root/lib/pacemaker/pcmk_sched_probes.c

/* */

DEFINITIONS

1. add_expected_result
2. pcmk__probe_resource_list
3. probe_then_start
4. guest_resource_will_stop
5. probe_action
6. pcmk__probe_rsc_on_node
7. probe_needed_before_action
8. add_probe_orderings_for_stops
9. add_start_orderings_for_probe
10. add_restart_orderings_for_probe
11. clear_actions_tracking_flag
12. add_start_restart_orderings_for_rsc
13. order_then_probes
14. pcmk__order_probes
15. pcmk__schedule_probes

   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 <glib.h>
  13 
  14 #include <crm/crm.h>
  15 #include <crm/pengine/status.h>
  16 #include <pacemaker-internal.h>
  17 #include "libpacemaker_private.h"
  18 
  19 /*!
  20  * \internal
  21  * \brief Add the expected result to a newly created probe
  22  *
  23  * \param[in,out] probe  Probe action to add expected result to
  24  * \param[in]     rsc    Resource that probe is for
  25  * \param[in]     node   Node that probe will run on
  26  */
  27 static void
  28 add_expected_result(pcmk_action_t *probe, const pcmk_resource_t *rsc,
     /*  */
  29                     const pcmk_node_t *node)
  30 {
  31     // Check whether resource is currently active on node
  32     pcmk_node_t *running = pe_find_node_id(rsc->running_on, node->details->id);
  33 
  34     // The expected result is what we think the resource's current state is
  35     if (running == NULL) {
  36         pe__add_action_expected_result(probe, CRM_EX_NOT_RUNNING);
  37 
  38     } else if (rsc->role == pcmk_role_promoted) {
  39         pe__add_action_expected_result(probe, CRM_EX_PROMOTED);
  40     }
  41 }
  42 
  43 /*!
  44  * \internal
  45  * \brief Create any needed robes on a node for a list of resources
  46  *
  47  * \param[in,out] rscs  List of resources to create probes for
  48  * \param[in,out] node  Node to create probes on
  49  *
  50  * \return true if any probe was created, otherwise false
  51  */
  52 bool
  53 pcmk__probe_resource_list(GList *rscs, pcmk_node_t *node)
     /*  */
  54 {
  55     bool any_created = false;
  56 
  57     for (GList *iter = rscs; iter != NULL; iter = iter->next) {
  58         pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
  59 
  60         if (rsc->cmds->create_probe(rsc, node)) {
  61             any_created = true;
  62         }
  63     }
  64     return any_created;
  65 }
  66 
  67 /*!
  68  * \internal
  69  * \brief Order one resource's start after another's start-up probe
  70  *
  71  * \param[in,out] rsc1  Resource that might get start-up probe
  72  * \param[in]     rsc2  Resource that might be started
  73  */
  74 static void
  75 probe_then_start(pcmk_resource_t *rsc1, pcmk_resource_t *rsc2)
     /*  */
  76 {
  77     if ((rsc1->allocated_to != NULL)
  78         && (g_hash_table_lookup(rsc1->known_on,
  79                                 rsc1->allocated_to->details->id) == NULL)) {
  80 
  81         pcmk__new_ordering(rsc1,
  82                            pcmk__op_key(rsc1->id, PCMK_ACTION_MONITOR, 0),
  83                            NULL,
  84                            rsc2, pcmk__op_key(rsc2->id, PCMK_ACTION_START, 0),
  85                            NULL,
  86                            pcmk__ar_ordered, rsc1->cluster);
  87     }
  88 }
  89 
  90 /*!
  91  * \internal
  92  * \brief Check whether a guest resource will stop
  93  *
  94  * \param[in] node  Guest node to check
  95  *
  96  * \return true if guest resource will likely stop, otherwise false
  97  */
  98 static bool
  99 guest_resource_will_stop(const pcmk_node_t *node)
     /*  */
 100 {
 101     const pcmk_resource_t *guest_rsc = node->details->remote_rsc->container;
 102 
 103     /* Ideally, we'd check whether the guest has a required stop, but that
 104      * information doesn't exist yet, so approximate it ...
 105      */
 106     return node->details->remote_requires_reset
 107            || node->details->unclean
 108            || pcmk_is_set(guest_rsc->flags, pcmk_rsc_failed)
 109            || (guest_rsc->next_role == pcmk_role_stopped)
 110 
 111            // Guest is moving
 112            || ((guest_rsc->role > pcmk_role_stopped)
 113                && (guest_rsc->allocated_to != NULL)
 114                && (pcmk__find_node_in_list(guest_rsc->running_on,
 115                    guest_rsc->allocated_to->details->uname) == NULL));
 116 }
 117 
 118 /*!
 119  * \internal
 120  * \brief Create a probe action for a resource on a node
 121  *
 122  * \param[in,out] rsc   Resource to create probe for
 123  * \param[in,out] node  Node to create probe on
 124  *
 125  * \return Newly created probe action
 126  */
 127 static pcmk_action_t *
 128 probe_action(pcmk_resource_t *rsc, pcmk_node_t *node)
     /*  */
 129 {
 130     pcmk_action_t *probe = NULL;
 131     char *key = pcmk__op_key(rsc->id, PCMK_ACTION_MONITOR, 0);
 132 
 133     crm_debug("Scheduling probe of %s %s on %s",
 134               pcmk_role_text(rsc->role), rsc->id, pcmk__node_name(node));
 135 
 136     probe = custom_action(rsc, key, PCMK_ACTION_MONITOR, node, FALSE,
 137                           rsc->cluster);
 138     pcmk__clear_action_flags(probe, pcmk_action_optional);
 139 
 140     pcmk__order_vs_unfence(rsc, node, probe, pcmk__ar_ordered);
 141     add_expected_result(probe, rsc, node);
 142     return probe;
 143 }
 144 
 145 /*!
 146  * \internal
 147  * \brief Create probes for a resource on a node, if needed
 148  *
 149  * \brief Schedule any probes needed for a resource on a node
 150  *
 151  * \param[in,out] rsc   Resource to create probe for
 152  * \param[in,out] node  Node to create probe on
 153  *
 154  * \return true if any probe was created, otherwise false
 155  */
 156 bool
 157 pcmk__probe_rsc_on_node(pcmk_resource_t *rsc, pcmk_node_t *node)
     /*  */
 158 {
 159     uint32_t flags = pcmk__ar_ordered;
 160     pcmk_action_t *probe = NULL;
 161     pcmk_node_t *allowed = NULL;
 162     pcmk_resource_t *top = uber_parent(rsc);
 163     const char *reason = NULL;
 164 
 165     pcmk__assert((rsc != NULL) && (node != NULL));
 166 
 167     if (!pcmk_is_set(rsc->cluster->flags, pcmk_sched_probe_resources)) {
 168         reason = "start-up probes are disabled";
 169         goto no_probe;
 170     }
 171 
 172     if (pcmk__is_pacemaker_remote_node(node)) {
 173         if (pcmk_is_set(rsc->flags, pcmk_rsc_fence_device)) {
 174             reason = "Pacemaker Remote nodes cannot run stonith agents";
 175             goto no_probe;
 176 
 177         } else if (pcmk__is_guest_or_bundle_node(node)
 178                    && pe__resource_contains_guest_node(rsc->cluster, rsc)) {
 179             reason = "guest nodes cannot run resources containing guest nodes";
 180             goto no_probe;
 181 
 182         } else if (rsc->is_remote_node) {
 183             reason = "Pacemaker Remote nodes cannot host remote connections";
 184             goto no_probe;
 185         }
 186     }
 187 
 188     // If this is a collective resource, probes are created for its children
 189     if (rsc->children != NULL) {
 190         return pcmk__probe_resource_list(rsc->children, node);
 191     }
 192 
 193     if ((rsc->container != NULL) && !rsc->is_remote_node) {
 194         reason = "resource is inside a container";
 195         goto no_probe;
 196 
 197     } else if (pcmk_is_set(rsc->flags, pcmk_rsc_removed)) {
 198         reason = "resource is orphaned";
 199         goto no_probe;
 200 
 201     } else if (g_hash_table_lookup(rsc->known_on, node->details->id) != NULL) {
 202         reason = "resource state is already known";
 203         goto no_probe;
 204     }
 205 
 206     allowed = g_hash_table_lookup(rsc->allowed_nodes, node->details->id);
 207 
 208     if (rsc->exclusive_discover || top->exclusive_discover) {
 209         // Exclusive discovery is enabled ...
 210 
 211         if (allowed == NULL) {
 212             // ... but this node is not allowed to run the resource
 213             reason = "resource has exclusive discovery but is not allowed "
 214                      "on node";
 215             goto no_probe;
 216 
 217         } else if (allowed->rsc_discover_mode != pcmk_probe_exclusive) {
 218             // ... but no constraint marks this node for discovery of resource
 219             reason = "resource has exclusive discovery but is not enabled "
 220                      "on node";
 221             goto no_probe;
 222         }
 223     }
 224 
 225     if (allowed == NULL) {
 226         allowed = node;
 227     }
 228     if (allowed->rsc_discover_mode == pcmk_probe_never) {
 229         reason = "node has discovery disabled";
 230         goto no_probe;
 231     }
 232 
 233     if (pcmk__is_guest_or_bundle_node(node)) {
 234         pcmk_resource_t *guest = node->details->remote_rsc->container;
 235 
 236         if (guest->role == pcmk_role_stopped) {
 237             // The guest is stopped, so we know no resource is active there
 238             reason = "node's guest is stopped";
 239             probe_then_start(guest, top);
 240             goto no_probe;
 241 
 242         } else if (guest_resource_will_stop(node)) {
 243             reason = "node's guest will stop";
 244 
 245             // Order resource start after guest stop (in case it's restarting)
 246             pcmk__new_ordering(guest,
 247                                pcmk__op_key(guest->id, PCMK_ACTION_STOP, 0),
 248                                NULL, top,
 249                                pcmk__op_key(top->id, PCMK_ACTION_START, 0),
 250                                NULL, pcmk__ar_ordered, rsc->cluster);
 251             goto no_probe;
 252         }
 253     }
 254 
 255     // We've eliminated all cases where a probe is not needed, so now it is
 256     probe = probe_action(rsc, node);
 257 
 258     /* Below, we will order the probe relative to start or reload. If this is a
 259      * clone instance, the start or reload is for the entire clone rather than
 260      * just the instance. Otherwise, the start or reload is for the resource
 261      * itself.
 262      */
 263     if (!pcmk__is_clone(top)) {
 264         top = rsc;
 265     }
 266 
 267     /* Prevent a start if the resource can't be probed, but don't cause the
 268      * resource or entire clone to stop if already active.
 269      */
 270     if (!pcmk_is_set(probe->flags, pcmk_action_runnable)
 271         && (top->running_on == NULL)) {
 272         pcmk__set_relation_flags(flags, pcmk__ar_unrunnable_first_blocks);
 273     }
 274 
 275     // Start or reload after probing the resource
 276     pcmk__new_ordering(rsc, NULL, probe,
 277                        top, pcmk__op_key(top->id, PCMK_ACTION_START, 0), NULL,
 278                        flags, rsc->cluster);
 279     pcmk__new_ordering(rsc, NULL, probe, top, reload_key(rsc), NULL,
 280                        pcmk__ar_ordered, rsc->cluster);
 281 
 282     return true;
 283 
 284 no_probe:
 285     pcmk__rsc_trace(rsc,
 286                     "Skipping probe for %s on %s because %s",
 287                     rsc->id, node->details->id, reason);
 288     return false;
 289 }
 290 
 291 /*!
 292  * \internal
 293  * \brief Check whether a probe should be ordered before another action
 294  *
 295  * \param[in] probe  Probe action to check
 296  * \param[in] then   Other action to check
 297  *
 298  * \return true if \p probe should be ordered before \p then, otherwise false
 299  */
 300 static bool
 301 probe_needed_before_action(const pcmk_action_t *probe,
     /*  */
 302                            const pcmk_action_t *then)
 303 {
 304     // Probes on a node are performed after unfencing it, not before
 305     if (pcmk__str_eq(then->task, PCMK_ACTION_STONITH, pcmk__str_none)
 306         && pcmk__same_node(probe->node, then->node)) {
 307         const char *op = g_hash_table_lookup(then->meta,
 308                                              PCMK__META_STONITH_ACTION);
 309 
 310         if (pcmk__str_eq(op, PCMK_ACTION_ON, pcmk__str_casei)) {
 311             return false;
 312         }
 313     }
 314 
 315     // Probes should be done on a node before shutting it down
 316     if (pcmk__str_eq(then->task, PCMK_ACTION_DO_SHUTDOWN, pcmk__str_none)
 317         && (probe->node != NULL) && (then->node != NULL)
 318         && !pcmk__same_node(probe->node, then->node)) {
 319         return false;
 320     }
 321 
 322     // Otherwise probes should always be done before any other action
 323     return true;
 324 }
 325 
 326 /*!
 327  * \internal
 328  * \brief Add implicit "probe then X" orderings for "stop then X" orderings
 329  *
 330  * If the state of a resource is not known yet, a probe will be scheduled,
 331  * expecting a "not running" result. If the probe fails, a stop will not be
 332  * scheduled until the next transition. Thus, if there are ordering constraints
 333  * like "stop this resource then do something else that's not for the same
 334  * resource", add implicit "probe this resource then do something" equivalents
 335  * so the relation is upheld until we know whether a stop is needed.
 336  *
 337  * \param[in,out] scheduler  Scheduler data
 338  */
 339 static void
 340 add_probe_orderings_for_stops(pcmk_scheduler_t *scheduler)
     /*  */
 341 {
 342     for (GList *iter = scheduler->ordering_constraints; iter != NULL;
 343          iter = iter->next) {
 344 
 345         pcmk__action_relation_t *order = iter->data;
 346         uint32_t order_flags = pcmk__ar_ordered;
 347         GList *probes = NULL;
 348         GList *then_actions = NULL;
 349         pcmk_action_t *first = NULL;
 350         pcmk_action_t *then = NULL;
 351 
 352         // Skip disabled orderings
 353         if (order->flags == pcmk__ar_none) {
 354             continue;
 355         }
 356 
 357         // Skip non-resource orderings, and orderings for the same resource
 358         if ((order->rsc1 == NULL) || (order->rsc1 == order->rsc2)) {
 359             continue;
 360         }
 361 
 362         // Skip invalid orderings (shouldn't be possible)
 363         first = order->action1;
 364         then = order->action2;
 365         if (((first == NULL) && (order->task1 == NULL))
 366             || ((then == NULL) && (order->task2 == NULL))) {
 367             continue;
 368         }
 369 
 370         // Skip orderings for first actions other than stop
 371         if ((first != NULL) && !pcmk__str_eq(first->task, PCMK_ACTION_STOP,
 372                                              pcmk__str_none)) {
 373             continue;
 374         } else if ((first == NULL)
 375                    && !pcmk__ends_with(order->task1,
 376                                        "_" PCMK_ACTION_STOP "_0")) {
 377             continue;
 378         }
 379 
 380         /* Do not imply a probe ordering for a resource inside of a stopping
 381          * container. Otherwise, it might introduce a transition loop, since a
 382          * probe could be scheduled after the container starts again.
 383          */
 384         if ((order->rsc2 != NULL) && (order->rsc1->container == order->rsc2)) {
 385 
 386             if ((then != NULL) && pcmk__str_eq(then->task, PCMK_ACTION_STOP,
 387                                                pcmk__str_none)) {
 388                 continue;
 389             } else if ((then == NULL)
 390                        && pcmk__ends_with(order->task2,
 391                                           "_" PCMK_ACTION_STOP "_0")) {
 392                 continue;
 393             }
 394         }
 395 
 396         // Preserve certain order options for future filtering
 397         if (pcmk_is_set(order->flags, pcmk__ar_if_first_unmigratable)) {
 398             pcmk__set_relation_flags(order_flags,
 399                                      pcmk__ar_if_first_unmigratable);
 400         }
 401         if (pcmk_is_set(order->flags, pcmk__ar_if_on_same_node)) {
 402             pcmk__set_relation_flags(order_flags, pcmk__ar_if_on_same_node);
 403         }
 404 
 405         // Preserve certain order types for future filtering
 406         if ((order->flags == pcmk__ar_if_required_on_same_node)
 407             || (order->flags == pcmk__ar_if_on_same_node_or_target)) {
 408             order_flags = order->flags;
 409         }
 410 
 411         // List all scheduled probes for the first resource
 412         probes = pe__resource_actions(order->rsc1, NULL, PCMK_ACTION_MONITOR,
 413                                       FALSE);
 414         if (probes == NULL) { // There aren't any
 415             continue;
 416         }
 417 
 418         // List all relevant "then" actions
 419         if (then != NULL) {
 420             then_actions = g_list_prepend(NULL, then);
 421 
 422         } else if (order->rsc2 != NULL) {
 423             then_actions = find_actions(order->rsc2->actions, order->task2,
 424                                         NULL);
 425             if (then_actions == NULL) { // There aren't any
 426                 g_list_free(probes);
 427                 continue;
 428             }
 429         }
 430 
 431         crm_trace("Implying 'probe then' orderings for '%s then %s' "
 432                   "(id=%d, type=%.6x)",
 433                   ((first == NULL)? order->task1 : first->uuid),
 434                   ((then == NULL)? order->task2 : then->uuid),
 435                   order->id, order->flags);
 436 
 437         for (GList *probe_iter = probes; probe_iter != NULL;
 438              probe_iter = probe_iter->next) {
 439 
 440             pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data;
 441 
 442             for (GList *then_iter = then_actions; then_iter != NULL;
 443                  then_iter = then_iter->next) {
 444 
 445                 pcmk_action_t *then = (pcmk_action_t *) then_iter->data;
 446 
 447                 if (probe_needed_before_action(probe, then)) {
 448                     order_actions(probe, then, order_flags);
 449                 }
 450             }
 451         }
 452 
 453         g_list_free(then_actions);
 454         g_list_free(probes);
 455     }
 456 }
 457 
 458 /*!
 459  * \internal
 460  * \brief Add necessary orderings between probe and starts of clone instances
 461  *
 462  * , in additon to the ordering with the parent resource added upon creating
 463  * the probe.
 464  *
 465  * \param[in,out] probe     Probe as 'first' action in an ordering
 466  * \param[in,out] after     'then' action wrapper in the ordering
 467  */
 468 static void
 469 add_start_orderings_for_probe(pcmk_action_t *probe,
     /*  */
 470                               pcmk__related_action_t *after)
 471 {
 472     uint32_t flags = pcmk__ar_ordered|pcmk__ar_unrunnable_first_blocks;
 473 
 474     /* Although the ordering between the probe of the clone instance and the
 475      * start of its parent has been added in pcmk__probe_rsc_on_node(), we
 476      * avoided enforcing `pcmk__ar_unrunnable_first_blocks` order type for that
 477      * as long as any of the clone instances are running to prevent them from
 478      * being unexpectedly stopped.
 479      *
 480      * On the other hand, we still need to prevent any inactive instances from
 481      * starting unless the probe is runnable so that we don't risk starting too
 482      * many instances before we know the state on all nodes.
 483      */
 484     if ((after->action->rsc->variant <= pcmk_rsc_variant_group)
 485         || pcmk_is_set(probe->flags, pcmk_action_runnable)
 486         // The order type is already enforced for its parent.
 487         || pcmk_is_set(after->type, pcmk__ar_unrunnable_first_blocks)
 488         || (pe__const_top_resource(probe->rsc, false) != after->action->rsc)
 489         || !pcmk__str_eq(after->action->task, PCMK_ACTION_START,
 490                          pcmk__str_none)) {
 491         return;
 492     }
 493 
 494     crm_trace("Adding probe start orderings for 'unrunnable %s@%s "
 495               "then instances of %s@%s'",
 496               probe->uuid, pcmk__node_name(probe->node),
 497               after->action->uuid, pcmk__node_name(after->action->node));
 498 
 499     for (GList *then_iter = after->action->actions_after; then_iter != NULL;
 500          then_iter = then_iter->next) {
 501 
 502         pcmk__related_action_t *then = then_iter->data;
 503 
 504         if (then->action->rsc->running_on
 505             || (pe__const_top_resource(then->action->rsc, false)
 506                 != after->action->rsc)
 507             || !pcmk__str_eq(then->action->task, PCMK_ACTION_START,
 508                              pcmk__str_none)) {
 509             continue;
 510         }
 511 
 512         crm_trace("Adding probe start ordering for 'unrunnable %s@%s "
 513                   "then %s@%s' (type=%#.6x)",
 514                   probe->uuid, pcmk__node_name(probe->node),
 515                   then->action->uuid, pcmk__node_name(then->action->node),
 516                   flags);
 517 
 518         /* Prevent the instance from starting if the instance can't, but don't
 519          * cause any other intances to stop if already active.
 520          */
 521         order_actions(probe, then->action, flags);
 522     }
 523 
 524     return;
 525 }
 526 
 527 /*!
 528  * \internal
 529  * \brief Order probes before restarts and re-promotes
 530  *
 531  * If a given ordering is a "probe then start" or "probe then promote" ordering,
 532  * add an implicit "probe then stop/demote" ordering in case the action is part
 533  * of a restart/re-promote, and do the same recursively for all actions ordered
 534  * after the "then" action.
 535  *
 536  * \param[in,out] probe     Probe as 'first' action in an ordering
 537  * \param[in,out] after     'then' action in the ordering
 538  */
 539 static void
 540 add_restart_orderings_for_probe(pcmk_action_t *probe, pcmk_action_t *after)
     /*  */
 541 {
 542     GList *iter = NULL;
 543     bool interleave = false;
 544     pcmk_resource_t *compatible_rsc = NULL;
 545 
 546     // Validate that this is a resource probe followed by some action
 547     if ((after == NULL) || (probe == NULL) || !pcmk__is_primitive(probe->rsc)
 548         || !pcmk__str_eq(probe->task, PCMK_ACTION_MONITOR, pcmk__str_none)) {
 549         return;
 550     }
 551 
 552     // Avoid running into any possible loop
 553     if (pcmk_is_set(after->flags, pcmk_action_detect_loop)) {
 554         return;
 555     }
 556     pcmk__set_action_flags(after, pcmk_action_detect_loop);
 557 
 558     crm_trace("Adding probe restart orderings for '%s@%s then %s@%s'",
 559               probe->uuid, pcmk__node_name(probe->node),
 560               after->uuid, pcmk__node_name(after->node));
 561 
 562     /* Add restart orderings if "then" is for a different primitive.
 563      * Orderings for collective resources will be added later.
 564      */
 565     if (pcmk__is_primitive(after->rsc) && (probe->rsc != after->rsc)) {
 566 
 567             GList *then_actions = NULL;
 568 
 569             if (pcmk__str_eq(after->task, PCMK_ACTION_START, pcmk__str_none)) {
 570                 then_actions = pe__resource_actions(after->rsc, NULL,
 571                                                     PCMK_ACTION_STOP, FALSE);
 572 
 573             } else if (pcmk__str_eq(after->task, PCMK_ACTION_PROMOTE,
 574                                     pcmk__str_none)) {
 575                 then_actions = pe__resource_actions(after->rsc, NULL,
 576                                                     PCMK_ACTION_DEMOTE, FALSE);
 577             }
 578 
 579             for (iter = then_actions; iter != NULL; iter = iter->next) {
 580                 pcmk_action_t *then = (pcmk_action_t *) iter->data;
 581 
 582                 // Skip pseudo-actions (for example, those implied by fencing)
 583                 if (!pcmk_is_set(then->flags, pcmk_action_pseudo)) {
 584                     order_actions(probe, then, pcmk__ar_ordered);
 585                 }
 586             }
 587             g_list_free(then_actions);
 588     }
 589 
 590     /* Detect whether "then" is an interleaved clone action. For these, we want
 591      * to add orderings only for the relevant instance.
 592      */
 593     if ((after->rsc != NULL)
 594         && (after->rsc->variant > pcmk_rsc_variant_group)) {
 595         const char *interleave_s = g_hash_table_lookup(after->rsc->meta,
 596                                                        PCMK_META_INTERLEAVE);
 597 
 598         interleave = crm_is_true(interleave_s);
 599         if (interleave) {
 600             compatible_rsc = pcmk__find_compatible_instance(probe->rsc,
 601                                                             after->rsc,
 602                                                             pcmk_role_unknown,
 603                                                             false);
 604         }
 605     }
 606 
 607     /* Now recursively do the same for all actions ordered after "then". This
 608      * also handles collective resources since the collective action will be
 609      * ordered before its individual instances' actions.
 610      */
 611     for (iter = after->actions_after; iter != NULL; iter = iter->next) {
 612         pcmk__related_action_t *after_wrapper = iter->data;
 613 
 614         /* pcmk__ar_first_implies_then is the reason why a required A.start
 615          * implies/enforces B.start to be required too, which is the cause of
 616          * B.restart/re-promote.
 617          *
 618          * Not sure about pcmk__ar_first_implies_same_node_then though. It's now
 619          * only used for unfencing case, which tends to introduce transition
 620          * loops...
 621          */
 622         if (!pcmk_is_set(after_wrapper->type, pcmk__ar_first_implies_then)) {
 623             /* The order type between a group/clone and its child such as
 624              * B.start-> B_child.start is:
 625              * pcmk__ar_then_implies_first_graphed
 626              * |pcmk__ar_unrunnable_first_blocks
 627              *
 628              * Proceed through the ordering chain and build dependencies with
 629              * its children.
 630              */
 631             if ((after->rsc == NULL)
 632                 || (after->rsc->variant < pcmk_rsc_variant_group)
 633                 || (probe->rsc->parent == after->rsc)
 634                 || (after_wrapper->action->rsc == NULL)
 635                 || (after_wrapper->action->rsc->variant > pcmk_rsc_variant_group)
 636                 || (after->rsc != after_wrapper->action->rsc->parent)) {
 637                 continue;
 638             }
 639 
 640             /* Proceed to the children of a group or a non-interleaved clone.
 641              * For an interleaved clone, proceed only to the relevant child.
 642              */
 643             if ((after->rsc->variant > pcmk_rsc_variant_group) && interleave
 644                 && ((compatible_rsc == NULL)
 645                     || (compatible_rsc != after_wrapper->action->rsc))) {
 646                 continue;
 647             }
 648         }
 649 
 650         crm_trace("Recursively adding probe restart orderings for "
 651                   "'%s@%s then %s@%s' (type=%#.6x)",
 652                   after->uuid, pcmk__node_name(after->node),
 653                   after_wrapper->action->uuid,
 654                   pcmk__node_name(after_wrapper->action->node),
 655                   after_wrapper->type);
 656 
 657         add_restart_orderings_for_probe(probe, after_wrapper->action);
 658     }
 659 }
 660 
 661 /*!
 662  * \internal
 663  * \brief Clear the tracking flag on all scheduled actions
 664  *
 665  * \param[in,out] scheduler  Scheduler data
 666  */
 667 static void
 668 clear_actions_tracking_flag(pcmk_scheduler_t *scheduler)
     /*  */
 669 {
 670     for (GList *iter = scheduler->actions; iter != NULL; iter = iter->next) {
 671         pcmk_action_t *action = iter->data;
 672 
 673         pcmk__clear_action_flags(action, pcmk_action_detect_loop);
 674     }
 675 }
 676 
 677 /*!
 678  * \internal
 679  * \brief Add start and restart orderings for probes scheduled for a resource
 680  *
 681  * \param[in,out] data       Resource whose probes should be ordered
 682  * \param[in]     user_data  Unused
 683  */
 684 static void
 685 add_start_restart_orderings_for_rsc(gpointer data, gpointer user_data)
     /*  */
 686 {
 687     pcmk_resource_t *rsc = data;
 688     GList *probes = NULL;
 689 
 690     // For collective resources, order each instance recursively
 691     if (!pcmk__is_primitive(rsc)) {
 692         g_list_foreach(rsc->children, add_start_restart_orderings_for_rsc,
 693                        NULL);
 694         return;
 695     }
 696 
 697     // Find all probes for given resource
 698     probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE);
 699 
 700     // Add probe restart orderings for each probe found
 701     for (GList *iter = probes; iter != NULL; iter = iter->next) {
 702         pcmk_action_t *probe = (pcmk_action_t *) iter->data;
 703 
 704         for (GList *then_iter = probe->actions_after; then_iter != NULL;
 705              then_iter = then_iter->next) {
 706 
 707             pcmk__related_action_t *then = then_iter->data;
 708 
 709             add_start_orderings_for_probe(probe, then);
 710             add_restart_orderings_for_probe(probe, then->action);
 711             clear_actions_tracking_flag(rsc->cluster);
 712         }
 713     }
 714 
 715     g_list_free(probes);
 716 }
 717 
 718 /*!
 719  * \internal
 720  * \brief Add "A then probe B" orderings for "A then B" orderings
 721  *
 722  * \param[in,out] scheduler  Scheduler data
 723  *
 724  * \note This function is currently disabled (see next comment).
 725  */
 726 static void
 727 order_then_probes(pcmk_scheduler_t *scheduler)
     /*  */
 728 {
 729 #if 0
 730     /* Given an ordering "A then B", we would prefer to wait for A to be started
 731      * before probing B.
 732      *
 733      * For example, if A is a filesystem which B can't even run without, it
 734      * would be helpful if the author of B's agent could assume that A is
 735      * running before B.monitor will be called.
 736      *
 737      * However, we can't _only_ probe after A is running, otherwise we wouldn't
 738      * detect the state of B if A could not be started. We can't even do an
 739      * opportunistic version of this, because B may be moving:
 740      *
 741      *   A.stop -> A.start -> B.probe -> B.stop -> B.start
 742      *
 743      * and if we add B.stop -> A.stop here, we get a loop:
 744      *
 745      *   A.stop -> A.start -> B.probe -> B.stop -> A.stop
 746      *
 747      * We could kill the "B.probe -> B.stop" dependency, but that could mean
 748      * stopping B "too" soon, because B.start must wait for the probe, and
 749      * we don't want to stop B if we can't start it.
 750      *
 751      * We could add the ordering only if A is an anonymous clone with
 752      * clone-max == node-max (since we'll never be moving it). However, we could
 753      * still be stopping one instance at the same time as starting another.
 754      *
 755      * The complexity of checking for allowed conditions combined with the ever
 756      * narrowing use case suggests that this code should remain disabled until
 757      * someone gets smarter.
 758      */
 759     for (GList *iter = scheduler->resources; iter != NULL; iter = iter->next) {
 760         pcmk_resource_t *rsc = (pcmk_resource_t *) iter->data;
 761 
 762         pcmk_action_t *start = NULL;
 763         GList *actions = NULL;
 764         GList *probes = NULL;
 765 
 766         actions = pe__resource_actions(rsc, NULL, PCMK_ACTION_START, FALSE);
 767 
 768         if (actions) {
 769             start = actions->data;
 770             g_list_free(actions);
 771         }
 772 
 773         if (start == NULL) {
 774             crm_debug("No start action for %s", rsc->id);
 775             continue;
 776         }
 777 
 778         probes = pe__resource_actions(rsc, NULL, PCMK_ACTION_MONITOR, FALSE);
 779 
 780         for (actions = start->actions_before; actions != NULL;
 781              actions = actions->next) {
 782 
 783             pcmk__related_action_t *before = actions->data;
 784 
 785             pcmk_action_t *first = before->action;
 786             pcmk_resource_t *first_rsc = first->rsc;
 787 
 788             if (first->required_runnable_before) {
 789                 for (GList *clone_actions = first->actions_before;
 790                      clone_actions != NULL;
 791                      clone_actions = clone_actions->next) {
 792 
 793                     before = clone_actions->data;
 794 
 795                     crm_trace("Testing '%s then %s' for %s",
 796                               first->uuid, before->action->uuid, start->uuid);
 797 
 798                     pcmk__assert(before->action->rsc != NULL);
 799                     first_rsc = before->action->rsc;
 800                     break;
 801                 }
 802 
 803             } else if (!pcmk__str_eq(first->task, PCMK_ACTION_START,
 804                                      pcmk__str_none)) {
 805                 crm_trace("Not a start op %s for %s", first->uuid, start->uuid);
 806             }
 807 
 808             if (first_rsc == NULL) {
 809                 continue;
 810 
 811             } else if (pe__const_top_resource(first_rsc, false)
 812                        == pe__const_top_resource(start->rsc, false)) {
 813                 crm_trace("Same parent %s for %s", first_rsc->id, start->uuid);
 814                 continue;
 815 
 816             } else if (!pcmk__is_clone(pe__const_top_resource(first_rsc,
 817                                                               false))) {
 818                 crm_trace("Not a clone %s for %s", first_rsc->id, start->uuid);
 819                 continue;
 820             }
 821 
 822             crm_debug("Applying %s before %s %d", first->uuid, start->uuid,
 823                       pe__const_top_resource(first_rsc, false)->variant);
 824 
 825             for (GList *probe_iter = probes; probe_iter != NULL;
 826                  probe_iter = probe_iter->next) {
 827 
 828                 pcmk_action_t *probe = (pcmk_action_t *) probe_iter->data;
 829 
 830                 crm_debug("Ordering %s before %s", first->uuid, probe->uuid);
 831                 order_actions(first, probe, pcmk__ar_ordered);
 832             }
 833         }
 834     }
 835 #endif
 836 }
 837 
 838 void
 839 pcmk__order_probes(pcmk_scheduler_t *scheduler)
     /*  */
 840 {
 841     // Add orderings for "probe then X"
 842     g_list_foreach(scheduler->resources, add_start_restart_orderings_for_rsc,
 843                    NULL);
 844     add_probe_orderings_for_stops(scheduler);
 845 
 846     order_then_probes(scheduler);
 847 }
 848 
 849 /*!
 850  * \internal
 851  * \brief Schedule any probes needed
 852  *
 853  * \param[in,out] scheduler  Scheduler data
 854  *
 855  * \note This may also schedule fencing of failed remote nodes.
 856  */
 857 void
 858 pcmk__schedule_probes(pcmk_scheduler_t *scheduler)
     /*  */
 859 {
 860     // Schedule probes on each node in the cluster as needed
 861     for (GList *iter = scheduler->nodes; iter != NULL; iter = iter->next) {
 862         pcmk_node_t *node = (pcmk_node_t *) iter->data;
 863         const char *probed = NULL;
 864 
 865         if (!node->details->online) { // Don't probe offline nodes
 866             if (pcmk__is_failed_remote_node(node)) {
 867                 pe_fence_node(scheduler, node,
 868                               "the connection is unrecoverable", FALSE);
 869             }
 870             continue;
 871 
 872         } else if (node->details->unclean) { // ... or nodes that need fencing
 873             continue;
 874 
 875         } else if (!node->details->rsc_discovery_enabled) {
 876             // The user requested that probes not be done on this node
 877             continue;
 878         }
 879 
 880         /* This is no longer needed for live clusters, since the probe_complete
 881          * node attribute will never be in the CIB. However this is still useful
 882          * for processing old saved CIBs (< 1.1.14), including the
 883          * reprobe-target_rc regression test.
 884          */
 885         probed = pcmk__node_attr(node, CRM_OP_PROBED, NULL,
 886                                  pcmk__rsc_node_current);
 887         if (probed != NULL && crm_is_true(probed) == FALSE) {
 888             pcmk_action_t *probe_op = NULL;
 889 
 890             probe_op = custom_action(NULL,
 891                                      crm_strdup_printf("%s-%s", CRM_OP_REPROBE,
 892                                                        node->details->uname),
 893                                      CRM_OP_REPROBE, node, FALSE, scheduler);
 894             pcmk__insert_meta(probe_op, PCMK__META_OP_NO_WAIT, PCMK_VALUE_TRUE);
 895             continue;
 896         }
 897 
 898         // Probe each resource in the cluster on this node, as needed
 899         pcmk__probe_resource_list(scheduler->resources, node);
 900     }
 901 }

/* */