pcmk_graph_consumer.c

Go to the documentation of this file.

/*
 * Copyright 2004-2021 the Pacemaker project contributors
 *
 * The version control history for this file may have further details.
 *
 * This source code is licensed under the GNU Lesser General Public License
 * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
 */

#include <crm_internal.h>

#include <sys/param.h>
#include <sys/stat.h>

#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-internal.h>


/*
 * Functions for updating graph
 */

static void
update_synapse_ready(synapse_t *synapse, int action_id)
{
    if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) {
        return; // All inputs have already been confirmed
    }
    pcmk__set_synapse_flags(synapse, pcmk__synapse_ready); // Presume ready until proven otherwise
    for (GList *lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) {
        crm_action_t *prereq = (crm_action_t *) lpc->data;

        if (prereq->id == action_id) {
            crm_trace("Confirming input %d of synapse %d",
                      action_id, synapse->id);
            crm__set_graph_action_flags(prereq, pcmk__graph_action_confirmed);

        } else if (!(pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed))) {
            pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready);
            crm_trace("Synapse %d still not ready after action %d",
                      synapse->id, action_id);
        }
    }
    if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) {
        crm_trace("Synapse %d is now ready to execute", synapse->id);
    }
}

static void
update_synapse_confirmed(synapse_t *synapse, int action_id)
{
    bool all_confirmed = true;

    for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) {
        crm_action_t *action = (crm_action_t *) lpc->data;

        if (action->id == action_id) {
            crm_trace("Confirmed action %d of synapse %d",
                      action_id, synapse->id);
            crm__set_graph_action_flags(action, pcmk__graph_action_confirmed);

        } else if (all_confirmed && !(pcmk_is_set(action->flags, pcmk__graph_action_confirmed))) {
            all_confirmed = false;
            crm_trace("Synapse %d still not confirmed after action %d",
                      synapse->id, action_id);
        }
    }

    if (all_confirmed && !(pcmk_is_set(synapse->flags, pcmk__synapse_confirmed))) {
        crm_trace("Confirmed synapse %d", synapse->id);
        pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed);
    }
}

void
pcmk__update_graph(crm_graph_t *graph, crm_action_t *action)
{
    for (GList *lpc = graph->synapses; lpc != NULL; lpc = lpc->next) {
        synapse_t *synapse = (synapse_t *) lpc->data;

        if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_failed)) {
            continue; // This synapse already completed

        } else if (pcmk_is_set(synapse->flags, pcmk__synapse_executed)) {
            update_synapse_confirmed(synapse, action->id);

        } else if (!(pcmk_is_set(action->flags, pcmk__graph_action_failed)) || (synapse->priority == INFINITY)) {
            update_synapse_ready(synapse, action->id);
        }
    }
}


/*
 * Functions for executing graph
 */

/* A transition graph consists of various types of actions. The library caller
 * registers execution functions for each action type, which will be stored
 * here.
 */
static crm_graph_functions_t *graph_fns = NULL;

void
pcmk__set_graph_functions(crm_graph_functions_t *fns)
{
    crm_debug("Setting custom functions for executing transition graphs");
    graph_fns = fns;

    CRM_ASSERT(graph_fns != NULL);
    CRM_ASSERT(graph_fns->rsc != NULL);
    CRM_ASSERT(graph_fns->crmd != NULL);
    CRM_ASSERT(graph_fns->pseudo != NULL);
    CRM_ASSERT(graph_fns->stonith != NULL);
}

static bool
should_fire_synapse(crm_graph_t *graph, synapse_t *synapse)
{
    GList *lpc = NULL;

    pcmk__set_synapse_flags(synapse, pcmk__synapse_ready);
    for (lpc = synapse->inputs; lpc != NULL; lpc = lpc->next) {
        crm_action_t *prereq = (crm_action_t *) lpc->data;

        if (!(pcmk_is_set(prereq->flags, pcmk__graph_action_confirmed))) {
            crm_trace("Input %d for synapse %d not yet confirmed",
                      prereq->id, synapse->id);
            pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready);
            break;

        } else if (pcmk_is_set(prereq->flags, pcmk__graph_action_failed) && !(pcmk_is_set(prereq->flags, pcmk__graph_action_can_fail))) {
            crm_trace("Input %d for synapse %d confirmed but failed",
                      prereq->id, synapse->id);
            pcmk__clear_synapse_flags(synapse, pcmk__synapse_ready);
            break;
        }
    }
    if (pcmk_is_set(synapse->flags, pcmk__synapse_ready)) {
        crm_trace("Synapse %d is ready to execute", synapse->id);
    } else {
        return false;
    }

    for (lpc = synapse->actions; lpc != NULL; lpc = lpc->next) {
        crm_action_t *a = (crm_action_t *) lpc->data;

        if (a->type == action_type_pseudo) {
            /* None of the below applies to pseudo ops */

        } else if (synapse->priority < graph->abort_priority) {
            crm_trace("Skipping synapse %d: priority %d is less than "
                      "abort priority %d",
                      synapse->id, synapse->priority, graph->abort_priority);
            graph->skipped++;
            return false;

        } else if (graph_fns->allowed && !(graph_fns->allowed(graph, a))) {
            crm_trace("Deferring synapse %d: not allowed", synapse->id);
            return false;
        }
    }

    return true;
}

static int
initiate_action(crm_graph_t *graph, crm_action_t *action)
{
    const char *id = ID(action->xml);

    CRM_CHECK(id != NULL, return EINVAL);
    CRM_CHECK(!pcmk_is_set(action->flags, pcmk__graph_action_executed),
              return pcmk_rc_already);

    crm__set_graph_action_flags(action, pcmk__graph_action_executed);
    switch (action->type) {
        case action_type_pseudo:
            crm_trace("Executing pseudo-action %d (%s)", action->id, id);
            return graph_fns->pseudo(graph, action)? pcmk_rc_ok : pcmk_rc_error;

        case action_type_rsc:
            crm_trace("Executing resource action %d (%s)", action->id, id);
            return graph_fns->rsc(graph, action)? pcmk_rc_ok : pcmk_rc_error;

        case action_type_crm:
            if (pcmk__str_eq(crm_element_value(action->xml, XML_LRM_ATTR_TASK),
                             CRM_OP_FENCE, pcmk__str_casei)) {
                crm_trace("Executing fencing action %d (%s)",
                          action->id, id);
                return graph_fns->stonith(graph, action)? pcmk_rc_ok : pcmk_rc_error;
            }
            crm_trace("Executing control action %d (%s)", action->id, id);
            return graph_fns->crmd(graph, action)? pcmk_rc_ok : pcmk_rc_error;

        default:
            crm_err("Unsupported graph action type <%s id='%s'> (bug?)",
                    crm_element_name(action->xml), id);
            return EINVAL;
    }
}

static int
fire_synapse(crm_graph_t *graph, synapse_t *synapse)
{
    pcmk__set_synapse_flags(synapse, pcmk__synapse_executed);
    for (GList *lpc = synapse->actions; lpc != NULL; lpc = lpc->next) {
        crm_action_t *action = (crm_action_t *) lpc->data;

        if (initiate_action(graph, action) != pcmk_rc_ok) {
            crm_err("Failed initiating <%s id=%d> in synapse %d",
                    crm_element_name(action->xml), action->id, synapse->id);
            pcmk__set_synapse_flags(synapse, pcmk__synapse_confirmed);
            crm__set_graph_action_flags(action, pcmk__graph_action_confirmed | pcmk__graph_action_failed);
            return pcmk_rc_error;
        }
    }
    return pcmk_rc_ok;
}

static gboolean
pseudo_action_dummy(crm_graph_t * graph, crm_action_t * action)
{
    static int fail = -1;

    if (fail < 0) {
        long long fail_ll;

        if ((pcmk__scan_ll(getenv("PE_fail"), &fail_ll, 0LL) == pcmk_rc_ok)
            && (fail_ll > 0LL) && (fail_ll <= INT_MAX)) {
            fail = (int) fail_ll;
        } else {
            fail = 0;
        }
    }

    if (action->id == fail) {
        crm_err("Dummy event handler: pretending action %d failed", action->id);
        crm__set_graph_action_flags(action, pcmk__graph_action_failed);
        graph->abort_priority = INFINITY;
    } else {
        crm_trace("Dummy event handler: action %d initiated", action->id);
    }
    crm__set_graph_action_flags(action, pcmk__graph_action_confirmed);
    pcmk__update_graph(graph, action);
    return TRUE;
}

static crm_graph_functions_t default_fns = {
    pseudo_action_dummy,
    pseudo_action_dummy,
    pseudo_action_dummy,
    pseudo_action_dummy
};

enum transition_status
pcmk__execute_graph(crm_graph_t *graph)
{
    GList *lpc = NULL;
    int log_level = LOG_DEBUG;
    enum transition_status pass_result = transition_active;
    const char *status = "In progress";

    if (graph_fns == NULL) {
        graph_fns = &default_fns;
    }
    if (graph == NULL) {
        return transition_complete;
    }

    graph->fired = 0;
    graph->pending = 0;
    graph->skipped = 0;
    graph->completed = 0;
    graph->incomplete = 0;

    // Count completed and in-flight synapses
    for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) {
        synapse_t *synapse = (synapse_t *) lpc->data;

        if (pcmk_is_set(synapse->flags, pcmk__synapse_confirmed)) {
            graph->completed++;

        } else if (!(pcmk_is_set(synapse->flags, pcmk__synapse_failed)) && pcmk_is_set(synapse->flags, pcmk__synapse_executed)) {
            graph->pending++;
        }
    }
    crm_trace("Executing graph %d (%d synapses already completed, %d pending)",
              graph->id, graph->completed, graph->pending);

    // Execute any synapses that are ready
    for (lpc = graph->synapses; lpc != NULL; lpc = lpc->next) {
        synapse_t *synapse = (synapse_t *) lpc->data;

        if ((graph->batch_limit > 0)
            && (graph->pending >= graph->batch_limit)) {

            crm_debug("Throttling graph execution: batch limit (%d) reached",
                      graph->batch_limit);
            break;

        } else if (pcmk_is_set(synapse->flags, pcmk__synapse_failed)) {
            graph->skipped++;
            continue;

        } else if (pcmk_any_flags_set(synapse->flags, pcmk__synapse_confirmed|pcmk__synapse_executed)) {
            continue; // Already handled

        } else if (should_fire_synapse(graph, synapse)) {
            graph->fired++;
            if (fire_synapse(graph, synapse) != pcmk_rc_ok) {
                crm_err("Synapse %d failed to fire", synapse->id);
                log_level = LOG_ERR;
                graph->abort_priority = INFINITY;
                graph->incomplete++;
                graph->fired--;
            }

            if (!(pcmk_is_set(synapse->flags, pcmk__synapse_confirmed))) {
                graph->pending++;
            }

        } else {
            crm_trace("Synapse %d cannot fire", synapse->id);
            graph->incomplete++;
        }
    }

    if ((graph->pending == 0) && (graph->fired == 0)) {
        graph->complete = TRUE;

        if ((graph->incomplete != 0) && (graph->abort_priority <= 0)) {
            log_level = LOG_WARNING;
            pass_result = transition_terminated;
            status = "Terminated";

        } else if (graph->skipped != 0) {
            log_level = LOG_NOTICE;
            pass_result = transition_complete;
            status = "Stopped";

        } else {
            log_level = LOG_NOTICE;
            pass_result = transition_complete;
            status = "Complete";
        }

    } else if (graph->fired == 0) {
        pass_result = transition_pending;
    }

    do_crm_log(log_level,
               "Transition %d (Complete=%d, Pending=%d,"
               " Fired=%d, Skipped=%d, Incomplete=%d, Source=%s): %s",
               graph->id, graph->completed, graph->pending, graph->fired,
               graph->skipped, graph->incomplete, graph->source, status);

    return pass_result;
}


/*
 * Functions for unpacking transition graph XML into structs
 */

static crm_action_t *
unpack_action(synapse_t *parent, xmlNode *xml_action)
{
    action_type_e action_type;
    crm_action_t *action = NULL;
    const char *element = TYPE(xml_action);
    const char *value = ID(xml_action);

    if (value == NULL) {
        crm_err("Ignoring transition graph action without id (bug?)");
        crm_log_xml_trace(xml_action, "invalid");
        return NULL;
    }

    if (pcmk__str_eq(element, XML_GRAPH_TAG_RSC_OP, pcmk__str_casei)) {
        action_type = action_type_rsc;

    } else if (pcmk__str_eq(element, XML_GRAPH_TAG_PSEUDO_EVENT,
                            pcmk__str_casei)) {
        action_type = action_type_pseudo;

    } else if (pcmk__str_eq(element, XML_GRAPH_TAG_CRM_EVENT,
                            pcmk__str_casei)) {
        action_type = action_type_crm;

    } else {
        crm_err("Ignoring transition graph action of unknown type '%s' (bug?)",
                element);
        crm_log_xml_trace(xml_action, "invalid");
        return NULL;
    }

    action = calloc(1, sizeof(crm_action_t));
    if (action == NULL) {
        crm_perror(LOG_CRIT, "Cannot unpack transition graph action");
        crm_log_xml_trace(xml_action, "lost");
        return NULL;
    }

    pcmk__scan_min_int(value, &(action->id), -1);
    action->type = action_type_rsc;
    action->xml = copy_xml(xml_action);
    action->synapse = parent;
    action->type = action_type;
    action->params = xml2list(action->xml);

    value = g_hash_table_lookup(action->params, "CRM_meta_timeout");
    pcmk__scan_min_int(value, &(action->timeout), 0);

    /* Take start-delay into account for the timeout of the action timer */
    value = g_hash_table_lookup(action->params, "CRM_meta_start_delay");
    {
        int start_delay;

        pcmk__scan_min_int(value, &start_delay, 0);
        action->timeout += start_delay;
    }

    if (pcmk__guint_from_hash(action->params,
                              CRM_META "_" XML_LRM_ATTR_INTERVAL, 0,
                              &(action->interval_ms)) != pcmk_rc_ok) {
        action->interval_ms = 0;
    }

    value = g_hash_table_lookup(action->params, "CRM_meta_can_fail");
    if (value != NULL) {

        gboolean can_fail = FALSE;
        crm_str_to_boolean(value, &can_fail);
        if (can_fail) {
            crm__set_graph_action_flags(action, pcmk__graph_action_can_fail);
        } else {
            crm__clear_graph_action_flags(action, pcmk__graph_action_can_fail);
        }

#ifndef PCMK__COMPAT_2_0
        if (pcmk_is_set(action->flags, pcmk__graph_action_can_fail)) {
            crm_warn("Support for the can_fail meta-attribute is deprecated"
                     " and will be removed in a future release");
        }
#endif
    }

    crm_trace("Action %d has timer set to %dms", action->id, action->timeout);

    return action;
}

static synapse_t *
unpack_synapse(crm_graph_t *new_graph, xmlNode *xml_synapse)
{
    const char *value = NULL;
    xmlNode *action_set = NULL;
    synapse_t *new_synapse = NULL;

    crm_trace("Unpacking synapse %s", ID(xml_synapse));

    new_synapse = calloc(1, sizeof(synapse_t));
    if (new_synapse == NULL) {
        return NULL;
    }

    pcmk__scan_min_int(ID(xml_synapse), &(new_synapse->id), 0);

    value = crm_element_value(xml_synapse, XML_CIB_ATTR_PRIORITY);
    pcmk__scan_min_int(value, &(new_synapse->priority), 0);

    CRM_CHECK(new_synapse->id >= 0, free(new_synapse);
                                    return NULL);

    new_graph->num_synapses++;

    crm_trace("Unpacking synapse %s action sets",
              crm_element_value(xml_synapse, XML_ATTR_ID));

    for (action_set = first_named_child(xml_synapse, "action_set");
         action_set != NULL; action_set = crm_next_same_xml(action_set)) {

        for (xmlNode *action = pcmk__xml_first_child(action_set);
             action != NULL; action = pcmk__xml_next(action)) {

            crm_action_t *new_action = unpack_action(new_synapse, action);

            if (new_action == NULL) {
                continue;
            }

            crm_trace("Adding action %d to synapse %d",
                      new_action->id, new_synapse->id);
            new_graph->num_actions++;
            new_synapse->actions = g_list_append(new_synapse->actions,
                                                 new_action);
        }
    }

    crm_trace("Unpacking synapse %s inputs", ID(xml_synapse));

    for (xmlNode *inputs = first_named_child(xml_synapse, "inputs");
         inputs != NULL; inputs = crm_next_same_xml(inputs)) {

        for (xmlNode *trigger = first_named_child(inputs, "trigger");
             trigger != NULL; trigger = crm_next_same_xml(trigger)) {

            for (xmlNode *input = pcmk__xml_first_child(trigger);
                 input != NULL; input = pcmk__xml_next(input)) {

                crm_action_t *new_input = unpack_action(new_synapse, input);

                if (new_input == NULL) {
                    continue;
                }

                crm_trace("Adding input %d to synapse %d",
                           new_input->id, new_synapse->id);

                new_synapse->inputs = g_list_append(new_synapse->inputs,
                                                    new_input);
            }
        }
    }

    return new_synapse;
}

crm_graph_t *
pcmk__unpack_graph(xmlNode *xml_graph, const char *reference)
{
    crm_graph_t *new_graph = NULL;
    const char *t_id = NULL;
    const char *time = NULL;

    new_graph = calloc(1, sizeof(crm_graph_t));
    if (new_graph == NULL) {
        return NULL;
    }

    new_graph->source = strdup((reference == NULL)? "unknown" : reference);
    if (new_graph->source == NULL) {
        free(new_graph);
        return NULL;
    }

    new_graph->id = -1;
    new_graph->abort_priority = 0;
    new_graph->network_delay = 0;
    new_graph->stonith_timeout = 0;
    new_graph->completion_action = tg_done;

    // Parse top-level attributes from <transition_graph>
    if (xml_graph != NULL) {
        t_id = crm_element_value(xml_graph, "transition_id");
        CRM_CHECK(t_id != NULL, free(new_graph);
                                return NULL);
        pcmk__scan_min_int(t_id, &(new_graph->id), -1);

        time = crm_element_value(xml_graph, "cluster-delay");
        CRM_CHECK(time != NULL, free(new_graph);
                                return NULL);
        new_graph->network_delay = crm_parse_interval_spec(time);

        time = crm_element_value(xml_graph, "stonith-timeout");
        if (time == NULL) {
            new_graph->stonith_timeout = new_graph->network_delay;
        } else {
            new_graph->stonith_timeout = crm_parse_interval_spec(time);
        }

        // Use 0 (dynamic limit) as default/invalid, -1 (no limit) as minimum
        t_id = crm_element_value(xml_graph, "batch-limit");
        if ((t_id == NULL)
            || (pcmk__scan_min_int(t_id, &(new_graph->batch_limit),
                                   -1) != pcmk_rc_ok)) {
            new_graph->batch_limit = 0;
        }

        t_id = crm_element_value(xml_graph, "migration-limit");
        pcmk__scan_min_int(t_id, &(new_graph->migration_limit), -1);
    }

    // Unpack each child <synapse> element
    for (xmlNode *synapse_xml = first_named_child(xml_graph, "synapse");
         synapse_xml != NULL; synapse_xml = crm_next_same_xml(synapse_xml)) {

        synapse_t *new_synapse = unpack_synapse(new_graph, synapse_xml);

        if (new_synapse != NULL) {
            new_graph->synapses = g_list_append(new_graph->synapses,
                                                new_synapse);
        }
    }

    crm_debug("Unpacked transition %d from %s: %d actions in %d synapses",
              new_graph->id, new_graph->source, new_graph->num_actions,
              new_graph->num_synapses);

    return new_graph;
}


/*
 * Functions for freeing transition graph objects
 */

static void
free_graph_action(gpointer user_data)
{
    crm_action_t *action = user_data;

    if ((action->timer != NULL) && (action->timer->source_id != 0)) {
        crm_warn("Cancelling timer for graph action %d", action->id);
        g_source_remove(action->timer->source_id);
    }
    if (action->params != NULL) {
        g_hash_table_destroy(action->params);
    }
    free_xml(action->xml);
    free(action->timer);
    free(action);
}

static void
free_graph_synapse(gpointer user_data)
{
    synapse_t *synapse = user_data;

    g_list_free_full(synapse->actions, free_graph_action);
    g_list_free_full(synapse->inputs, free_graph_action);
    free(synapse);
}

void
pcmk__free_graph(crm_graph_t *graph)
{
    if (graph != NULL) {
        g_list_free_full(graph->synapses, free_graph_synapse);
        free(graph->source);
        free(graph);
    }
}


/*
 * Other transition graph utilities
 */

lrmd_event_data_t *
pcmk__event_from_graph_action(xmlNode *resource, crm_action_t *action,
                              int status, int rc, const char *exit_reason)
{
    lrmd_event_data_t *op = NULL;
    GHashTableIter iter;
    const char *name = NULL;
    const char *value = NULL;
    xmlNode *action_resource = NULL;

    CRM_CHECK(action != NULL, return NULL);
    CRM_CHECK(action->type == action_type_rsc, return NULL);

    action_resource = first_named_child(action->xml, XML_CIB_TAG_RESOURCE);
    CRM_CHECK(action_resource != NULL, crm_log_xml_warn(action->xml, "invalid");
                                       return NULL);

    op = lrmd_new_event(ID(action_resource),
                        crm_element_value(action->xml, XML_LRM_ATTR_TASK),
                        action->interval_ms);
    lrmd__set_result(op, rc, status, exit_reason);
    op->t_run = time(NULL);
    op->t_rcchange = op->t_run;
    op->params = pcmk__strkey_table(free, free);

    g_hash_table_iter_init(&iter, action->params);
    while (g_hash_table_iter_next(&iter, (void **)&name, (void **)&value)) {
        g_hash_table_insert(op->params, strdup(name), strdup(value));
    }

    for (xmlNode *xop = pcmk__xml_first_child(resource); xop != NULL;
         xop = pcmk__xml_next(xop)) {
        int tmp = 0;

        crm_element_value_int(xop, XML_LRM_ATTR_CALLID, &tmp);
        crm_debug("Got call_id=%d for %s", tmp, ID(resource));
        if (tmp > op->call_id) {
            op->call_id = tmp;
        }
    }

    op->call_id++;
    return op;
}