687     if (rc != pcmk_ok) {
 688         fsa_data_t *msg_data = NULL;    // for register_fsa_error() macro
 689 
 690         crm_crit("join-%d node history update (via CIB call %d) for node %s "
 691                  "failed: %s",
 692                  current_join_id, call_id, node, pcmk_strerror(rc));
 693         crm_log_xml_debug(msg, "failed");
 694         register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
 695     }
 696 
 697     crm_debug("join-%d node history update (via CIB call %d) for node %s "
 698               "complete",
 699               current_join_id, call_id, node);
 700     check_join_state(controld_globals.fsa_state, __func__);
 701 }
 702 
 703 /*      A_DC_JOIN_PROCESS_ACK   */
 704 void
 705 do_dc_join_ack(long long action,
     /*  */
 706                enum crmd_fsa_cause cause,
 707                enum crmd_fsa_state cur_state,
 708                enum crmd_fsa_input current_input, fsa_data_t * msg_data)
 709 {
 710     int join_id = -1;
 711     ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg);
 712 
 713     const char *op = crm_element_value(join_ack->msg, PCMK__XA_CRM_TASK);
 714     char *join_from = crm_element_value_copy(join_ack->msg, PCMK__XA_SRC);
 715     crm_node_t *peer = NULL;
 716 
 717     enum controld_section_e section = controld_section_lrm;
 718     char *xpath = NULL;
 719     xmlNode *state = join_ack->xml;
 720     xmlNode *execd_state = NULL;
 721 
 722     cib_t *cib = controld_globals.cib_conn;
 723     int rc = pcmk_ok;
 724 
 725     // Sanity checks
 726     if (join_from == NULL) {
 727         crm_warn("Ignoring message received without node identification");
 728         goto done;
 729     }
 730     if (op == NULL) {
 731         crm_warn("Ignoring message received from %s without task", join_from);
 732         goto done;
 733     }
 734 
 735     if (strcmp(op, CRM_OP_JOIN_CONFIRM)) {
 736         crm_debug("Ignoring '%s' message from %s while waiting for '%s'",
 737                   op, join_from, CRM_OP_JOIN_CONFIRM);
 738         goto done;
 739     }
 740 
 741     if (crm_element_value_int(join_ack->msg, PCMK__XA_JOIN_ID, &join_id) != 0) {
 742         crm_warn("Ignoring join confirmation from %s without valid join ID",
 743                  join_from);
 744         goto done;
 745     }
 746 
 747     peer = pcmk__get_node(0, join_from, NULL, pcmk__node_search_cluster_member);
 748     if (peer->join != crm_join_finalized) {
 749         crm_info("Ignoring out-of-sequence join-%d confirmation from %s "
 750                  "(currently %s not %s)",
 751                  join_id, join_from, crm_join_phase_str(peer->join),
 752                  crm_join_phase_str(crm_join_finalized));
 753         goto done;
 754     }
 755 
 756     if (join_id != current_join_id) {
 757         crm_err("Rejecting join-%d confirmation from %s "
 758                 "because currently on join-%d",
 759                 join_id, join_from, current_join_id);
 760         crm_update_peer_join(__func__, peer, crm_join_nack);
 761         goto done;
 762     }
 763 
 764     crm_update_peer_join(__func__, peer, crm_join_confirmed);
 765 
 766     /* Update CIB with node's current executor state. A new transition will be
 767      * triggered later, when the CIB manager notifies us of the change.
 768      *
 769      * The delete and modify requests are part of an atomic transaction.
 770      */
 771     rc = cib->cmds->init_transaction(cib);
 772     if (rc != pcmk_ok) {
 773         goto done;
 774     }
 775 
 776     // Delete relevant parts of node's current executor state from CIB
 777     if (pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) {
 778         section = controld_section_lrm_unlocked;
 779     }
 780     controld_node_state_deletion_strings(join_from, section, &xpath, NULL);
 781 
 782     rc = cib->cmds->remove(cib, xpath, NULL,
 783                            cib_scope_local
 784                            |cib_xpath
 785                            |cib_multiple
 786                            |cib_transaction);
 787     if (rc != pcmk_ok) {
 788         goto done;
 789     }
 790 
 791     // Update CIB with node's latest known executor state
 792     if (pcmk__str_eq(join_from, controld_globals.our_nodename,
 793                      pcmk__str_casei)) {
 794 
 795         // Use the latest possible state if processing our own join ack
 796         execd_state = controld_query_executor_state();
 797 
 798         if (execd_state != NULL) {
 799             crm_debug("Updating local node history for join-%d from query "
 800                       "result",
 801                       current_join_id);
 802             state = execd_state;
 803 
 804         } else {
 805             crm_warn("Updating local node history from join-%d confirmation "
 806                      "because query failed",
 807                      current_join_id);
 808         }
 809 
 810     } else {
 811         crm_debug("Updating node history for %s from join-%d confirmation",
 812                   join_from, current_join_id);
 813     }
 814 
 815     rc = cib->cmds->modify(cib, PCMK_XE_STATUS, state,
 816                            cib_scope_local|cib_can_create|cib_transaction);
 817     free_xml(execd_state);
 818     if (rc != pcmk_ok) {
 819         goto done;
 820     }
 821 
 822     // Commit the transaction
 823     rc = cib->cmds->end_transaction(cib, true, cib_scope_local);
 824     fsa_register_cib_callback(rc, join_from, join_node_state_commit_callback);
 825 
 826     if (rc > 0) {
 827         // join_from will be freed after callback
 828         join_from = NULL;
 829         rc = pcmk_ok;
 830     }
 831 
 832 done:
 833     if (rc != pcmk_ok) {
 834         crm_crit("join-%d node history update for node %s failed: %s",
 835                  current_join_id, join_from, pcmk_strerror(rc));
 836         register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
 837     }
 838     free(join_from);
 839     free(xpath);
 840 }
 841 
 842 void
 843 finalize_join_for(gpointer key, gpointer value, gpointer user_data)
     /*  */
 844 {
 845     xmlNode *acknak = NULL;
 846     xmlNode *tmp1 = NULL;
 847     crm_node_t *join_node = value;
 848     const char *join_to = join_node->uname;
 849     bool integrated = false;
 850 
 851     switch (join_node->join) {
 852         case crm_join_integrated:
 853             integrated = true;
 854             break;
 855         case crm_join_nack:
 856         case crm_join_nack_quiet:
 857             break;
 858         default:
 859             crm_trace("Not updating non-integrated and non-nacked node %s (%s) "
 860                       "for join-%d", join_to,
 861                       crm_join_phase_str(join_node->join), current_join_id);
 862             return;
 863     }
 864 
 865     /* Update the <node> element with the node's name and UUID, in case they
 866      * weren't known before
 867      */
 868     crm_trace("Updating node name and UUID in CIB for %s", join_to);
 869     tmp1 = pcmk__xe_create(NULL, PCMK_XE_NODE);
 870     crm_xml_add(tmp1, PCMK_XA_ID, pcmk__cluster_node_uuid(join_node));
 871     crm_xml_add(tmp1, PCMK_XA_UNAME, join_to);
 872     fsa_cib_anon_update(PCMK_XE_NODES, tmp1);
 873     free_xml(tmp1);
 874 
 875     if (join_node->join == crm_join_nack_quiet) {
 876         crm_trace("Not sending nack message to node %s with feature set older "
 877                   "than 3.17.0", join_to);
 878         return;
 879     }
 880 
 881     join_node = pcmk__get_node(0, join_to, NULL,
 882                                pcmk__node_search_cluster_member);
 883     if (!pcmk__cluster_is_node_active(join_node)) {
 884         /*
 885          * NACK'ing nodes that the membership layer doesn't know about yet
 886          * simply creates more churn
 887          *
 888          * Better to leave them waiting and let the join restart when
 889          * the new membership event comes in
 890          *
 891          * All other NACKs (due to versions etc) should still be processed
 892          */
 893         pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_PENDING);
 894         return;
 895     }
 896 
 897     // Acknowledge or nack node's join request
 898     crm_debug("%sing join-%d request from %s",
 899               integrated? "Acknowledg" : "Nack", current_join_id, join_to);
 900     acknak = create_dc_message(CRM_OP_JOIN_ACKNAK, join_to);
 901     pcmk__xe_set_bool_attr(acknak, CRM_OP_JOIN_ACKNAK, integrated);
 902 
 903     if (integrated) {
 904         // No change needed for a nacked node
 905         crm_update_peer_join(__func__, join_node, crm_join_finalized);
 906         pcmk__update_peer_expected(__func__, join_node, CRMD_JOINSTATE_MEMBER);
 907 
 908         /* Iterate through the remote peer cache and add information on which
 909          * node hosts each to the ACK message.  This keeps new controllers in
 910          * sync with what has already happened.
 911          */
 912         if (pcmk__cluster_num_remote_nodes() > 0) {
 913             GHashTableIter iter;
 914             crm_node_t *node = NULL;
 915             xmlNode *remotes = pcmk__xe_create(acknak, PCMK_XE_NODES);
 916 
 917             g_hash_table_iter_init(&iter, crm_remote_peer_cache);
 918             while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
 919                 xmlNode *remote = NULL;
 920 
 921                 if (!node->conn_host) {
 922                     continue;
 923                 }
 924 
 925                 remote = pcmk__xe_create(remotes, PCMK_XE_NODE);
 926                 pcmk__xe_set_props(remote,
 927                                    PCMK_XA_ID, node->uname,
 928                                    PCMK__XA_NODE_STATE, node->state,
 929                                    PCMK__XA_CONNECTION_HOST, node->conn_host,
 930                                    NULL);
 931             }
 932         }
 933     }
 934     pcmk__cluster_send_message(join_node, crm_msg_crmd, acknak);
 935     free_xml(acknak);
 936     return;
 937 }
 938 
 939 gboolean
 940 check_join_state(enum crmd_fsa_state cur_state, const char *source)
     /*  */
 941 {
 942     static unsigned long long highest_seq = 0;
 943 
 944     if (controld_globals.membership_id != crm_peer_seq) {
 945         crm_debug("join-%d: Membership changed from %llu to %llu "
 946                   CRM_XS " highest=%llu state=%s for=%s",
 947                   current_join_id, controld_globals.membership_id, crm_peer_seq,
 948                   highest_seq, fsa_state2string(cur_state), source);
 949         if(highest_seq < crm_peer_seq) {
 950             /* Don't spam the FSA with duplicates */
 951             highest_seq = crm_peer_seq;
 952             register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
 953         }
 954 
 955     } else if (cur_state == S_INTEGRATION) {
 956         if (crmd_join_phase_count(crm_join_welcomed) == 0) {
 957             int count = crmd_join_phase_count(crm_join_integrated);
 958 
 959             crm_debug("join-%d: Integration of %d peer%s complete "
 960                       CRM_XS " state=%s for=%s",
 961                       current_join_id, count, pcmk__plural_s(count),
 962                       fsa_state2string(cur_state), source);
 963             register_fsa_input_before(C_FSA_INTERNAL, I_INTEGRATED, NULL);
 964             return TRUE;
 965         }
 966 
 967     } else if (cur_state == S_FINALIZE_JOIN) {
 968         if (!pcmk_is_set(controld_globals.fsa_input_register, R_HAVE_CIB)) {
 969             crm_debug("join-%d: Delaying finalization until we have CIB "
 970                       CRM_XS " state=%s for=%s",
 971                       current_join_id, fsa_state2string(cur_state), source);
 972             return TRUE;
 973 
 974         } else if (crmd_join_phase_count(crm_join_welcomed) != 0) {
 975             int count = crmd_join_phase_count(crm_join_welcomed);
 976 
 977             crm_debug("join-%d: Still waiting on %d welcomed node%s "
 978                       CRM_XS " state=%s for=%s",
 979                       current_join_id, count, pcmk__plural_s(count),
 980                       fsa_state2string(cur_state), source);
 981             crmd_join_phase_log(LOG_DEBUG);
 982 
 983         } else if (crmd_join_phase_count(crm_join_integrated) != 0) {
 984             int count = crmd_join_phase_count(crm_join_integrated);
 985 
 986             crm_debug("join-%d: Still waiting on %d integrated node%s "
 987                       CRM_XS " state=%s for=%s",
 988                       current_join_id, count, pcmk__plural_s(count),
 989                       fsa_state2string(cur_state), source);
 990             crmd_join_phase_log(LOG_DEBUG);
 991 
 992         } else if (crmd_join_phase_count(crm_join_finalized) != 0) {
 993             int count = crmd_join_phase_count(crm_join_finalized);
 994 
 995             crm_debug("join-%d: Still waiting on %d finalized node%s "
 996                       CRM_XS " state=%s for=%s",
 997                       current_join_id, count, pcmk__plural_s(count),
 998                       fsa_state2string(cur_state), source);
 999             crmd_join_phase_log(LOG_DEBUG);
1000 
1001         } else {
1002             crm_debug("join-%d: Complete " CRM_XS " state=%s for=%s",
1003                       current_join_id, fsa_state2string(cur_state), source);
1004             register_fsa_input_later(C_FSA_INTERNAL, I_FINALIZED, NULL);
1005             return TRUE;
1006         }
1007     }
1008 
1009     return FALSE;
1010 }
1011 
1012 void
1013 do_dc_join_final(long long action,
     /*  */
1014                  enum crmd_fsa_cause cause,
1015                  enum crmd_fsa_state cur_state,
1016                  enum crmd_fsa_input current_input, fsa_data_t * msg_data)
1017 {
1018     crm_debug("Ensuring DC, quorum and node attributes are up-to-date");
1019     crm_update_quorum(crm_have_quorum, TRUE);
1020 }
1021 
1022 int crmd_join_phase_count(enum crm_join_phase phase)
     /*  */
1023 {
1024     int count = 0;
1025     crm_node_t *peer;
1026     GHashTableIter iter;
1027 
1028     g_hash_table_iter_init(&iter, crm_peer_cache);
1029     while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
1030         if(peer->join == phase) {
1031             count++;
1032         }
1033     }
1034     return count;
1035 }
1036 
1037 void crmd_join_phase_log(int level)
     /*  */
1038 {
1039     crm_node_t *peer;
1040     GHashTableIter iter;
1041 
1042     g_hash_table_iter_init(&iter, crm_peer_cache);
1043     while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &peer)) {
1044         do_crm_log(level, "join-%d: %s=%s", current_join_id, peer->uname,
1045                    crm_join_phase_str(peer->join));
1046     }
1047 }
/* */
root/daemons/controld/controld_join_dc.c

DEFINITIONS
