From 7a813755269f00d7b815e819636841af991762c0 Mon Sep 17 00:00:00 2001 From: Ken Gaillot Date: Mon, 11 Dec 2017 12:23:06 -0600 Subject: [PATCH] Fix: tools: crm_resource --cleanup The new "failures only" mode of crm_resource --cleanup had multiple issues, including not working without --resource specified, comparing a user-provided interval string against a milliseconds interval, and considering no interval specified as all intervals rather than 0 but only when clearing LRM history entries. --- tools/crm_resource.c | 35 +++--- tools/crm_resource.h | 9 +- tools/crm_resource_runtime.c | 258 ++++++++++++++++++++++++++++++------------- 3 files changed, 202 insertions(+), 100 deletions(-) diff --git a/tools/crm_resource.c b/tools/crm_resource.c index 4ddcef4..5152004 100644 --- a/tools/crm_resource.c +++ b/tools/crm_resource.c @@ -1092,14 +1092,20 @@ main(int argc, char **argv) rc = cli_resource_delete_attribute(rsc, rsc_id, prop_set, prop_id, prop_name, cib_conn, &data_set); - } else if (rsc_cmd == 'C' && just_errors) { + } else if ((rsc_cmd == 'C') && rsc) { + if (do_force == FALSE) { + rsc = uber_parent(rsc); + } crmd_replies_needed = 0; - rc = cli_resource_delete_failures(crmd_channel, host_uname, rsc, operation, - interval, &data_set); + crm_debug("%s of %s (%s requested) on %s", + (just_errors? "Clearing failures" : "Re-checking the state"), + rsc->id, rsc_id, (host_uname? host_uname : "all hosts")); + rc = cli_resource_delete(crmd_channel, host_uname, rsc, operation, + interval, just_errors, &data_set); - if(rsc && (rc == pcmk_ok) && (BE_QUIET == FALSE)) { - /* Now check XML_RSC_ATTR_TARGET_ROLE and XML_RSC_ATTR_MANAGED */ + if ((rc == pcmk_ok) && !BE_QUIET) { + // Show any reasons why resource might stay stopped cli_resource_check(cib_conn, rsc); } @@ -1107,22 +1113,9 @@ main(int argc, char **argv) start_mainloop(); } - } else if ((rsc_cmd == 'C') && rsc) { - if(do_force == FALSE) { - rsc = uber_parent(rsc); - } - - crm_debug("Re-checking the state of %s (%s requested) on %s", - rsc->id, rsc_id, host_uname); - crmd_replies_needed = 0; - rc = cli_resource_delete(crmd_channel, host_uname, rsc, operation, - interval, &data_set); - - if(rc == pcmk_ok && BE_QUIET == FALSE) { - /* Now check XML_RSC_ATTR_TARGET_ROLE and XML_RSC_ATTR_MANAGED */ - cli_resource_check(cib_conn, rsc); - } - + } else if (rsc_cmd == 'C' && just_errors) { + rc = cli_cleanup_all(crmd_channel, host_uname, operation, interval, + &data_set); if (rc == pcmk_ok) { start_mainloop(); } diff --git a/tools/crm_resource.h b/tools/crm_resource.h index e28c9ef..0ac51f2 100644 --- a/tools/crm_resource.h +++ b/tools/crm_resource.h @@ -75,10 +75,11 @@ int cli_resource_search(resource_t *rsc, const char *requested_name, pe_working_set_t *data_set); int cli_resource_delete(crm_ipc_t *crmd_channel, const char *host_uname, resource_t *rsc, const char *operation, - const char *interval, pe_working_set_t *data_set); -int cli_resource_delete_failures(crm_ipc_t *crmd_channel, const char *host_uname, - resource_t *rsc, const char *operation, - const char *interval, pe_working_set_t *data_set); + const char *interval, bool just_failures, + pe_working_set_t *data_set); +int cli_cleanup_all(crm_ipc_t *crmd_channel, const char *node_name, + const char *operation, const char *interval, + pe_working_set_t *data_set); int cli_resource_restart(resource_t * rsc, const char *host, int timeout_ms, cib_t * cib); int cli_resource_move(resource_t *rsc, const char *rsc_id, const char *host_name, cib_t *cib, diff --git a/tools/crm_resource_runtime.c b/tools/crm_resource_runtime.c index 1048636..bdebb0b 100644 --- a/tools/crm_resource_runtime.c +++ b/tools/crm_resource_runtime.c @@ -532,15 +532,129 @@ rsc_fail_name(resource_t *rsc) return is_set(rsc->flags, pe_rsc_unique)? strdup(name) : clone_strip(name); } +static int +clear_rsc_history(crm_ipc_t *crmd_channel, const char *host_uname, + const char *rsc_id, pe_working_set_t *data_set) +{ + int rc = pcmk_ok; + + /* Erase the resource's entire LRM history in the CIB, even if we're only + * clearing a single operation's fail count. If we erased only entries for a + * single operation, we might wind up with a wrong idea of the current + * resource state, and we might not re-probe the resource. + */ + rc = send_lrm_rsc_op(crmd_channel, CRM_OP_LRM_DELETE, host_uname, rsc_id, + TRUE, data_set); + if (rc != pcmk_ok) { + return rc; + } + crmd_replies_needed++; + + crm_trace("Processing %d mainloop inputs", crmd_replies_needed); + while (g_main_context_iteration(NULL, FALSE)) { + crm_trace("Processed mainloop input, %d still remaining", + crmd_replies_needed); + } + + if (crmd_replies_needed < 0) { + crmd_replies_needed = 0; + } + return rc; +} + +static int +clear_rsc_failures(crm_ipc_t *crmd_channel, const char *node_name, + const char *rsc_id, const char *operation, + const char *interval, pe_working_set_t *data_set) +{ + int rc = pcmk_ok; + const char *failed_value = NULL; + const char *interval_ms_str = NULL; + GHashTable *rscs = NULL; + GHashTableIter iter; + + /* Create a hash table to use as a set of resources to clean. This lets us + * clean each resource only once (per node) regardless of how many failed + * operations it has. + */ + rscs = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, NULL); + + // Normalize interval to milliseconds for comparison to history entry + if (operation) { + interval_ms_str = crm_strdup_printf("%llu", crm_get_interval(interval)); + } + + for (xmlNode *xml_op = __xml_first_child(data_set->failed); xml_op != NULL; + xml_op = __xml_next(xml_op)) { + + // No resource specified means all resources match + failed_value = crm_element_value(xml_op, XML_LRM_ATTR_RSCID); + if (rsc_id == NULL) { + rsc_id = failed_value; + } else if (safe_str_neq(rsc_id, failed_value)) { + continue; + } + + // Host name should always have been provided by this point + failed_value = crm_element_value(xml_op, XML_ATTR_UNAME); + if (safe_str_neq(node_name, failed_value)) { + continue; + } + + // No operation specified means all operations match + if (operation) { + failed_value = crm_element_value(xml_op, XML_LRM_ATTR_TASK); + if (safe_str_neq(operation, failed_value)) { + continue; + } + + // Interval (if operation was specified) defaults to 0 (not all) + failed_value = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL); + if (safe_str_neq(interval_ms_str, failed_value)) { + continue; + } + } + + g_hash_table_add(rscs, (gpointer) rsc_id); + } + + g_hash_table_iter_init(&iter, rscs); + while (g_hash_table_iter_next(&iter, (gpointer *) &rsc_id, NULL)) { + crm_debug("Erasing failures of %s on %s", rsc_id, node_name); + rc = clear_rsc_history(crmd_channel, node_name, rsc_id, data_set); + if (rc != pcmk_ok) { + return rc; + } + } + g_hash_table_destroy(rscs); + return rc; +} + +static int +clear_rsc_fail_attrs(resource_t *rsc, const char *operation, + const char *interval, node_t *node) +{ + int rc = pcmk_ok; + int attr_options = attrd_opt_none; + char *rsc_name = rsc_fail_name(rsc); + + if (is_remote_node(node)) { + attr_options |= attrd_opt_remote; + } + rc = attrd_clear_delegate(NULL, node->details->uname, rsc_name, operation, + interval, NULL, attr_options); + free(rsc_name); + return rc; +} + int cli_resource_delete(crm_ipc_t *crmd_channel, const char *host_uname, resource_t *rsc, const char *operation, - const char *interval, pe_working_set_t *data_set) + const char *interval, bool just_failures, + pe_working_set_t *data_set) { int rc = pcmk_ok; node_t *node = NULL; - char *rsc_name = NULL; - int attr_options = attrd_opt_none; if (rsc == NULL) { return -ENXIO; @@ -552,8 +666,8 @@ cli_resource_delete(crm_ipc_t *crmd_channel, const char *host_uname, resource_t *child = (resource_t *) lpc->data; rc = cli_resource_delete(crmd_channel, host_uname, child, operation, - interval, data_set); - if(rc != pcmk_ok) { + interval, just_failures, data_set); + if (rc != pcmk_ok) { return rc; } } @@ -585,8 +699,13 @@ cli_resource_delete(crm_ipc_t *crmd_channel, const char *host_uname, node = (node_t *) lpc->data; if (node->details->online) { - cli_resource_delete(crmd_channel, node->details->uname, rsc, - operation, interval, data_set); + rc = cli_resource_delete(crmd_channel, node->details->uname, + rsc, operation, interval, + just_failures, data_set); + } + if (rc != pcmk_ok) { + g_list_free(nodes); + return rc; } } @@ -611,102 +730,91 @@ cli_resource_delete(crm_ipc_t *crmd_channel, const char *host_uname, if (crmd_channel == NULL) { printf("Dry run: skipping clean-up of %s on %s due to CIB_file\n", rsc->id, host_uname); - return rc; - } + return pcmk_ok; + } - /* Erase the resource's entire LRM history in the CIB, even if we're only - * clearing a single operation's fail count. If we erased only entries for a - * single operation, we might wind up with a wrong idea of the current - * resource state, and we might not re-probe the resource. - */ - rc = send_lrm_rsc_op(crmd_channel, CRM_OP_LRM_DELETE, host_uname, rsc->id, - TRUE, data_set); + rc = clear_rsc_fail_attrs(rsc, operation, interval, node); if (rc != pcmk_ok) { - printf("Unable to clean up %s history on %s: %s\n", - rsc->id, host_uname, pcmk_strerror(rc)); + printf("Unable to clean up %s failures on %s: %s\n", + rsc->id, host_uname, pcmk_strerror(rc)); return rc; } - crmd_replies_needed++; - crm_trace("Processing %d mainloop inputs", crmd_replies_needed); - while(g_main_context_iteration(NULL, FALSE)) { - crm_trace("Processed mainloop input, %d still remaining", - crmd_replies_needed); - } - - if(crmd_replies_needed < 0) { - crmd_replies_needed = 0; - } - - rsc_name = rsc_fail_name(rsc); - if (is_remote_node(node)) { - attr_options |= attrd_opt_remote; + if (just_failures) { + rc = clear_rsc_failures(crmd_channel, host_uname, rsc->id, operation, + interval, data_set); + } else { + rc = clear_rsc_history(crmd_channel, host_uname, rsc->id, data_set); } - rc = attrd_clear_delegate(NULL, host_uname, rsc_name, operation, interval, - NULL, attr_options); if (rc != pcmk_ok) { - printf("Cleaned %s history on %s, but unable to clear failures: %s\n", + printf("Cleaned %s failures on %s, but unable to clean history: %s\n", rsc->id, host_uname, pcmk_strerror(rc)); } else { printf("Cleaned up %s on %s\n", rsc->id, host_uname); } - free(rsc_name); - return rc; } int -cli_resource_delete_failures(crm_ipc_t *crmd_channel, const char *host_uname, - resource_t *rsc, const char *operation, - const char *interval, pe_working_set_t *data_set) +cli_cleanup_all(crm_ipc_t *crmd_channel, const char *node_name, + const char *operation, const char *interval, + pe_working_set_t *data_set) { + int attr_options = attrd_opt_none; int rc = pcmk_ok; + const char *display_name = node_name? node_name : "all nodes"; - if (rsc == NULL) { - return -ENXIO; - - } else if (rsc->children) { - GListPtr lpc = NULL; + if (crmd_channel == NULL) { + printf("Dry run: skipping clean-up of %s due to CIB_file\n", + display_name); + return pcmk_ok; + } + crmd_replies_needed = 0; - for (lpc = rsc->children; lpc != NULL; lpc = lpc->next) { - resource_t *child = (resource_t *) lpc->data; + if (node_name) { + node_t *node = pe_find_node(data_set->nodes, node_name); - rc = cli_resource_delete_failures(crmd_channel, host_uname, child, operation, - interval, data_set); - if(rc != pcmk_ok) { - return rc; - } + if (node == NULL) { + CMD_ERR("Unknown node: %s", node_name); + return -ENXIO; + } + if (is_remote_node(node)) { + attr_options |= attrd_opt_remote; } - return pcmk_ok; } - for (xmlNode *xml_op = __xml_first_child(data_set->failed); xml_op != NULL; - xml_op = __xml_next(xml_op)) { - - const char *node = crm_element_value(xml_op, XML_ATTR_UNAME); - const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); - const char *task_interval = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL); - const char *resource_name = crm_element_value(xml_op, XML_LRM_ATTR_RSCID); + rc = attrd_clear_delegate(NULL, node_name, NULL, operation, interval, + NULL, attr_options); + if (rc != pcmk_ok) { + printf("Unable to clean up all failures on %s: %s\n", + display_name, pcmk_strerror(rc)); + return rc; + } - if(resource_name == NULL) { - continue; - } else if(host_uname && safe_str_neq(host_uname, node)) { - continue; - } else if(rsc->id && safe_str_neq(rsc->id, resource_name)) { - continue; - } else if(operation && safe_str_neq(operation, task)) { - continue; - } else if(interval && safe_str_neq(interval, task_interval)) { - continue; + if (node_name) { + rc = clear_rsc_failures(crmd_channel, node_name, NULL, + operation, interval, data_set); + if (rc != pcmk_ok) { + printf("Cleaned all resource failures on %s, but unable to clean history: %s\n", + node_name, pcmk_strerror(rc)); + return rc; } + } else { + for (GList *iter = data_set->nodes; iter; iter = iter->next) { + pe_node_t *node = (pe_node_t *) iter->data; - crm_debug("Erasing %s failure for %s (%s detected) on %s", - task, rsc->id, resource_name, node); - rc = cli_resource_delete(crmd_channel, node, rsc, task, - task_interval, data_set); + rc = clear_rsc_failures(crmd_channel, node->details->uname, NULL, + operation, interval, data_set); + if (rc != pcmk_ok) { + printf("Cleaned all resource failures on all nodes, but unable to clean history on %s: %s\n", + node->details->uname, pcmk_strerror(rc)); + return rc; + } + } } - return rc; + printf("Cleaned up all resources on %s\n", display_name); + return pcmk_ok; } void -- 1.8.3.1