Esempio n. 1
0
void
create_node_entry(const char *uuid, const char *uname, const char *type)
{

    /* make sure a node entry exists for the new node
     *
     * this will add anyone except the first ever node in the cluster
     *   since it will also be the DC which doesnt go through the
     *   join process (with itself).  We can include a special case
     *   later if desired.
     */
    xmlNode *tmp1 = create_xml_node(NULL, XML_CIB_TAG_NODE);

    crm_trace("Creating node entry for %s", uname);
    set_uuid(tmp1, XML_ATTR_UUID, uname);

    crm_xml_add(tmp1, XML_ATTR_UNAME, uname);
    crm_xml_add(tmp1, XML_ATTR_TYPE, type);

    fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1,
                        cib_scope_local | cib_quorum_override | cib_can_create);

    free_xml(tmp1);

}
Esempio n. 2
0
void
finalize_join_for(gpointer key, gpointer value, gpointer user_data)
{
    xmlNode *acknak = NULL;
    xmlNode *tmp1 = NULL;
    crm_node_t *join_node = value;
    const char *join_to = join_node->uname;

    if(join_node->join != crm_join_integrated) {
        crm_trace("Skipping %s in state %d", join_to, join_node->join);
        return;
    }

    /* make sure a node entry exists for the new node */
    crm_trace("Creating node entry for %s", join_to);

    tmp1 = create_xml_node(NULL, XML_CIB_TAG_NODE);
    set_uuid(tmp1, XML_ATTR_UUID, join_node);
    crm_xml_add(tmp1, XML_ATTR_UNAME, join_to);

    fsa_cib_anon_update(XML_CIB_TAG_NODES, tmp1,
                        cib_scope_local | cib_quorum_override | cib_can_create);
    free_xml(tmp1);

    join_node = crm_get_peer(0, join_to);
    if (crm_is_peer_active(join_node) == FALSE) {
        /*
         * NACK'ing nodes that the membership layer doesn't know about yet
         * simply creates more churn
         *
         * Better to leave them waiting and let the join restart when
         * the new membership event comes in
         *
         * All other NACKs (due to versions etc) should still be processed
         */
        crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_PENDING);
        return;
    }

    /* send the ack/nack to the node */
    acknak = create_request(CRM_OP_JOIN_ACKNAK, NULL, join_to,
                            CRM_SYSTEM_CRMD, CRM_SYSTEM_DC, NULL);
    crm_xml_add_int(acknak, F_CRM_JOIN_ID, current_join_id);

    crm_debug("join-%d: ACK'ing join request from %s",
              current_join_id, join_to);
    crm_xml_add(acknak, CRM_OP_JOIN_ACKNAK, XML_BOOLEAN_TRUE);
    crm_update_peer_join(__FUNCTION__, join_node, crm_join_finalized);
    crm_update_peer_expected(__FUNCTION__, join_node, CRMD_JOINSTATE_MEMBER);

    send_cluster_message(crm_get_peer(0, join_to), crm_msg_crmd, acknak, TRUE);
    free_xml(acknak);
    return;
}
Esempio n. 3
0
enum crmd_fsa_input
handle_shutdown_request(xmlNode * stored_msg)
{
    /* handle here to avoid potential version issues
     *   where the shutdown message/proceedure may have
     *   been changed in later versions.
     *
     * This way the DC is always in control of the shutdown
     */

    char *now_s = NULL;
    time_t now = time(NULL);
    xmlNode *node_state = NULL;
    const char *host_from = crm_element_value(stored_msg, F_CRM_HOST_FROM);

    if (host_from == NULL) {
        /* we're shutting down and the DC */
        host_from = fsa_our_uname;
    }

    crm_info("Creating shutdown request for %s (state=%s)", host_from, fsa_state2string(fsa_state));

    crm_log_xml(LOG_MSG, "message", stored_msg);

    node_state = create_node_state(host_from, NULL, NULL, NULL, NULL,
                                   CRMD_STATE_INACTIVE, FALSE, __FUNCTION__);

    fsa_cib_anon_update(XML_CIB_TAG_STATUS, node_state, cib_quorum_override);
    crm_log_xml_debug_2(node_state, "Shutdown update");
    free_xml(node_state);

    now_s = crm_itoa(now);
    update_attrd(host_from, XML_CIB_ATTR_SHUTDOWN, now_s, NULL);
    crm_free(now_s);

    /* will be picked up by the TE as long as its running */
    return I_NULL;
}
Esempio n. 4
0
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
    uint32_t old = 0;
    uint32_t changed = 0;
    bool appeared = FALSE;
    bool is_remote = is_set(node->flags, crm_remote_node);
    const char *status = NULL;

    /* Crmd waits to receive some information from the membership layer before
     * declaring itself operational. If this is being called for a cluster node,
     * indicate that we have it.
     */
    if (!is_remote) {
        set_bit(fsa_input_register, R_PEER_DATA);
    }

    if (node->uname == NULL) {
        return;
    }

    switch (type) {
        case crm_status_uname:
            /* If we've never seen the node, then it also won't be in the status section */
            crm_info("%s node %s is now %s",
                     (is_remote? "Remote" : "Cluster"),
                     node->uname, state_text(node->state));
            return;

        case crm_status_rstate:
        case crm_status_nstate:
            /* This callback should not be called unless the state actually
             * changed, but here's a failsafe just in case.
             */
            CRM_CHECK(safe_str_neq(data, node->state), return);

            crm_info("%s node %s is now %s (was %s)",
                     (is_remote? "Remote" : "Cluster"),
                     node->uname, state_text(node->state), state_text(data));

            if (safe_str_eq(CRM_NODE_MEMBER, node->state)) {
                appeared = TRUE;
                if (!is_remote) {
                    remove_stonith_cleanup(node->uname);
                }
            }

            crmd_alert_node_event(node);
            break;

        case crm_status_processes:
            if (data) {
                old = *(const uint32_t *)data;
                changed = node->processes ^ old;
            }

            status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
            crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
                     node->uname, peer2text(proc_flags), status,
                     AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);

            if ((changed & proc_flags) == 0) {
                /* Peer process did not change */
                crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
                return;
            } else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
                crm_trace("Not connected");
                return;
            } else if (fsa_state == S_STOPPING) {
                crm_trace("Stopping");
                return;
            }

            appeared = (node->processes & proc_flags) != 0;
            if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
                /* Did we get evicted? */
                crm_notice("Our peer connection failed");
                register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);

            } else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
                /* Did the DC leave us? */
                crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
                register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);

                /* @COMPAT DC < 1.1.13: If a DC shuts down normally, we don't
                 * want to fence it. Newer DCs will send their shutdown request
                 * to all peers, who will update the DC's expected state to
                 * down, thus avoiding fencing. We can safely erase the DC's
                 * transient attributes when it leaves in that case. However,
                 * the only way to avoid fencing older DCs is to leave the
                 * transient attributes intact until it rejoins.
                 */
                if (compare_version(fsa_our_dc_version, "3.0.9") > 0) {
                    erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
                }

            } else if(AM_I_DC && appeared == FALSE) {
                crm_info("Peer %s left us", node->uname);
                erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
            }
            break;
    }

    if (AM_I_DC) {
        xmlNode *update = NULL;
        int flags = node_update_peer;
        gboolean alive = is_remote? appeared : crm_is_peer_active(node);
        crm_action_t *down = match_down_event(node->uuid, appeared);

        crm_trace("Alive=%d, appeared=%d, down=%d",
                  alive, appeared, (down? down->id : -1));

        if (alive && type == crm_status_processes) {
            register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
        }

        if (down) {
            const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);

            if (safe_str_eq(task, CRM_OP_FENCE)) {

                /* tengine_stonith_callback() confirms fence actions */
                crm_trace("Updating CIB %s stonithd reported fencing of %s complete",
                          (down->confirmed? "after" : "before"), node->uname);

            } else if ((alive == FALSE) && safe_str_eq(task, CRM_OP_SHUTDOWN)) {
                crm_notice("%s of peer %s is complete "CRM_XS" op=%d",
                           task, node->uname, down->id);

                /* down->confirmed = TRUE; */
                stop_te_timer(down->timer);

                if (!is_remote) {
                    flags |= node_update_join | node_update_expected;
                    crmd_peer_down(node, FALSE);
                    check_join_state(fsa_state, __FUNCTION__);
                }

                update_graph(transition_graph, down);
                trigger_graph();

            } else {
                crm_trace("Node %s is %salive, was expected to %s (op %d)",
                          node->uname, (alive? "" : "not "), task, down->id);
            }

        } else if (appeared == FALSE) {
            crm_notice("Stonith/shutdown of %s not matched", node->uname);

            if (!is_remote) {
                crm_update_peer_join(__FUNCTION__, node, crm_join_none);
                check_join_state(fsa_state, __FUNCTION__);
            }

            abort_transition(INFINITY, tg_restart, "Node failure", NULL);
            fail_incompletable_actions(transition_graph, node->uuid);

        } else {
            crm_trace("Node %s came up, was not expected to be down",
                      node->uname);
        }

        if (is_remote) {
            /* A pacemaker_remote node won't have its cluster status updated
             * in the CIB by membership-layer callbacks, so do it here.
             */
            flags |= node_update_cluster;

            /* Trigger resource placement on newly integrated nodes */
            if (appeared) {
                abort_transition(INFINITY, tg_restart,
                                 "pacemaker_remote node integrated", NULL);
            }
        }

        /* Update the CIB node state */
        update = create_node_state_update(node, flags, NULL, __FUNCTION__);
        fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
                            cib_scope_local | cib_quorum_override | cib_can_create);
        free_xml(update);
    }

    trigger_fsa(fsa_source);
}
Esempio n. 5
0
void
peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
    uint32_t old = 0;
    uint32_t changed = 0;
    bool appeared = FALSE;
    const char *status = NULL;

    set_bit(fsa_input_register, R_PEER_DATA);
    if (node->uname == NULL) {
        return;
    }

    switch (type) {
        case crm_status_uname:
            /* If we've never seen the node, then it also wont be in the status section */
            crm_info("%s is now %s", node->uname, node->state);
            return;
        case crm_status_rstate:
            crm_info("Remote node %s is now %s (was %s)", node->uname, node->state, (const char *)data);
            /* Keep going */
        case crm_status_nstate:
            crm_info("%s is now %s (was %s)", node->uname, node->state, (const char *)data);
            if (safe_str_eq(data, node->state)) {
                /* State did not change */
                return;
            } else if(safe_str_eq(CRM_NODE_MEMBER, node->state)) {
                appeared = TRUE;
            }
            break;
        case crm_status_processes:
            if (data) {
                old = *(const uint32_t *)data;
                changed = node->processes ^ old;
            }

            /* crmd_proc_update(node, proc_flags); */
            status = (node->processes & proc_flags) ? ONLINESTATUS : OFFLINESTATUS;
            crm_info("Client %s/%s now has status [%s] (DC=%s, changed=%6x)",
                     node->uname, peer2text(proc_flags), status,
                     AM_I_DC ? "true" : crm_str(fsa_our_dc), changed);

            if ((changed & proc_flags) == 0) {
                /* Peer process did not change */
                crm_trace("No change %6x %6x %6x", old, node->processes, proc_flags);
                return;
            } else if (is_not_set(fsa_input_register, R_CIB_CONNECTED)) {
                crm_trace("Not connected");
                return;
            } else if (fsa_state == S_STOPPING) {
                crm_trace("Stopping");
                return;
            }

            appeared = (node->processes & proc_flags) != 0;
            if (safe_str_eq(node->uname, fsa_our_uname) && (node->processes & proc_flags) == 0) {
                /* Did we get evicted? */
                crm_notice("Our peer connection failed");
                register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ERROR, NULL);

            } else if (safe_str_eq(node->uname, fsa_our_dc) && crm_is_peer_active(node) == FALSE) {
                /* Did the DC leave us? */
                crm_notice("Our peer on the DC (%s) is dead", fsa_our_dc);
                register_fsa_input(C_CRMD_STATUS_CALLBACK, I_ELECTION, NULL);

            } else if(AM_I_DC && appeared == FALSE) {
                crm_info("Peer %s left us", node->uname);
                /* crm_update_peer_join(__FUNCTION__, node, crm_join_none); */
            }
            break;
    }

    if (AM_I_DC) {
        xmlNode *update = NULL;
        int flags = node_update_peer;
        gboolean alive = crm_is_peer_active(node);
        crm_action_t *down = match_down_event(0, node->uuid, NULL, appeared);

        crm_trace("Alive=%d, appear=%d, down=%p", alive, appeared, down);

        if (alive && type == crm_status_processes) {
            register_fsa_input_before(C_FSA_INTERNAL, I_NODE_JOIN, NULL);
        }

        if (down) {
            const char *task = crm_element_value(down->xml, XML_LRM_ATTR_TASK);

            if (alive && safe_str_eq(task, CRM_OP_FENCE)) {
                crm_info("Node return implies stonith of %s (action %d) completed", node->uname,
                         down->id);
                erase_status_tag(node->uname, XML_CIB_TAG_LRM, cib_scope_local);
                erase_status_tag(node->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local);
                /* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
                down->sent_update = TRUE;       /* Prevent tengine_stonith_callback() from calling send_stonith_update() */

            } else if (safe_str_eq(task, CRM_OP_FENCE)) {
                crm_trace("Waiting for stonithd to report the fencing of %s is complete", node->uname); /* via tengine_stonith_callback() */

            } else if (alive == FALSE) {
                crm_notice("%s of %s (op %d) is complete", task, node->uname, down->id);
                /* down->confirmed = TRUE; Only stonith-ng returning should imply completion */
                stop_te_timer(down->timer);

                flags |= node_update_join | node_update_expected;
                crmd_peer_down(node, FALSE);
                check_join_state(fsa_state, __FUNCTION__);

                update_graph(transition_graph, down);
                trigger_graph();

            } else {
                crm_trace("Other %p", down);
            }

        } else if (appeared == FALSE) {
            crm_notice("Stonith/shutdown of %s not matched", node->uname);

            crm_update_peer_join(__FUNCTION__, node, crm_join_none);
            check_join_state(fsa_state, __FUNCTION__);

            abort_transition(INFINITY, tg_restart, "Node failure", NULL);
            fail_incompletable_actions(transition_graph, node->uuid);

        } else {
            crm_trace("Other %p", down);
        }

        update = do_update_node_cib(node, flags, NULL, __FUNCTION__);
        fsa_cib_anon_update(XML_CIB_TAG_STATUS, update,
                            cib_scope_local | cib_quorum_override | cib_can_create);
        free_xml(update);
    }

    trigger_fsa(fsa_source);
}