void send_stonith_update(stonith_ops_t * op) { enum cib_errors rc = cib_ok; const char *target = op->node_name; const char *uuid = op->node_uuid; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = create_xml_node(NULL, XML_CIB_TAG_STATE); CRM_CHECK(op->node_name != NULL, return); CRM_CHECK(op->node_uuid != NULL, return); crm_xml_add(node_state, XML_ATTR_UUID, uuid); crm_xml_add(node_state, XML_ATTR_UNAME, target); crm_xml_add(node_state, XML_CIB_ATTR_HASTATE, DEADSTATUS); crm_xml_add(node_state, XML_CIB_ATTR_INCCM, XML_BOOLEAN_NO); crm_xml_add(node_state, XML_CIB_ATTR_CRMDSTATE, OFFLINESTATUS); crm_xml_add(node_state, XML_CIB_ATTR_JOINSTATE, CRMD_JOINSTATE_DOWN); crm_xml_add(node_state, XML_CIB_ATTR_EXPSTATE, CRMD_JOINSTATE_DOWN); crm_xml_add(node_state, XML_ATTR_ORIGIN, __FUNCTION__); rc = fsa_cib_conn->cmds->update( fsa_cib_conn, XML_CIB_TAG_STATUS, node_state, cib_quorum_override|cib_scope_local|cib_can_create); if(rc < cib_ok) { crm_err("CIB update failed: %s", cib_error2string(rc)); abort_transition( INFINITY, tg_shutdown, "CIB update failed", node_state); } else { /* delay processing the trigger until the update completes */ add_cib_op_callback(fsa_cib_conn, rc, FALSE, crm_strdup(target), cib_fencing_updated); } erase_status_tag(op->node_name, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(op->node_name, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); free_xml(node_state); #if 0 /* Make sure the membership cache is accurate */ crm_update_peer(0, 0, 0, -1, 0, uuid, target, NULL, CRM_NODE_LOST); #endif return; }
void send_stonith_update(crm_action_t * action, const char *target, const char *uuid) { int rc = pcmk_ok; crm_node_t *peer = NULL; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = NULL; CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); if(get_node_uuid(0, target) == NULL) { set_node_uuid(target, uuid); } /* Make sure the membership and join caches are accurate */ peer = crm_get_peer(0, target); if(peer->uuid == NULL) { crm_info("Recording uuid '%s' for node '%s'", uuid, target); peer->uuid = strdup(uuid); } crm_update_peer_proc(__FUNCTION__, peer, crm_proc_none, NULL); crm_update_peer_state(__FUNCTION__, peer, CRM_NODE_LOST, 0); crm_update_peer_expected(__FUNCTION__, peer, CRMD_JOINSTATE_DOWN); erase_node_from_join(target); node_state = do_update_node_cib(peer, node_update_cluster|node_update_peer|node_update_join|node_update_expected, NULL, __FUNCTION__); /* Force our known ID */ crm_xml_add(node_state, XML_ATTR_UUID, uuid); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, node_state, cib_quorum_override | cib_scope_local | cib_can_create); /* Delay processing the trigger until the update completes */ crm_debug("Sending fencing update %d for %s", rc, target); add_cib_op_callback(fsa_cib_conn, rc, FALSE, strdup(target), cib_fencing_updated); /* Make sure it sticks */ /* fsa_cib_conn->cmds->bump_epoch(fsa_cib_conn, cib_quorum_override|cib_scope_local); */ erase_status_tag(target, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(target, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); free_xml(node_state); return; }
void send_stonith_update(crm_action_t * action, const char *target, const char *uuid) { enum cib_errors rc = cib_ok; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = create_xml_node(NULL, XML_CIB_TAG_STATE); CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); crm_xml_add(node_state, XML_ATTR_UUID, uuid); crm_xml_add(node_state, XML_ATTR_UNAME, target); crm_xml_add(node_state, XML_CIB_ATTR_HASTATE, DEADSTATUS); crm_xml_add(node_state, XML_CIB_ATTR_INCCM, XML_BOOLEAN_NO); crm_xml_add(node_state, XML_CIB_ATTR_CRMDSTATE, OFFLINESTATUS); crm_xml_add(node_state, XML_CIB_ATTR_JOINSTATE, CRMD_JOINSTATE_DOWN); crm_xml_add(node_state, XML_CIB_ATTR_EXPSTATE, CRMD_JOINSTATE_DOWN); crm_xml_add(node_state, XML_ATTR_ORIGIN, __FUNCTION__); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, node_state, cib_quorum_override | cib_scope_local | cib_can_create); /* Delay processing the trigger until the update completes */ crm_info("Sending fencing update %d for %s", rc, target); add_cib_op_callback(fsa_cib_conn, rc, FALSE, crm_strdup(target), cib_fencing_updated); /* Make sure it sticks */ /* fsa_cib_conn->cmds->bump_epoch(fsa_cib_conn, cib_quorum_override|cib_scope_local); */ erase_status_tag(target, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(target, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); free_xml(node_state); /* Make sure the membership cache is accurate */ crm_update_peer(0, 0, 0, -1, crm_proc_none, uuid, target, NULL, CRM_NODE_LOST); return; }
/*! * \internal * \brief Handle cluster communication related to pacemaker_remote node leaving * * \param[in] node_name Name of lost node * \param[in] opts Whether to keep or erase LRM history */ static void remote_node_down(const char *node_name, const enum down_opts opts) { xmlNode *update; int call_id = 0; int call_opt = crmd_cib_smart_opt(); crm_node_t *node; /* Purge node from attrd's memory */ update_attrd_remote_node_removed(node_name, NULL); /* Purge node's transient attributes */ erase_status_tag(node_name, XML_TAG_TRANSIENT_NODEATTRS, call_opt); /* Normally, the LRM operation history should be kept until the node comes * back up. However, after a successful fence, we want to clear it, so we * don't think resources are still running on the node. */ if (opts == DOWN_ERASE_LRM) { erase_status_tag(node_name, XML_CIB_TAG_LRM, call_opt); } /* Ensure node is in the remote peer cache with lost state */ node = crm_remote_peer_get(node_name); CRM_CHECK(node != NULL, return); crm_update_peer_state(__FUNCTION__, node, CRM_NODE_LOST, 0); /* Notify DC */ send_remote_state_message(node_name, FALSE); /* Update CIB node state */ update = create_xml_node(NULL, XML_CIB_TAG_STATUS); create_node_state_update(node, node_update_cluster, update, __FUNCTION__); fsa_cib_update(XML_CIB_TAG_STATUS, update, call_opt, call_id, NULL); if (call_id < 0) { crm_perror(LOG_ERR, "%s CIB node state update", node_name); } free_xml(update); }
/* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; int call_id = 0; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *op = crm_element_value(join_ack->msg, F_CRM_TASK); const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM); crm_node_t *peer = crm_get_peer(0, join_from); if (safe_str_neq(op, CRM_OP_JOIN_CONFIRM) || peer == NULL) { crm_debug("Ignoring op=%s message from %s", op, join_from); return; } crm_trace("Processing ack from %s", join_from); crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id); if (peer->join != crm_join_finalized) { crm_info("Join not in progress: ignoring join-%d from %s (phase = %d)", join_id, join_from, peer->join); return; } else if (join_id != current_join_id) { crm_err("Invalid response from %s: join-%d vs. join-%d", join_from, join_id, current_join_id); crm_update_peer_join(__FUNCTION__, peer, crm_join_nack); return; } crm_update_peer_join(__FUNCTION__, peer, crm_join_confirmed); crm_info("join-%d: Updating node state to %s for %s", join_id, CRMD_JOINSTATE_MEMBER, join_from); /* update CIB with the current LRM status from the node * We dont need to notify the TE of these updates, a transition will * be started in due time */ erase_status_tag(join_from, XML_CIB_TAG_LRM, cib_scope_local); fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml, cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL); fsa_register_cib_callback(call_id, FALSE, NULL, join_update_complete_callback); crm_debug("join-%d: Registered callback for LRM update %d", join_id, call_id); }
/* CRM_OP_JOIN_ACKNACKを送信後、ノードからCRM_OP_JOIN_CONFIRMメッセージが送られて来たときの処理 */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t *msg_data) { int join_id = -1; int call_id = 0; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_id_s = NULL; const char *join_state = NULL; const char *op = crm_element_value(join_ack->msg, F_CRM_TASK); const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM); if(safe_str_neq(op, CRM_OP_JOIN_CONFIRM)) { /* CRM_OP_JOIN_CONFIRMメッセージ以外のメッセージは無視してログを出力 */ crm_debug("Ignoring op=%s message from %s", op, join_from); return; } crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id); join_id_s = crm_element_value(join_ack->msg, F_CRM_JOIN_ID); /* now update them to "member" */ crm_debug_2("Processing ack from %s", join_from); /* CRM_OP_JOIN_CONFIRMを送ってきたノードがfinalized_nodesハッシュテーブルに含まれるかサーチする */ join_state = (const char *) g_hash_table_lookup(finalized_nodes, join_from); if(join_state == NULL) { /* 含まれていない場合は、無視 */ crm_err("Join not in progress: ignoring join-%d from %s", join_id, join_from); return; } else if(safe_str_neq(join_state, CRMD_JOINSTATE_MEMBER)) { crm_err("Node %s wasnt invited to join the cluster",join_from); /* 含まれていて、CRMD_JOINSTATE_MEMBER状態の場合は、finalized_nodesハッシュテーブルからCRM_OP_JOIN_CONFIRMを送ってきたノードを削除 */ g_hash_table_remove(finalized_nodes, join_from); return; } else if(join_id != current_join_id) { crm_err("Invalid response from %s: join-%d vs. join-%d", join_from, join_id, current_join_id); /* JOIN_IDが異なる場合も、finalized_nodesハッシュテーブルからCRM_OP_JOIN_CONFIRMを送ってきたノードを削除 */ g_hash_table_remove(finalized_nodes, join_from); return; } /* 上記以外の場合も、finalized_nodesハッシュテーブルからCRM_OP_JOIN_CONFIRMを送ってきたノードを削除 */ g_hash_table_remove(finalized_nodes, join_from); /* confirmed_nodesハッシュテーブルにRM_OP_JOIN_CONFIRMを送ってきたノードでサーチ */ if(g_hash_table_lookup(confirmed_nodes, join_from) != NULL) { /* 既に登録済みの場合は、エラーメッセージを表示 */ crm_err("join-%d: hash already contains confirmation from %s", join_id, join_from); } /* confirmed_nodesハッシュテーブルにCRM_OP_JOIN_CONFIRMを送ってきたノードを追加 */ g_hash_table_insert( confirmed_nodes, crm_strdup(join_from), crm_strdup(join_id_s)); /* join完了ログを出す */ crm_info("join-%d: Updating node state to %s for %s", join_id, CRMD_JOINSTATE_MEMBER, join_from); /* update CIB with the current LRM status from the node * We dont need to notify the TE of these updates, a transition will * be started in due time */ /* CIBからCRM_OP_JOIN_CONFIRMを送ってきたノードのXML_CIB_TAG_LRM情報を消去 */ erase_status_tag(join_from, XML_CIB_TAG_LRM, cib_scope_local); /* CIBを更新 */ /* CRM_OP_JOIN_CONFIRMを送ってきたノードのXML_CIB_TAG_LRM情報も追加される */ /* クラスタ構成ノードのLRMD情報が最新で更新 */ fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml, cib_scope_local|cib_quorum_override|cib_can_create, call_id); /* CIBの更新コールバックをセット */ add_cib_op_callback( fsa_cib_conn, call_id, FALSE, NULL, join_update_complete_callback); crm_debug("join-%d: Registered callback for LRM update %d", join_id, call_id); }
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); }
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); }
void send_stonith_update(crm_action_t * action, const char *target, const char *uuid) { int rc = pcmk_ok; crm_node_t *peer = NULL; /* We (usually) rely on the membership layer to do node_update_cluster, * and the peer status callback to do node_update_peer, because the node * might have already rejoined before we get the stonith result here. */ int flags = node_update_join | node_update_expected; /* zero out the node-status & remove all LRM status info */ xmlNode *node_state = NULL; CRM_CHECK(target != NULL, return); CRM_CHECK(uuid != NULL, return); /* Make sure the membership and join caches are accurate */ peer = crm_get_peer_full(0, target, CRM_GET_PEER_ANY); CRM_CHECK(peer != NULL, return); if (peer->state == NULL) { /* Usually, we rely on the membership layer to update the cluster state * in the CIB. However, if the node has never been seen, do it here, so * the node is not considered unclean. */ flags |= node_update_cluster; } if (peer->uuid == NULL) { crm_info("Recording uuid '%s' for node '%s'", uuid, target); peer->uuid = strdup(uuid); } crmd_peer_down(peer, TRUE); /* Generate a node state update for the CIB */ node_state = do_update_node_cib(peer, flags, NULL, __FUNCTION__); /* we have to mark whether or not remote nodes have already been fenced */ if (peer->flags & crm_remote_node) { time_t now = time(NULL); char *now_s = crm_itoa(now); crm_xml_add(node_state, XML_NODE_IS_FENCED, now_s); free(now_s); } /* Force our known ID */ crm_xml_add(node_state, XML_ATTR_UUID, uuid); rc = fsa_cib_conn->cmds->update(fsa_cib_conn, XML_CIB_TAG_STATUS, node_state, cib_quorum_override | cib_scope_local | cib_can_create); /* Delay processing the trigger until the update completes */ crm_debug("Sending fencing update %d for %s", rc, target); fsa_register_cib_callback(rc, FALSE, strdup(target), cib_fencing_updated); /* Make sure it sticks */ /* fsa_cib_conn->cmds->bump_epoch(fsa_cib_conn, cib_quorum_override|cib_scope_local); */ erase_status_tag(peer->uname, XML_CIB_TAG_LRM, cib_scope_local); erase_status_tag(peer->uname, XML_TAG_TRANSIENT_NODEATTRS, cib_scope_local); free_xml(node_state); return; }
/* aka. this is notification that we have (or have not) been accepted */ void do_cl_join_finalize_respond(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *tmp1 = NULL; gboolean was_nack = TRUE; static gboolean first_join = TRUE; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); const char *start_state = daemon_option("node_start_state"); int join_id = -1; const char *op = crm_element_value(input->msg, F_CRM_TASK); const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK); const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM); if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) { crm_trace("Ignoring op=%s message", op); return; } /* calculate if it was an ack or a nack */ if (crm_is_true(ack_nack)) { was_nack = FALSE; } crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id); if (was_nack) { crm_err("Shutting down because cluster join with leader %s failed " CRM_XS" join-%d NACK'd", welcome_from, join_id); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) { crm_warn("Discarding our own welcome - we're no longer the DC"); return; } if (update_dc(input->msg) == FALSE) { crm_warn("Discarding %s from node %s (expected from %s)", op, welcome_from, fsa_our_dc); return; } update_dc_expected(input->msg); /* send our status section to the DC */ tmp1 = do_lrm_query(TRUE, fsa_our_uname); if (tmp1 != NULL) { xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id); crm_debug("Confirming join-%d: sending local operation history to %s", join_id, fsa_our_dc); /* * If this is the node's first join since the crmd started on it, clear * any previous transient node attributes, to handle the case where * the node restarted so quickly that the cluster layer didn't notice. * * Do not remove the resources though, they'll be cleaned up in * do_dc_join_ack(). Removing them here creates a race condition if the * crmd is being recovered. Instead of a list of active resources from * the lrmd, we may end up with a blank status section. If we are _NOT_ * lucky, we will probe for the "wrong" instance of anonymous clones and * end up with multiple active instances on the machine. */ if (first_join && is_not_set(fsa_input_register, R_SHUTDOWN)) { first_join = FALSE; erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0); update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE); update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE); if (start_state) { set_join_state(start_state); } } send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE); free_xml(reply); if (AM_I_DC == FALSE) { register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__); update_attrd(NULL, NULL, NULL, NULL, FALSE); } free_xml(tmp1); } else { crm_err("Could not confirm join-%d with %s: Local operation history failed", join_id, fsa_our_dc); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } }
/* A_DC_JOIN_PROCESS_ACK */ void do_dc_join_ack(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { int join_id = -1; int call_id = 0; ha_msg_input_t *join_ack = fsa_typed_data(fsa_dt_ha_msg); const char *join_id_s = NULL; const char *join_state = NULL; const char *op = crm_element_value(join_ack->msg, F_CRM_TASK); const char *join_from = crm_element_value(join_ack->msg, F_CRM_HOST_FROM); if (safe_str_neq(op, CRM_OP_JOIN_CONFIRM)) { crm_debug("Ignoring op=%s message from %s", op, join_from); return; } crm_element_value_int(join_ack->msg, F_CRM_JOIN_ID, &join_id); join_id_s = crm_element_value(join_ack->msg, F_CRM_JOIN_ID); /* now update them to "member" */ crm_trace("Processing ack from %s", join_from); join_state = (const char *) g_hash_table_lookup(finalized_nodes, join_from); if (join_state == NULL) { crm_err("Join not in progress: ignoring join-%d from %s", join_id, join_from); return; } else if (safe_str_neq(join_state, CRMD_JOINSTATE_MEMBER)) { crm_err("Node %s wasnt invited to join the cluster", join_from); g_hash_table_remove(finalized_nodes, join_from); return; } else if (join_id != current_join_id) { crm_err("Invalid response from %s: join-%d vs. join-%d", join_from, join_id, current_join_id); g_hash_table_remove(finalized_nodes, join_from); return; } g_hash_table_remove(finalized_nodes, join_from); if (g_hash_table_lookup(confirmed_nodes, join_from) != NULL) { crm_err("join-%d: hash already contains confirmation from %s", join_id, join_from); } g_hash_table_insert(confirmed_nodes, strdup(join_from), strdup(join_id_s)); crm_info("join-%d: Updating node state to %s for %s", join_id, CRMD_JOINSTATE_MEMBER, join_from); /* update CIB with the current LRM status from the node * We dont need to notify the TE of these updates, a transition will * be started in due time */ erase_status_tag(join_from, XML_CIB_TAG_LRM, cib_scope_local); fsa_cib_update(XML_CIB_TAG_STATUS, join_ack->xml, cib_scope_local | cib_quorum_override | cib_can_create, call_id, NULL); add_cib_op_callback(fsa_cib_conn, call_id, FALSE, NULL, join_update_complete_callback); crm_debug("join-%d: Registered callback for LRM update %d", join_id, call_id); }
/* aka. this is notification that we have (or have not) been accepted */ void do_cl_join_finalize_respond(long long action, enum crmd_fsa_cause cause, enum crmd_fsa_state cur_state, enum crmd_fsa_input current_input, fsa_data_t * msg_data) { xmlNode *tmp1 = NULL; gboolean was_nack = TRUE; static gboolean first_join = TRUE; ha_msg_input_t *input = fsa_typed_data(fsa_dt_ha_msg); int join_id = -1; const char *op = crm_element_value(input->msg, F_CRM_TASK); const char *ack_nack = crm_element_value(input->msg, CRM_OP_JOIN_ACKNAK); const char *welcome_from = crm_element_value(input->msg, F_CRM_HOST_FROM); if (safe_str_neq(op, CRM_OP_JOIN_ACKNAK)) { crm_trace("Ignoring op=%s message", op); return; } /* calculate if it was an ack or a nack */ if (crm_is_true(ack_nack)) { was_nack = FALSE; } crm_element_value_int(input->msg, F_CRM_JOIN_ID, &join_id); if (was_nack) { crm_err("Join (join-%d) with leader %s failed (NACK'd): Shutting down", join_id, welcome_from); register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL); return; } if (AM_I_DC == FALSE && safe_str_eq(welcome_from, fsa_our_uname)) { crm_warn("Discarding our own welcome - we're no longer the DC"); return; } if (update_dc(input->msg) == FALSE) { crm_warn("Discarding %s from %s (expected %s)", op, welcome_from, fsa_our_dc); return; } /* send our status section to the DC */ crm_debug("Confirming join join-%d: %s", join_id, crm_element_value(input->msg, F_CRM_TASK)); tmp1 = do_lrm_query(TRUE, fsa_our_uname); if (tmp1 != NULL) { xmlNode *reply = create_request(CRM_OP_JOIN_CONFIRM, tmp1, fsa_our_dc, CRM_SYSTEM_DC, CRM_SYSTEM_CRMD, NULL); crm_xml_add_int(reply, F_CRM_JOIN_ID, join_id); crm_debug("join-%d: Join complete." " Sending local LRM status to %s", join_id, fsa_our_dc); if (first_join) { first_join = FALSE; /* * Clear any previous transient node attribute and lrm operations * * Corosync has a nasty habit of not being able to tell if a * node is returning or didn't leave in the first place. * This confuses Pacemaker because it never gets a "node up" * event which is normally used to clean up the status section. * * Do not remove the resources though, they'll be cleaned up in * do_dc_join_ack(). Removing them here creates a race * condition if the crmd is being recovered. * Instead of a list of active resources from the lrmd * we may end up with a blank status section. * If we are _NOT_ lucky, we will probe for the "wrong" instance * of anonymous clones and end up with multiple active * instances on the machine. */ erase_status_tag(fsa_our_uname, XML_TAG_TRANSIENT_NODEATTRS, 0); /* Just in case attrd was still around too */ if (is_not_set(fsa_input_register, R_SHUTDOWN)) { update_attrd(fsa_our_uname, "terminate", NULL, NULL, FALSE); update_attrd(fsa_our_uname, XML_CIB_ATTR_SHUTDOWN, "0", NULL, FALSE); } } send_cluster_message(crm_get_peer(0, fsa_our_dc), crm_msg_crmd, reply, TRUE); free_xml(reply); if (AM_I_DC == FALSE) { register_fsa_input_adv(cause, I_NOT_DC, NULL, A_NOTHING, TRUE, __FUNCTION__); update_attrd(NULL, NULL, NULL, NULL, FALSE); } free_xml(tmp1); } else { crm_err("Could not send our LRM state to the DC"); register_fsa_error(C_FSA_INTERNAL, I_FAIL, NULL); } }