diff --git a/daemons/controld/controld_throttle.c b/daemons/controld/controld_throttle.c index eb45696761..f56d4deb08 100644 --- a/daemons/controld/controld_throttle.c +++ b/daemons/controld/controld_throttle.c @@ -1,534 +1,540 @@ /* * Copyright 2013-2018 Andrew Beekhof * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include enum throttle_state_e { throttle_extreme = 0x1000, throttle_high = 0x0100, throttle_med = 0x0010, throttle_low = 0x0001, throttle_none = 0x0000, }; struct throttle_record_s { int max; enum throttle_state_e mode; char *node; }; static int throttle_job_max = 0; static float throttle_load_target = 0.0; #define THROTTLE_FACTOR_LOW 1.2 #define THROTTLE_FACTOR_MEDIUM 1.6 #define THROTTLE_FACTOR_HIGH 2.0 static GHashTable *throttle_records = NULL; static mainloop_timer_t *throttle_timer = NULL; #if SUPPORT_PROCFS /*! * \internal * \brief Return name of /proc file containing the CIB daemon's load statistics * * \return Newly allocated memory with file name on success, NULL otherwise * * \note It is the caller's responsibility to free the return value. * This will return NULL if the daemon is being run via valgrind. * This should be called only on Linux systems. */ static char * find_cib_loadfile(void) { int pid = crm_procfs_pid_of("pacemaker-based"); return pid? crm_strdup_printf("/proc/%d/stat", pid) : NULL; } static bool throttle_cib_load(float *load) { /* /proc/[pid]/stat Status information about the process. This is used by ps(1). It is defined in /usr/src/linux/fs/proc/array.c. The fields, in order, with their proper scanf(3) format specifiers, are: pid %d (1) The process ID. comm %s (2) The filename of the executable, in parentheses. This is visible whether or not the executable is swapped out. state %c (3) One character from the string "RSDZTW" where R is running, S is sleeping in an interruptible wait, D is waiting in uninterruptible disk sleep, Z is zombie, T is traced or stopped (on a signal), and W is paging. ppid %d (4) The PID of the parent. pgrp %d (5) The process group ID of the process. session %d (6) The session ID of the process. tty_nr %d (7) The controlling terminal of the process. (The minor device number is contained in the combination of bits 31 to 20 and 7 to 0; the major device number is in bits 15 to 8.) tpgid %d (8) The ID of the foreground process group of the controlling terminal of the process. flags %u (%lu before Linux 2.6.22) (9) The kernel flags word of the process. For bit meanings, see the PF_* defines in the Linux kernel source file include/linux/sched.h. Details depend on the kernel version. minflt %lu (10) The number of minor faults the process has made which have not required loading a memory page from disk. cminflt %lu (11) The number of minor faults that the process's waited-for children have made. majflt %lu (12) The number of major faults the process has made which have required loading a memory page from disk. cmajflt %lu (13) The number of major faults that the process's waited-for children have made. utime %lu (14) Amount of time that this process has been scheduled in user mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). This includes guest time, guest_time (time spent running a virtual CPU, see below), so that applications that are not aware of the guest time field do not lose that time from their calculations. stime %lu (15) Amount of time that this process has been scheduled in kernel mode, measured in clock ticks (divide by sysconf(_SC_CLK_TCK)). */ static char *loadfile = NULL; static time_t last_call = 0; static long ticks_per_s = 0; static unsigned long last_utime, last_stime; char buffer[64*1024]; FILE *stream = NULL; time_t now = time(NULL); if(load == NULL) { return FALSE; } else { *load = 0.0; } if(loadfile == NULL) { last_call = 0; last_utime = 0; last_stime = 0; loadfile = find_cib_loadfile(); if (loadfile == NULL) { crm_warn("Couldn't find CIB load file"); return FALSE; } ticks_per_s = sysconf(_SC_CLK_TCK); crm_trace("Found %s", loadfile); } stream = fopen(loadfile, "r"); if(stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_strerror(rc), rc); free(loadfile); loadfile = NULL; return FALSE; } if(fgets(buffer, sizeof(buffer), stream)) { char *comm = calloc(1, 256); char state = 0; int rc = 0, pid = 0, ppid = 0, pgrp = 0, session = 0, tty_nr = 0, tpgid = 0; unsigned long flags = 0, minflt = 0, cminflt = 0, majflt = 0, cmajflt = 0, utime = 0, stime = 0; rc = sscanf(buffer, "%d %[^ ] %c %d %d %d %d %d %lu %lu %lu %lu %lu %lu %lu", &pid, comm, &state, &ppid, &pgrp, &session, &tty_nr, &tpgid, &flags, &minflt, &cminflt, &majflt, &cmajflt, &utime, &stime); free(comm); if(rc != 15) { crm_err("Only %d of 15 fields found in %s", rc, loadfile); fclose(stream); return FALSE; } else if(last_call > 0 && last_call < now && last_utime <= utime && last_stime <= stime) { time_t elapsed = now - last_call; unsigned long delta_utime = utime - last_utime; unsigned long delta_stime = stime - last_stime; *load = (delta_utime + delta_stime); /* Cast to a float before division */ *load /= ticks_per_s; *load /= elapsed; crm_debug("cib load: %f (%lu ticks in %lds)", *load, delta_utime + delta_stime, (long)elapsed); } else { crm_debug("Init %lu + %lu ticks at %ld (%lu tps)", utime, stime, (long)now, ticks_per_s); } last_call = now; last_utime = utime; last_stime = stime; fclose(stream); return TRUE; } fclose(stream); return FALSE; } static bool throttle_load_avg(float *load) { char buffer[256]; FILE *stream = NULL; const char *loadfile = "/proc/loadavg"; if(load == NULL) { return FALSE; } stream = fopen(loadfile, "r"); if(stream == NULL) { int rc = errno; crm_warn("Couldn't read %s: %s (%d)", loadfile, pcmk_strerror(rc), rc); return FALSE; } if(fgets(buffer, sizeof(buffer), stream)) { char *nl = strstr(buffer, "\n"); /* Grab the 1-minute average, ignore the rest */ *load = strtof(buffer, NULL); if(nl) { nl[0] = 0; } fclose(stream); return TRUE; } fclose(stream); return FALSE; } /*! * \internal * \brief Check a load value against throttling thresholds * * \param[in] load Load value to check * \param[in] desc Description of metric (for logging) * \param[in] thresholds Low/medium/high/extreme thresholds * * \return Throttle mode corresponding to load value */ static enum throttle_state_e throttle_check_thresholds(float load, const char *desc, float thresholds[4]) { if (load > thresholds[3]) { crm_notice("Extreme %s detected: %f", desc, load); return throttle_extreme; } else if (load > thresholds[2]) { crm_notice("High %s detected: %f", desc, load); return throttle_high; } else if (load > thresholds[1]) { crm_info("Moderate %s detected: %f", desc, load); return throttle_med; } else if (load > thresholds[0]) { crm_debug("Noticeable %s detected: %f", desc, load); return throttle_low; } crm_trace("Negligible %s detected: %f", desc, load); return throttle_none; } static enum throttle_state_e throttle_handle_load(float load, const char *desc, int cores) { float normalize; float thresholds[4]; if (cores == 1) { /* On a single core machine, a load of 1.0 is already too high */ normalize = 0.6; } else { /* Normalize the load to be per-core */ normalize = cores; } thresholds[0] = throttle_load_target * normalize * THROTTLE_FACTOR_LOW; thresholds[1] = throttle_load_target * normalize * THROTTLE_FACTOR_MEDIUM; thresholds[2] = throttle_load_target * normalize * THROTTLE_FACTOR_HIGH; thresholds[3] = load + 1.0; /* never extreme */ return throttle_check_thresholds(load, desc, thresholds); } #endif static enum throttle_state_e throttle_mode(void) { #if SUPPORT_PROCFS unsigned int cores; float load; float thresholds[4]; enum throttle_state_e mode = throttle_none; cores = crm_procfs_num_cores(); if(throttle_cib_load(&load)) { float cib_max_cpu = 0.95; /* The CIB is a single-threaded task and thus cannot consume * more than 100% of a CPU (and 1/cores of the overall system * load). * * On a many-cored system, the CIB might therefore be maxed out * (causing operations to fail or appear to fail) even though * the overall system load is still reasonable. * * Therefore, the 'normal' thresholds can not apply here, and we * need a special case. */ if(cores == 1) { cib_max_cpu = 0.4; } if(throttle_load_target > 0.0 && throttle_load_target < cib_max_cpu) { cib_max_cpu = throttle_load_target; } thresholds[0] = cib_max_cpu * 0.8; thresholds[1] = cib_max_cpu * 0.9; thresholds[2] = cib_max_cpu; /* Can only happen on machines with a low number of cores */ thresholds[3] = cib_max_cpu * 1.5; mode |= throttle_check_thresholds(load, "CIB load", thresholds); } if(throttle_load_target <= 0) { /* If we ever make this a valid value, the cluster will at least behave as expected */ return mode; } if(throttle_load_avg(&load)) { crm_debug("Current load is %f across %u core(s)", load, cores); mode |= throttle_handle_load(load, "CPU load", cores); } if(mode & throttle_extreme) { return throttle_extreme; } else if(mode & throttle_high) { return throttle_high; } else if(mode & throttle_med) { return throttle_med; } else if(mode & throttle_low) { return throttle_low; } #endif // SUPPORT_PROCFS return throttle_none; } static void throttle_send_command(enum throttle_state_e mode) { xmlNode *xml = NULL; static enum throttle_state_e last = -1; if(mode != last) { crm_info("New throttle mode: %.4x (was %.4x)", mode, last); last = mode; xml = create_request(CRM_OP_THROTTLE, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL); crm_xml_add_int(xml, F_CRM_THROTTLE_MODE, mode); crm_xml_add_int(xml, F_CRM_THROTTLE_MAX, throttle_job_max); send_cluster_message(NULL, crm_msg_crmd, xml, TRUE); free_xml(xml); } } static gboolean throttle_timer_cb(gpointer data) { throttle_send_command(throttle_mode()); return TRUE; } static void throttle_record_free(gpointer p) { struct throttle_record_s *r = p; free(r->node); free(r); } void throttle_set_load_target(float target) { throttle_load_target = target; } void throttle_update_job_max(const char *preference) { int max = 0; throttle_job_max = 2 * crm_procfs_num_cores(); if(preference) { /* Global preference from the CIB */ max = crm_int_helper(preference, NULL); if(max > 0) { throttle_job_max = max; } } preference = getenv("PCMK_node_action_limit"); if(preference) { /* Per-node override */ max = crm_int_helper(preference, NULL); if(max > 0) { throttle_job_max = max; } } } void throttle_init(void) { if(throttle_records == NULL) { throttle_records = g_hash_table_new_full( crm_str_hash, g_str_equal, NULL, throttle_record_free); throttle_timer = mainloop_timer_add("throttle", 30 * 1000, TRUE, throttle_timer_cb, NULL); } throttle_update_job_max(NULL); mainloop_timer_start(throttle_timer); } void throttle_fini(void) { - mainloop_timer_del(throttle_timer); throttle_timer = NULL; - g_hash_table_destroy(throttle_records); throttle_records = NULL; + if (throttle_timer != NULL) { + mainloop_timer_del(throttle_timer); + throttle_timer = NULL; + } + if (throttle_records != NULL) { + g_hash_table_destroy(throttle_records); + throttle_records = NULL; + } } int throttle_get_total_job_limit(int l) { /* Cluster-wide limit */ GHashTableIter iter; int limit = l; int peers = crm_active_peers(); struct throttle_record_s *r = NULL; g_hash_table_iter_init(&iter, throttle_records); while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &r)) { switch(r->mode) { case throttle_extreme: if(limit == 0 || limit > peers/4) { limit = QB_MAX(1, peers/4); } break; case throttle_high: if(limit == 0 || limit > peers/2) { limit = QB_MAX(1, peers/2); } break; default: break; } } if(limit == l) { /* crm_trace("No change to batch-limit=%d", limit); */ } else if(l == 0) { crm_trace("Using batch-limit=%d", limit); } else { crm_trace("Using batch-limit=%d instead of %d", limit, l); } return limit; } int throttle_get_job_limit(const char *node) { int jobs = 1; struct throttle_record_s *r = NULL; r = g_hash_table_lookup(throttle_records, node); if(r == NULL) { r = calloc(1, sizeof(struct throttle_record_s)); r->node = strdup(node); r->mode = throttle_low; r->max = throttle_job_max; crm_trace("Defaulting to local values for unknown node %s", node); g_hash_table_insert(throttle_records, r->node, r); } switch(r->mode) { case throttle_extreme: case throttle_high: jobs = 1; /* At least one job must always be allowed */ break; case throttle_med: jobs = QB_MAX(1, r->max / 4); break; case throttle_low: jobs = QB_MAX(1, r->max / 2); break; case throttle_none: jobs = QB_MAX(1, r->max); break; default: crm_err("Unknown throttle mode %.4x on %s", r->mode, node); break; } return jobs; } void throttle_update(xmlNode *xml) { int max = 0; enum throttle_state_e mode = 0; struct throttle_record_s *r = NULL; const char *from = crm_element_value(xml, F_CRM_HOST_FROM); crm_element_value_int(xml, F_CRM_THROTTLE_MODE, (int*)&mode); crm_element_value_int(xml, F_CRM_THROTTLE_MAX, &max); r = g_hash_table_lookup(throttle_records, from); if(r == NULL) { r = calloc(1, sizeof(struct throttle_record_s)); r->node = strdup(from); g_hash_table_insert(throttle_records, r->node, r); } r->max = max; r->mode = mode; crm_debug("Host %s supports a maximum of %d jobs and throttle mode %.4x. New job limit is %d", from, max, mode, throttle_get_job_limit(from)); } diff --git a/lib/pengine/container.c b/lib/pengine/container.c index 5eab244f9d..2fe4a28544 100644 --- a/lib/pengine/container.c +++ b/lib/pengine/container.c @@ -1,1464 +1,1466 @@ /* * Copyright 2004-2018 Andrew Beekhof * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #define VARIANT_CONTAINER 1 #include "./variant.h" void tuple_free(container_grouping_t *tuple); static char * next_ip(const char *last_ip) { unsigned int oct1 = 0; unsigned int oct2 = 0; unsigned int oct3 = 0; unsigned int oct4 = 0; int rc = sscanf(last_ip, "%u.%u.%u.%u", &oct1, &oct2, &oct3, &oct4); if (rc != 4) { /*@ TODO check for IPv6 */ return NULL; } else if (oct3 > 253) { return NULL; } else if (oct4 > 253) { ++oct3; oct4 = 1; } else { ++oct4; } return crm_strdup_printf("%u.%u.%u.%u", oct1, oct2, oct3, oct4); } static int allocate_ip(container_variant_data_t *data, container_grouping_t *tuple, char *buffer, int max) { if(data->ip_range_start == NULL) { return 0; } else if(data->ip_last) { tuple->ipaddr = next_ip(data->ip_last); } else { tuple->ipaddr = strdup(data->ip_range_start); } data->ip_last = tuple->ipaddr; #if 0 return snprintf(buffer, max, " --add-host=%s-%d:%s --link %s-docker-%d:%s-link-%d", data->prefix, tuple->offset, tuple->ipaddr, data->prefix, tuple->offset, data->prefix, tuple->offset); #else if (data->type == PE_CONTAINER_TYPE_DOCKER) { if (data->add_host == FALSE) { return 0; } return snprintf(buffer, max, " --add-host=%s-%d:%s", data->prefix, tuple->offset, tuple->ipaddr); } else if (data->type == PE_CONTAINER_TYPE_RKT) { return snprintf(buffer, max, " --hosts-entry=%s=%s-%d", tuple->ipaddr, data->prefix, tuple->offset); } else { return 0; } #endif } static xmlNode * create_resource(const char *name, const char *provider, const char *kind) { xmlNode *rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); crm_xml_add(rsc, XML_ATTR_ID, name); crm_xml_add(rsc, XML_AGENT_ATTR_CLASS, PCMK_RESOURCE_CLASS_OCF); crm_xml_add(rsc, XML_AGENT_ATTR_PROVIDER, provider); crm_xml_add(rsc, XML_ATTR_TYPE, kind); return rsc; } /*! * \internal * \brief Check whether cluster can manage resource inside container * * \param[in] data Container variant data * * \return TRUE if networking configuration is acceptable, FALSE otherwise * * \note The resource is manageable if an IP range or control port has been * specified. If a control port is used without an IP range, replicas per * host must be 1. */ static bool valid_network(container_variant_data_t *data) { if(data->ip_range_start) { return TRUE; } if(data->control_port) { if(data->replicas_per_host > 1) { pe_err("Specifying the 'control-port' for %s requires 'replicas-per-host=1'", data->prefix); data->replicas_per_host = 1; /* @TODO to be sure: clear_bit(rsc->flags, pe_rsc_unique); */ } return TRUE; } return FALSE; } static bool create_ip_resource( resource_t *parent, container_variant_data_t *data, container_grouping_t *tuple, pe_working_set_t * data_set) { if(data->ip_range_start) { char *id = NULL; xmlNode *xml_ip = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-ip-%s", data->prefix, tuple->ipaddr); crm_xml_sanitize_id(id); xml_ip = create_resource(id, "heartbeat", "IPaddr2"); free(id); xml_obj = create_xml_node(xml_ip, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, tuple->offset); crm_create_nvpair_xml(xml_obj, NULL, "ip", tuple->ipaddr); if(data->host_network) { crm_create_nvpair_xml(xml_obj, NULL, "nic", data->host_network); } if(data->host_netmask) { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", data->host_netmask); } else { crm_create_nvpair_xml(xml_obj, NULL, "cidr_netmask", "32"); } xml_obj = create_xml_node(xml_ip, "operations"); crm_create_op_xml(xml_obj, ID(xml_ip), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? crm_log_xml_trace(xml_ip, "Container-ip"); if (common_unpack(xml_ip, &tuple->ip, parent, data_set) == false) { return FALSE; } parent->children = g_list_append(parent->children, tuple->ip); } return TRUE; } static bool create_docker_resource( resource_t *parent, container_variant_data_t *data, container_grouping_t *tuple, pe_working_set_t * data_set) { int offset = 0, max = 4096; char *buffer = calloc(1, max+1); int doffset = 0, dmax = 1024; char *dbuffer = calloc(1, dmax+1); char *id = NULL; xmlNode *xml_docker = NULL; xmlNode *xml_obj = NULL; id = crm_strdup_printf("%s-docker-%d", data->prefix, tuple->offset); crm_xml_sanitize_id(id); xml_docker = create_resource(id, "heartbeat", "docker"); free(id); xml_obj = create_xml_node(xml_docker, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, tuple->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", XML_BOOLEAN_TRUE); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", XML_BOOLEAN_FALSE); crm_create_nvpair_xml(xml_obj, NULL, "reuse", XML_BOOLEAN_FALSE); offset += snprintf(buffer+offset, max-offset, " --restart=no"); /* Set a container hostname only if we have an IP to map it to. * The user can set -h or --uts=host themselves if they want a nicer * name for logs, but this makes applications happy who need their * hostname to match the IP they bind to. */ if (data->ip_range_start != NULL) { offset += snprintf(buffer+offset, max-offset, " -h %s-%d", data->prefix, tuple->offset); } offset += snprintf(buffer+offset, max-offset, " -e PCMK_stderr=1"); if(data->docker_network) { // offset += snprintf(buffer+offset, max-offset, " --link-local-ip=%s", tuple->ipaddr); offset += snprintf(buffer+offset, max-offset, " --net=%s", data->docker_network); } if(data->control_port) { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%s", data->control_port); } else { offset += snprintf(buffer+offset, max-offset, " -e PCMK_remote_port=%d", DEFAULT_REMOTE_PORT); } for(GListPtr pIter = data->mounts; pIter != NULL; pIter = pIter->next) { container_mount_t *mount = pIter->data; if(mount->flags) { char *source = crm_strdup_printf( "%s/%s-%d", mount->source, data->prefix, tuple->offset); if(doffset > 0) { doffset += snprintf(dbuffer+doffset, dmax-doffset, ","); } doffset += snprintf(dbuffer+doffset, dmax-doffset, "%s", source); offset += snprintf(buffer+offset, max-offset, " -v %s:%s", source, mount->target); free(source); } else { offset += snprintf(buffer+offset, max-offset, " -v %s:%s", mount->source, mount->target); } if(mount->options) { offset += snprintf(buffer+offset, max-offset, ":%s", mount->options); } } for(GListPtr pIter = data->ports; pIter != NULL; pIter = pIter->next) { container_port_t *port = pIter->data; if(tuple->ipaddr) { offset += snprintf(buffer+offset, max-offset, " -p %s:%s:%s", tuple->ipaddr, port->source, port->target); } else if(safe_str_neq(data->docker_network, "host")) { // No need to do port mapping if net=host offset += snprintf(buffer+offset, max-offset, " -p %s:%s", port->source, port->target); } } if(data->docker_run_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->docker_run_options); } if(data->docker_host_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->docker_host_options); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", buffer); free(buffer); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", dbuffer); free(dbuffer); if(tuple->child) { if(data->docker_run_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->docker_run_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive, we'll * monitor the child independently */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); /* } else if(child && data->untrusted) { * Support this use-case? * * The ability to have resources started/stopped by us, but * unable to set attributes, etc. * * Arguably better to control API access this with ACLs like * "normal" remote nodes * * crm_create_nvpair_xml(xml_obj, NULL, * "run_cmd", * "/usr/libexec/pacemaker/pacemaker-execd"); * crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", * "/usr/libexec/pacemaker/lrmd_internal_ctl -c poke"); */ } else { if(data->docker_run_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->docker_run_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want * to know if it is alive */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_docker, "operations"); crm_create_op_xml(xml_obj, ID(xml_docker), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? crm_log_xml_trace(xml_docker, "Container-docker"); if (common_unpack(xml_docker, &tuple->docker, parent, data_set) == FALSE) { return FALSE; } parent->children = g_list_append(parent->children, tuple->docker); return TRUE; } static bool create_rkt_resource( resource_t *parent, container_variant_data_t *data, container_grouping_t *tuple, pe_working_set_t * data_set) { int offset = 0, max = 4096; char *buffer = calloc(1, max+1); int doffset = 0, dmax = 1024; char *dbuffer = calloc(1, dmax+1); char *id = NULL; xmlNode *xml_docker = NULL; xmlNode *xml_obj = NULL; int volid = 0; id = crm_strdup_printf("%s-rkt-%d", data->prefix, tuple->offset); crm_xml_sanitize_id(id); xml_docker = create_resource(id, "heartbeat", "rkt"); free(id); xml_obj = create_xml_node(xml_docker, XML_TAG_ATTR_SETS); crm_xml_set_id(xml_obj, "%s-attributes-%d", data->prefix, tuple->offset); crm_create_nvpair_xml(xml_obj, NULL, "image", data->image); crm_create_nvpair_xml(xml_obj, NULL, "allow_pull", "true"); crm_create_nvpair_xml(xml_obj, NULL, "force_kill", "false"); crm_create_nvpair_xml(xml_obj, NULL, "reuse", "false"); /* Set a container hostname only if we have an IP to map it to. * The user can set -h or --uts=host themselves if they want a nicer * name for logs, but this makes applications happy who need their * hostname to match the IP they bind to. */ if (data->ip_range_start != NULL) { offset += snprintf(buffer+offset, max-offset, " --hostname=%s-%d", data->prefix, tuple->offset); } offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_stderr=1"); if(data->docker_network) { // offset += snprintf(buffer+offset, max-offset, " --link-local-ip=%s", tuple->ipaddr); offset += snprintf(buffer+offset, max-offset, " --net=%s", data->docker_network); } if(data->control_port) { offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_remote_port=%s", data->control_port); } else { offset += snprintf(buffer+offset, max-offset, " --environment=PCMK_remote_port=%d", DEFAULT_REMOTE_PORT); } for(GListPtr pIter = data->mounts; pIter != NULL; pIter = pIter->next) { container_mount_t *mount = pIter->data; if(mount->flags) { char *source = crm_strdup_printf( "%s/%s-%d", mount->source, data->prefix, tuple->offset); if(doffset > 0) { doffset += snprintf(dbuffer+doffset, dmax-doffset, ","); } doffset += snprintf(dbuffer+doffset, dmax-doffset, "%s", source); offset += snprintf(buffer+offset, max-offset, " --volume vol%d,kind=host,source=%s", volid, source); if(mount->options) { offset += snprintf(buffer+offset, max-offset, ",%s", mount->options); } offset += snprintf(buffer+offset, max-offset, " --mount volume=vol%d,target=%s", volid, mount->target); free(source); } else { offset += snprintf(buffer+offset, max-offset, " --volume vol%d,kind=host,source=%s", volid, mount->source); if(mount->options) { offset += snprintf(buffer+offset, max-offset, ",%s", mount->options); } offset += snprintf(buffer+offset, max-offset, " --mount volume=vol%d,target=%s", volid, mount->target); } volid++; } for(GListPtr pIter = data->ports; pIter != NULL; pIter = pIter->next) { container_port_t *port = pIter->data; if(tuple->ipaddr) { offset += snprintf(buffer+offset, max-offset, " --port=%s:%s:%s", port->target, tuple->ipaddr, port->source); } else { offset += snprintf(buffer+offset, max-offset, " --port=%s:%s", port->target, port->source); } } if(data->docker_run_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->docker_run_options); } if(data->docker_host_options) { offset += snprintf(buffer+offset, max-offset, " %s", data->docker_host_options); } crm_create_nvpair_xml(xml_obj, NULL, "run_opts", buffer); free(buffer); crm_create_nvpair_xml(xml_obj, NULL, "mount_points", dbuffer); free(dbuffer); if(tuple->child) { if(data->docker_run_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->docker_run_command); } else { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", SBIN_DIR "/pacemaker-remoted"); } /* TODO: Allow users to specify their own? * * We just want to know if the container is alive, we'll * monitor the child independently */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); /* } else if(child && data->untrusted) { * Support this use-case? * * The ability to have resources started/stopped by us, but * unable to set attributes, etc. * * Arguably better to control API access this with ACLs like * "normal" remote nodes * * crm_create_nvpair_xml(xml_obj, NULL, * "run_cmd", * "/usr/libexec/pacemaker/pacemaker-execd"); * crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", * "/usr/libexec/pacemaker/lrmd_internal_ctl -c poke"); */ } else { if(data->docker_run_command) { crm_create_nvpair_xml(xml_obj, NULL, "run_cmd", data->docker_run_command); } /* TODO: Allow users to specify their own? * * We don't know what's in the container, so we just want * to know if it is alive */ crm_create_nvpair_xml(xml_obj, NULL, "monitor_cmd", "/bin/true"); } xml_obj = create_xml_node(xml_docker, "operations"); crm_create_op_xml(xml_obj, ID(xml_docker), "monitor", "60s", NULL); // TODO: Other ops? Timeouts and intervals from underlying resource? crm_log_xml_trace(xml_docker, "Container-rkt"); if (common_unpack(xml_docker, &tuple->docker, parent, data_set) == FALSE) { return FALSE; } parent->children = g_list_append(parent->children, tuple->docker); return TRUE; } /*! * \brief Ban a node from a resource's (and its children's) allowed nodes list * * \param[in,out] rsc Resource to modify * \param[in] uname Name of node to ban */ static void disallow_node(resource_t *rsc, const char *uname) { gpointer match = g_hash_table_lookup(rsc->allowed_nodes, uname); if (match) { ((pe_node_t *) match)->weight = -INFINITY; ((pe_node_t *) match)->rsc_discover_mode = pe_discover_never; } if (rsc->children) { GListPtr child; for (child = rsc->children; child != NULL; child = child->next) { disallow_node((resource_t *) (child->data), uname); } } } static bool create_remote_resource( resource_t *parent, container_variant_data_t *data, container_grouping_t *tuple, pe_working_set_t * data_set) { if (tuple->child && valid_network(data)) { GHashTableIter gIter; GListPtr rsc_iter = NULL; node_t *node = NULL; xmlNode *xml_remote = NULL; char *id = crm_strdup_printf("%s-%d", data->prefix, tuple->offset); char *port_s = NULL; const char *uname = NULL; const char *connect_name = NULL; if (remote_id_conflict(id, data_set)) { free(id); // The biggest hammer we have id = crm_strdup_printf("pcmk-internal-%s-remote-%d", tuple->child->id, tuple->offset); CRM_ASSERT(remote_id_conflict(id, data_set) == FALSE); } /* REMOTE_CONTAINER_HACK: Using "#uname" as the server name when the * connection does not have its own IP is a magic string that we use to * support nested remotes (i.e. a bundle running on a remote node). */ connect_name = (tuple->ipaddr? tuple->ipaddr : "#uname"); if (data->control_port == NULL) { port_s = crm_itoa(DEFAULT_REMOTE_PORT); } /* This sets tuple->docker as tuple->remote's container, which is * similar to what happens with guest nodes. This is how the PE knows * that the bundle node is fenced by recovering docker, and that * remote should be ordered relative to docker. */ xml_remote = pe_create_remote_xml(NULL, id, tuple->docker->id, NULL, NULL, NULL, connect_name, (data->control_port? data->control_port : port_s)); free(port_s); /* Abandon our created ID, and pull the copy from the XML, because we * need something that will get freed during data set cleanup to use as * the node ID and uname. */ free(id); id = NULL; uname = ID(xml_remote); /* Ensure a node has been created for the guest (it may have already * been, if it has a permanent node attribute), and ensure its weight is * -INFINITY so no other resources can run on it. */ node = pe_find_node(data_set->nodes, uname); if (node == NULL) { node = pe_create_node(uname, uname, "remote", "-INFINITY", data_set); } else { node->weight = -INFINITY; } node->rsc_discover_mode = pe_discover_never; /* unpack_remote_nodes() ensures that each remote node and guest node * has a pe_node_t entry. Ideally, it would do the same for bundle nodes. * Unfortunately, a bundle has to be mostly unpacked before it's obvious * what nodes will be needed, so we do it just above. * * Worse, that means that the node may have been utilized while * unpacking other resources, without our weight correction. The most * likely place for this to happen is when common_unpack() calls * resource_location() to set a default score in symmetric clusters. * This adds a node *copy* to each resource's allowed nodes, and these * copies will have the wrong weight. * * As a hacky workaround, fix those copies here. * * @TODO Possible alternative: ensure bundles are unpacked before other * resources, so the weight is correct before any copies are made. */ for (rsc_iter = data_set->resources; rsc_iter; rsc_iter = rsc_iter->next) { disallow_node((resource_t *) (rsc_iter->data), uname); } tuple->node = node_copy(node); tuple->node->weight = 500; tuple->node->rsc_discover_mode = pe_discover_exclusive; /* Ensure the node shows up as allowed and with the correct discovery set */ - g_hash_table_destroy(tuple->child->allowed_nodes); + if (tuple->child->allowed_nodes != NULL) { + g_hash_table_destroy(tuple->child->allowed_nodes); + } tuple->child->allowed_nodes = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free); g_hash_table_insert(tuple->child->allowed_nodes, (gpointer) tuple->node->details->id, node_copy(tuple->node)); { node_t *copy = node_copy(tuple->node); copy->weight = -INFINITY; g_hash_table_insert(tuple->child->parent->allowed_nodes, (gpointer) tuple->node->details->id, copy); } crm_log_xml_trace(xml_remote, "Container-remote"); if (common_unpack(xml_remote, &tuple->remote, parent, data_set) == FALSE) { return FALSE; } g_hash_table_iter_init(&gIter, tuple->remote->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&node)) { if(is_remote_node(node)) { /* Remote resources can only run on 'normal' cluster node */ node->weight = -INFINITY; } } tuple->node->details->remote_rsc = tuple->remote; tuple->remote->container = tuple->docker; // Ensures is_container_remote_node() functions correctly immediately /* A bundle's #kind is closer to "container" (guest node) than the * "remote" set by pe_create_node(). */ g_hash_table_insert(tuple->node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); /* One effect of this is that setup_container() will add * tuple->remote to tuple->docker's fillers, which will make * rsc_contains_remote_node() true for tuple->docker. * * tuple->child does NOT get added to tuple->docker's fillers. * The only noticeable effect if it did would be for its fail count to * be taken into account when checking tuple->docker's migration * threshold. */ parent->children = g_list_append(parent->children, tuple->remote); } return TRUE; } static bool create_container( resource_t *parent, container_variant_data_t *data, container_grouping_t *tuple, pe_working_set_t * data_set) { if (data->type == PE_CONTAINER_TYPE_DOCKER && create_docker_resource(parent, data, tuple, data_set) == FALSE) { return FALSE; } if (data->type == PE_CONTAINER_TYPE_RKT && create_rkt_resource(parent, data, tuple, data_set) == FALSE) { return FALSE; } if(create_ip_resource(parent, data, tuple, data_set) == FALSE) { return FALSE; } if(create_remote_resource(parent, data, tuple, data_set) == FALSE) { return FALSE; } if(tuple->child && tuple->ipaddr) { add_hash_param(tuple->child->meta, "external-ip", tuple->ipaddr); } if(tuple->remote) { /* * Allow the remote connection resource to be allocated to a * different node than the one on which the docker container * is active. * * This makes it possible to have Pacemaker Remote nodes running * containers with pacemaker-remoted inside in order to start * services inside those containers. */ set_bit(tuple->remote->flags, pe_rsc_allow_remote_remotes); } return TRUE; } static void mount_add(container_variant_data_t *container_data, const char *source, const char *target, const char *options, int flags) { container_mount_t *mount = calloc(1, sizeof(container_mount_t)); mount->source = strdup(source); mount->target = strdup(target); if (options) { mount->options = strdup(options); } mount->flags = flags; container_data->mounts = g_list_append(container_data->mounts, mount); } static void mount_free(container_mount_t *mount) { free(mount->source); free(mount->target); free(mount->options); free(mount); } static void port_free(container_port_t *port) { free(port->source); free(port->target); free(port); } static container_grouping_t * tuple_for_remote(resource_t *remote) { resource_t *top = remote; container_variant_data_t *container_data = NULL; if (top == NULL) { return NULL; } while (top->parent != NULL) { top = top->parent; } get_container_variant_data(container_data, top); for (GListPtr gIter = container_data->tuples; gIter != NULL; gIter = gIter->next) { container_grouping_t *tuple = (container_grouping_t *)gIter->data; if(tuple->remote == remote) { return tuple; } } CRM_LOG_ASSERT(FALSE); return NULL; } bool container_fix_remote_addr(resource_t *rsc) { const char *name; const char *value; const char *attr_list[] = { XML_ATTR_TYPE, XML_AGENT_ATTR_CLASS, XML_AGENT_ATTR_PROVIDER }; const char *value_list[] = { "remote", PCMK_RESOURCE_CLASS_OCF, "pacemaker" }; if(rsc == NULL) { return FALSE; } name = "addr"; value = g_hash_table_lookup(rsc->parameters, name); if (safe_str_eq(value, "#uname") == FALSE) { return FALSE; } for (int lpc = 0; lpc < DIMOF(attr_list); lpc++) { name = attr_list[lpc]; value = crm_element_value(rsc->xml, attr_list[lpc]); if (safe_str_eq(value, value_list[lpc]) == FALSE) { return FALSE; } } return TRUE; } const char * container_fix_remote_addr_in(resource_t *rsc, xmlNode *xml, const char *field) { // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside pe_node_t *node = NULL; container_grouping_t *tuple = NULL; if(container_fix_remote_addr(rsc) == FALSE) { return NULL; } tuple = tuple_for_remote(rsc); if(tuple == NULL) { return NULL; } node = tuple->docker->allocated_to; if(node == NULL && tuple->docker->running_on) { /* If it won't be running anywhere after the * transition, go with where it's running now. */ node = tuple->docker->running_on->data; } if(node == NULL) { crm_trace("Cannot fix address for %s", tuple->remote->id); return NULL; } crm_trace("Fixing addr for %s on %s", rsc->id, node->details->uname); if(xml != NULL && field != NULL) { crm_xml_add(xml, field, node->details->uname); } return node->details->uname; } gboolean container_unpack(resource_t * rsc, pe_working_set_t * data_set) { const char *value = NULL; xmlNode *xml_obj = NULL; xmlNode *xml_resource = NULL; container_variant_data_t *container_data = NULL; CRM_ASSERT(rsc != NULL); pe_rsc_trace(rsc, "Processing resource %s...", rsc->id); container_data = calloc(1, sizeof(container_variant_data_t)); rsc->variant_opaque = container_data; container_data->prefix = strdup(rsc->id); xml_obj = first_named_child(rsc->xml, "docker"); if (xml_obj != NULL) { container_data->type = PE_CONTAINER_TYPE_DOCKER; } else { xml_obj = first_named_child(rsc->xml, "rkt"); if (xml_obj != NULL) { container_data->type = PE_CONTAINER_TYPE_RKT; } else { return FALSE; } } value = crm_element_value(xml_obj, XML_RSC_ATTR_PROMOTED_MAX); if (value == NULL) { // @COMPAT deprecated since 2.0.0 value = crm_element_value(xml_obj, "masters"); } container_data->promoted_max = crm_parse_int(value, "0"); if (container_data->promoted_max < 0) { pe_err("%s for %s must be nonnegative integer, using 0", XML_RSC_ATTR_PROMOTED_MAX, rsc->id); container_data->promoted_max = 0; } value = crm_element_value(xml_obj, "replicas"); if ((value == NULL) && container_data->promoted_max) { container_data->replicas = container_data->promoted_max; } else { container_data->replicas = crm_parse_int(value, "1"); } if (container_data->replicas < 1) { pe_err("'replicas' for %s must be positive integer, using 1", rsc->id); container_data->replicas = 1; } /* * Communication between containers on the same host via the * floating IPs only works if docker is started with: * --userland-proxy=false --ip-masq=false */ value = crm_element_value(xml_obj, "replicas-per-host"); container_data->replicas_per_host = crm_parse_int(value, "1"); if (container_data->replicas_per_host < 1) { pe_err("'replicas-per-host' for %s must be positive integer, using 1", rsc->id); container_data->replicas_per_host = 1; } if (container_data->replicas_per_host == 1) { clear_bit(rsc->flags, pe_rsc_unique); } container_data->docker_run_command = crm_element_value_copy(xml_obj, "run-command"); container_data->docker_run_options = crm_element_value_copy(xml_obj, "options"); container_data->image = crm_element_value_copy(xml_obj, "image"); container_data->docker_network = crm_element_value_copy(xml_obj, "network"); xml_obj = first_named_child(rsc->xml, "network"); if(xml_obj) { container_data->ip_range_start = crm_element_value_copy(xml_obj, "ip-range-start"); container_data->host_netmask = crm_element_value_copy(xml_obj, "host-netmask"); container_data->host_network = crm_element_value_copy(xml_obj, "host-interface"); container_data->control_port = crm_element_value_copy(xml_obj, "control-port"); value = crm_element_value(xml_obj, "add-host"); if (check_boolean(value) == FALSE) { container_data->add_host = TRUE; } else { crm_str_to_boolean(value, &container_data->add_host); } for (xmlNode *xml_child = __xml_first_child_element(xml_obj); xml_child != NULL; xml_child = __xml_next_element(xml_child)) { container_port_t *port = calloc(1, sizeof(container_port_t)); port->source = crm_element_value_copy(xml_child, "port"); if(port->source == NULL) { port->source = crm_element_value_copy(xml_child, "range"); } else { port->target = crm_element_value_copy(xml_child, "internal-port"); } if(port->source != NULL && strlen(port->source) > 0) { if(port->target == NULL) { port->target = strdup(port->source); } container_data->ports = g_list_append(container_data->ports, port); } else { pe_err("Invalid port directive %s", ID(xml_child)); port_free(port); } } } xml_obj = first_named_child(rsc->xml, "storage"); for (xmlNode *xml_child = __xml_first_child_element(xml_obj); xml_child != NULL; xml_child = __xml_next_element(xml_child)) { const char *source = crm_element_value(xml_child, "source-dir"); const char *target = crm_element_value(xml_child, "target-dir"); const char *options = crm_element_value(xml_child, "options"); int flags = 0; if (source == NULL) { source = crm_element_value(xml_child, "source-dir-root"); flags = 1; } if (source && target) { mount_add(container_data, source, target, options, flags); } else { pe_err("Invalid mount directive %s", ID(xml_child)); } } xml_obj = first_named_child(rsc->xml, "primitive"); if (xml_obj && valid_network(container_data)) { char *value = NULL; xmlNode *xml_set = NULL; xml_resource = create_xml_node(NULL, XML_CIB_TAG_INCARNATION); /* @COMPAT We no longer use the tag, but we need to keep it as * part of the resource name, so that bundles don't restart in a rolling * upgrade. (It also avoids needing to change regression tests.) */ crm_xml_set_id(xml_resource, "%s-%s", container_data->prefix, (container_data->promoted_max? "master" : (const char *)xml_resource->name)); xml_set = create_xml_node(xml_resource, XML_TAG_META_SETS); crm_xml_set_id(xml_set, "%s-%s-meta", container_data->prefix, xml_resource->name); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_ORDERED, XML_BOOLEAN_TRUE); value = crm_itoa(container_data->replicas); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_MAX, value); free(value); value = crm_itoa(container_data->replicas_per_host); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_INCARNATION_NODEMAX, value); free(value); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_UNIQUE, (container_data->replicas_per_host > 1)? XML_BOOLEAN_TRUE : XML_BOOLEAN_FALSE); if (container_data->promoted_max) { crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTABLE, XML_BOOLEAN_TRUE); value = crm_itoa(container_data->promoted_max); crm_create_nvpair_xml(xml_set, NULL, XML_RSC_ATTR_PROMOTED_MAX, value); free(value); } //crm_xml_add(xml_obj, XML_ATTR_ID, container_data->prefix); add_node_copy(xml_resource, xml_obj); } else if(xml_obj) { pe_err("Cannot control %s inside %s without either ip-range-start or control-port", rsc->id, ID(xml_obj)); return FALSE; } if(xml_resource) { int lpc = 0; GListPtr childIter = NULL; resource_t *new_rsc = NULL; container_port_t *port = NULL; int offset = 0, max = 1024; char *buffer = NULL; if (common_unpack(xml_resource, &new_rsc, rsc, data_set) == FALSE) { pe_err("Failed unpacking resource %s", ID(rsc->xml)); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } return FALSE; } container_data->child = new_rsc; /* Currently, we always map the default authentication key location * into the same location inside the container. * * Ideally, we would respect the host's PCMK_authkey_location, but: * - it may be different on different nodes; * - the actual connection will do extra checking to make sure the key * file exists and is readable, that we can't do here on the DC * - tools such as crm_resource and crm_simulate may not have the same * environment variables as the cluster, causing operation digests to * differ * * Always using the default location inside the container is fine, * because we control the pacemaker_remote environment, and it avoids * having to pass another environment variable to the container. * * @TODO A better solution may be to have only pacemaker_remote use the * environment variable, and have the cluster nodes use a new * cluster option for key location. This would introduce the limitation * of the location being the same on all cluster nodes, but that's * reasonable. */ mount_add(container_data, DEFAULT_REMOTE_KEY_LOCATION, DEFAULT_REMOTE_KEY_LOCATION, NULL, 0); mount_add(container_data, CRM_BUNDLE_DIR, "/var/log", NULL, 1); port = calloc(1, sizeof(container_port_t)); if(container_data->control_port) { port->source = strdup(container_data->control_port); } else { /* If we wanted to respect PCMK_remote_port, we could use * crm_default_remote_port() here and elsewhere in this file instead * of DEFAULT_REMOTE_PORT. * * However, it gains nothing, since we control both the container * environment and the connection resource parameters, and the user * can use a different port if desired by setting control-port. */ port->source = crm_itoa(DEFAULT_REMOTE_PORT); } port->target = strdup(port->source); container_data->ports = g_list_append(container_data->ports, port); buffer = calloc(1, max+1); for(childIter = container_data->child->children; childIter != NULL; childIter = childIter->next) { container_grouping_t *tuple = calloc(1, sizeof(container_grouping_t)); tuple->child = childIter->data; tuple->child->exclusive_discover = TRUE; tuple->offset = lpc++; // Ensure the child's notify gets set based on the underlying primitive's value if(is_set(tuple->child->flags, pe_rsc_notify)) { set_bit(container_data->child->flags, pe_rsc_notify); } offset += allocate_ip(container_data, tuple, buffer+offset, max-offset); container_data->tuples = g_list_append(container_data->tuples, tuple); container_data->attribute_target = g_hash_table_lookup(tuple->child->meta, XML_RSC_ATTR_TARGET); } container_data->docker_host_options = buffer; if(container_data->attribute_target) { g_hash_table_replace(rsc->meta, strdup(XML_RSC_ATTR_TARGET), strdup(container_data->attribute_target)); g_hash_table_replace(container_data->child->meta, strdup(XML_RSC_ATTR_TARGET), strdup(container_data->attribute_target)); } } else { // Just a naked container, no pacemaker-remote int offset = 0, max = 1024; char *buffer = calloc(1, max+1); for(int lpc = 0; lpc < container_data->replicas; lpc++) { container_grouping_t *tuple = calloc(1, sizeof(container_grouping_t)); tuple->offset = lpc; offset += allocate_ip(container_data, tuple, buffer+offset, max-offset); container_data->tuples = g_list_append(container_data->tuples, tuple); } container_data->docker_host_options = buffer; } for (GListPtr gIter = container_data->tuples; gIter != NULL; gIter = gIter->next) { container_grouping_t *tuple = (container_grouping_t *)gIter->data; if (create_container(rsc, container_data, tuple, data_set) == FALSE) { pe_err("Failed unpacking resource %s", rsc->id); rsc->fns->free(rsc); return FALSE; } } if(container_data->child) { rsc->children = g_list_append(rsc->children, container_data->child); } return TRUE; } static int tuple_rsc_active(resource_t *rsc, gboolean all) { if (rsc) { gboolean child_active = rsc->fns->active(rsc, all); if (child_active && !all) { return TRUE; } else if (!child_active && all) { return FALSE; } } return -1; } gboolean container_active(resource_t * rsc, gboolean all) { container_variant_data_t *container_data = NULL; GListPtr iter = NULL; get_container_variant_data(container_data, rsc); for (iter = container_data->tuples; iter != NULL; iter = iter->next) { container_grouping_t *tuple = (container_grouping_t *)(iter->data); int rsc_active; rsc_active = tuple_rsc_active(tuple->ip, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = tuple_rsc_active(tuple->child, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = tuple_rsc_active(tuple->docker, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } rsc_active = tuple_rsc_active(tuple->remote, all); if (rsc_active >= 0) { return (gboolean) rsc_active; } } /* If "all" is TRUE, we've already checked that no resources were inactive, * so return TRUE; if "all" is FALSE, we didn't find any active resources, * so return FALSE. */ return all; } /*! * \internal * \brief Find the container child corresponding to a given node * * \param[in] bundle Top-level bundle resource * \param[in] node Node to search for * * \return Container child if found, NULL otherwise */ resource_t * find_container_child(const resource_t *bundle, const node_t *node) { container_variant_data_t *container_data = NULL; CRM_ASSERT(bundle && node); get_container_variant_data(container_data, bundle); for (GListPtr gIter = container_data->tuples; gIter != NULL; gIter = gIter->next) { container_grouping_t *tuple = (container_grouping_t *)gIter->data; CRM_ASSERT(tuple && tuple->node); if (tuple->node->details == node->details) { return tuple->child; } } return NULL; } static void print_rsc_in_list(resource_t *rsc, const char *pre_text, long options, void *print_data) { if (rsc != NULL) { if (options & pe_print_html) { status_print("
  • "); } rsc->fns->print(rsc, pre_text, options, print_data); if (options & pe_print_html) { status_print("
  • \n"); } } } static const char* container_type_as_string(enum container_type t) { if (t == PE_CONTAINER_TYPE_DOCKER) { return PE_CONTAINER_TYPE_DOCKER_S; } else if (t == PE_CONTAINER_TYPE_RKT) { return PE_CONTAINER_TYPE_RKT_S; } else { return PE_CONTAINER_TYPE_UNKNOWN_S; } } static void container_print_xml(resource_t * rsc, const char *pre_text, long options, void *print_data) { container_variant_data_t *container_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (pre_text == NULL) { pre_text = ""; } child_text = crm_concat(pre_text, " ", ' '); get_container_variant_data(container_data, rsc); status_print("%sid); // Always lowercase the container technology type for use as XML value status_print("type=\""); for (const char *c = container_type_as_string(container_data->type); *c; ++c) { status_print("%c", tolower(*c)); } status_print("\" "); status_print("image=\"%s\" ", container_data->image); status_print("unique=\"%s\" ", is_set(rsc->flags, pe_rsc_unique)? "true" : "false"); status_print("managed=\"%s\" ", is_set(rsc->flags, pe_rsc_managed) ? "true" : "false"); status_print("failed=\"%s\" ", is_set(rsc->flags, pe_rsc_failed) ? "true" : "false"); status_print(">\n"); for (GListPtr gIter = container_data->tuples; gIter != NULL; gIter = gIter->next) { container_grouping_t *tuple = (container_grouping_t *)gIter->data; CRM_ASSERT(tuple); status_print("%s \n", pre_text, tuple->offset); print_rsc_in_list(tuple->ip, child_text, options, print_data); print_rsc_in_list(tuple->child, child_text, options, print_data); print_rsc_in_list(tuple->docker, child_text, options, print_data); print_rsc_in_list(tuple->remote, child_text, options, print_data); status_print("%s \n", pre_text); } status_print("%s\n", pre_text); free(child_text); } static void tuple_print(container_grouping_t * tuple, const char *pre_text, long options, void *print_data) { node_t *node = NULL; resource_t *rsc = tuple->child; int offset = 0; char buffer[LINE_MAX]; if(rsc == NULL) { rsc = tuple->docker; } if(tuple->remote) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(tuple->remote)); } else { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_printable_id(tuple->docker)); } if(tuple->ipaddr) { offset += snprintf(buffer + offset, LINE_MAX - offset, " (%s)", tuple->ipaddr); } if (tuple->docker->running_on) { node = tuple->docker->running_on->data; } common_print(rsc, pre_text, buffer, node, options, print_data); } void container_print(resource_t * rsc, const char *pre_text, long options, void *print_data) { container_variant_data_t *container_data = NULL; char *child_text = NULL; CRM_CHECK(rsc != NULL, return); if (options & pe_print_xml) { container_print_xml(rsc, pre_text, options, print_data); return; } get_container_variant_data(container_data, rsc); if (pre_text == NULL) { pre_text = " "; } status_print("%s%s container%s: %s [%s]%s%s\n", pre_text, container_type_as_string(container_data->type), container_data->replicas>1?" set":"", rsc->id, container_data->image, is_set(rsc->flags, pe_rsc_unique) ? " (unique)" : "", is_set(rsc->flags, pe_rsc_managed) ? "" : " (unmanaged)"); if (options & pe_print_html) { status_print("
    \n
      \n"); } for (GListPtr gIter = container_data->tuples; gIter != NULL; gIter = gIter->next) { container_grouping_t *tuple = (container_grouping_t *)gIter->data; CRM_ASSERT(tuple); if (options & pe_print_html) { status_print("
    • "); } if (is_set(options, pe_print_implicit)) { child_text = crm_strdup_printf(" %s", pre_text); if(g_list_length(container_data->tuples) > 1) { status_print(" %sReplica[%d]\n", pre_text, tuple->offset); } if (options & pe_print_html) { status_print("
      \n
        \n"); } print_rsc_in_list(tuple->ip, child_text, options, print_data); print_rsc_in_list(tuple->docker, child_text, options, print_data); print_rsc_in_list(tuple->remote, child_text, options, print_data); print_rsc_in_list(tuple->child, child_text, options, print_data); if (options & pe_print_html) { status_print("
      \n"); } } else { child_text = crm_strdup_printf("%s ", pre_text); tuple_print(tuple, child_text, options, print_data); } free(child_text); if (options & pe_print_html) { status_print("
    • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } } void tuple_free(container_grouping_t *tuple) { if(tuple == NULL) { return; } if(tuple->node) { free(tuple->node); tuple->node = NULL; } if(tuple->ip) { free_xml(tuple->ip->xml); tuple->ip->xml = NULL; tuple->ip->fns->free(tuple->ip); tuple->ip = NULL; } if(tuple->docker) { free_xml(tuple->docker->xml); tuple->docker->xml = NULL; tuple->docker->fns->free(tuple->docker); tuple->docker = NULL; } if(tuple->remote) { free_xml(tuple->remote->xml); tuple->remote->xml = NULL; tuple->remote->fns->free(tuple->remote); tuple->remote = NULL; } free(tuple->ipaddr); free(tuple); } void container_free(resource_t * rsc) { container_variant_data_t *container_data = NULL; CRM_CHECK(rsc != NULL, return); get_container_variant_data(container_data, rsc); pe_rsc_trace(rsc, "Freeing %s", rsc->id); free(container_data->prefix); free(container_data->image); free(container_data->control_port); free(container_data->host_network); free(container_data->host_netmask); free(container_data->ip_range_start); free(container_data->docker_network); free(container_data->docker_run_options); free(container_data->docker_run_command); free(container_data->docker_host_options); g_list_free_full(container_data->tuples, (GDestroyNotify)tuple_free); g_list_free_full(container_data->mounts, (GDestroyNotify)mount_free); g_list_free_full(container_data->ports, (GDestroyNotify)port_free); g_list_free(rsc->children); if(container_data->child) { free_xml(container_data->child->xml); container_data->child->xml = NULL; container_data->child->fns->free(container_data->child); } common_free(rsc); } enum rsc_role_e container_resource_state(const resource_t * rsc, gboolean current) { enum rsc_role_e container_role = RSC_ROLE_UNKNOWN; return container_role; } /*! * \brief Get the number of configured replicas in a bundle * * \param[in] rsc Bundle resource * * \return Number of configured replicas, or 0 on error */ int pe_bundle_replicas(const resource_t *rsc) { if ((rsc == NULL) || (rsc->variant != pe_container)) { return 0; } else { container_variant_data_t *container_data = NULL; get_container_variant_data(container_data, rsc); return container_data->replicas; } } diff --git a/lib/pengine/native.c b/lib/pengine/native.c index a1ad4f12ee..7097abd19c 100644 --- a/lib/pengine/native.c +++ b/lib/pengine/native.c @@ -1,949 +1,951 @@ /* * Copyright 2004-2018 Andrew Beekhof * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #define VARIANT_NATIVE 1 #include "./variant.h" void native_add_running(resource_t * rsc, node_t * node, pe_working_set_t * data_set) { GListPtr gIter = rsc->running_on; CRM_CHECK(node != NULL, return); for (; gIter != NULL; gIter = gIter->next) { node_t *a_node = (node_t *) gIter->data; CRM_CHECK(a_node != NULL, return); if (safe_str_eq(a_node->details->id, node->details->id)) { return; } } pe_rsc_trace(rsc, "Adding %s to %s %s", rsc->id, node->details->uname, is_set(rsc->flags, pe_rsc_managed)?"":"(unmanaged)"); rsc->running_on = g_list_append(rsc->running_on, node); if (rsc->variant == pe_native) { node->details->running_rsc = g_list_append(node->details->running_rsc, rsc); } if (rsc->variant == pe_native && node->details->maintenance) { clear_bit(rsc->flags, pe_rsc_managed); } if (is_not_set(rsc->flags, pe_rsc_managed)) { resource_t *p = rsc->parent; pe_rsc_info(rsc, "resource %s isn't managed", rsc->id); resource_location(rsc, node, INFINITY, "not_managed_default", data_set); while(p && node->details->online) { /* add without the additional location constraint */ p->running_on = g_list_append(p->running_on, node); p = p->parent; } return; } if (rsc->variant == pe_native && g_list_length(rsc->running_on) > 1) { switch (rsc->recovery_type) { case recovery_stop_only: { GHashTableIter gIter; node_t *local_node = NULL; /* make sure it doesn't come up again */ - g_hash_table_destroy(rsc->allowed_nodes); + if (rsc->allowed_nodes != NULL) { + g_hash_table_destroy(rsc->allowed_nodes); + } rsc->allowed_nodes = node_hash_from_list(data_set->nodes); g_hash_table_iter_init(&gIter, rsc->allowed_nodes); while (g_hash_table_iter_next(&gIter, NULL, (void **)&local_node)) { local_node->weight = -INFINITY; } } break; case recovery_stop_start: break; case recovery_block: clear_bit(rsc->flags, pe_rsc_managed); set_bit(rsc->flags, pe_rsc_block); /* If the resource belongs to a group or bundle configured with * multiple-active=block, block the entire entity. */ if (rsc->parent && (rsc->parent->variant == pe_group || rsc->parent->variant == pe_container) && rsc->parent->recovery_type == recovery_block) { GListPtr gIter = rsc->parent->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child = (resource_t *) gIter->data; clear_bit(child->flags, pe_rsc_managed); set_bit(child->flags, pe_rsc_block); } } break; } crm_debug("%s is active on %d nodes including %s: %s", rsc->id, g_list_length(rsc->running_on), node->details->uname, recovery2text(rsc->recovery_type)); } else { pe_rsc_trace(rsc, "Resource %s is active on: %s", rsc->id, node->details->uname); } if (rsc->parent != NULL) { native_add_running(rsc->parent, node, data_set); } } extern void force_non_unique_clone(resource_t * rsc, const char *rid, pe_working_set_t * data_set); gboolean native_unpack(resource_t * rsc, pe_working_set_t * data_set) { resource_t *parent = uber_parent(rsc); native_variant_data_t *native_data = NULL; const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); pe_rsc_trace(rsc, "Processing resource %s...", rsc->id); native_data = calloc(1, sizeof(native_variant_data_t)); rsc->variant_opaque = native_data; // Cloned LSB primitives must be anonymous if (is_set(rsc->flags, pe_rsc_unique) && rsc->parent && safe_str_eq(class, PCMK_RESOURCE_CLASS_LSB)) { force_non_unique_clone(parent, rsc->id, data_set); } // Only OCF primitives can be promotable clones if (is_set(parent->flags, pe_rsc_promotable) && safe_str_neq(class, PCMK_RESOURCE_CLASS_OCF)) { pe_err("Resource %s is of type %s and therefore " "cannot be used as a promotable clone resource", rsc->id, class); return FALSE; } return TRUE; } static bool rsc_is_on_node(resource_t *rsc, const node_t *node, int flags) { pe_rsc_trace(rsc, "Checking whether %s is on %s", rsc->id, node->details->uname); if (is_set(flags, pe_find_current) && rsc->running_on) { for (GListPtr iter = rsc->running_on; iter; iter = iter->next) { node_t *loc = (node_t *) iter->data; if (loc->details == node->details) { return TRUE; } } } else if (is_set(flags, pe_find_inactive) && (rsc->running_on == NULL)) { return TRUE; } else if (is_not_set(flags, pe_find_current) && rsc->allocated_to && (rsc->allocated_to->details == node->details)) { return TRUE; } return FALSE; } resource_t * native_find_rsc(resource_t * rsc, const char *id, const node_t *on_node, int flags) { bool match = FALSE; resource_t *result = NULL; CRM_CHECK(id && rsc && rsc->id, return NULL); if (flags & pe_find_clone) { const char *rid = ID(rsc->xml); if (!pe_rsc_is_clone(uber_parent(rsc))) { match = FALSE; } else if (!strcmp(id, rsc->id) || safe_str_eq(id, rid)) { match = TRUE; } } else if (!strcmp(id, rsc->id)) { match = TRUE; } else if (is_set(flags, pe_find_renamed) && rsc->clone_name && strcmp(rsc->clone_name, id) == 0) { match = TRUE; } else if (is_set(flags, pe_find_any) || (is_set(flags, pe_find_anon) && is_not_set(rsc->flags, pe_rsc_unique))) { match = pe_base_name_eq(rsc, id); } if (match && on_node) { bool match_node = rsc_is_on_node(rsc, on_node, flags); if (match_node == FALSE) { match = FALSE; } } if (match) { return rsc; } for (GListPtr gIter = rsc->children; gIter != NULL; gIter = gIter->next) { resource_t *child = (resource_t *) gIter->data; result = rsc->fns->find_rsc(child, id, on_node, flags); if (result) { return result; } } return NULL; } char * native_parameter(resource_t * rsc, node_t * node, gboolean create, const char *name, pe_working_set_t * data_set) { char *value_copy = NULL; const char *value = NULL; GHashTable *hash = NULL; GHashTable *local_hash = NULL; CRM_CHECK(rsc != NULL, return NULL); CRM_CHECK(name != NULL && strlen(name) != 0, return NULL); pe_rsc_trace(rsc, "Looking up %s in %s", name, rsc->id); if (create || g_hash_table_size(rsc->parameters) == 0) { if (node != NULL) { pe_rsc_trace(rsc, "Creating hash with node %s", node->details->uname); } else { pe_rsc_trace(rsc, "Creating default hash"); } local_hash = crm_str_table_new(); get_rsc_attributes(local_hash, rsc, node, data_set); hash = local_hash; } else { hash = rsc->parameters; } value = g_hash_table_lookup(hash, name); if (value == NULL) { /* try meta attributes instead */ value = g_hash_table_lookup(rsc->meta, name); } if (value != NULL) { value_copy = strdup(value); } if (local_hash != NULL) { g_hash_table_destroy(local_hash); } return value_copy; } gboolean native_active(resource_t * rsc, gboolean all) { GListPtr gIter = rsc->running_on; for (; gIter != NULL; gIter = gIter->next) { node_t *a_node = (node_t *) gIter->data; if (a_node->details->unclean) { crm_debug("Resource %s: node %s is unclean", rsc->id, a_node->details->uname); return TRUE; } else if (a_node->details->online == FALSE) { crm_debug("Resource %s: node %s is offline", rsc->id, a_node->details->uname); } else { crm_debug("Resource %s active on %s", rsc->id, a_node->details->uname); return TRUE; } } return FALSE; } struct print_data_s { long options; void *print_data; }; static void native_print_attr(gpointer key, gpointer value, gpointer user_data) { long options = ((struct print_data_s *)user_data)->options; void *print_data = ((struct print_data_s *)user_data)->print_data; status_print("Option: %s = %s\n", (char *)key, (char *)value); } static const char * native_pending_state(resource_t * rsc) { const char *pending_state = NULL; if (safe_str_eq(rsc->pending_task, CRMD_ACTION_START)) { pending_state = "Starting"; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_STOP)) { pending_state = "Stopping"; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_MIGRATE)) { pending_state = "Migrating"; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_MIGRATED)) { /* Work might be done in here. */ pending_state = "Migrating"; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_PROMOTE)) { pending_state = "Promoting"; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_DEMOTE)) { pending_state = "Demoting"; } return pending_state; } static const char * native_pending_task(resource_t * rsc) { const char *pending_task = NULL; if (safe_str_eq(rsc->pending_task, CRMD_ACTION_NOTIFY)) { /* "Notifying" is not very useful to be shown. */ pending_task = NULL; } else if (safe_str_eq(rsc->pending_task, CRMD_ACTION_STATUS)) { pending_task = "Monitoring"; /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * unpack.c:unpack_rsc_op(). */ /* } else if (safe_str_eq(rsc->pending_task, "probe")) { pending_task = "Checking"; */ } return pending_task; } static enum rsc_role_e native_displayable_role(resource_t *rsc) { enum rsc_role_e role = rsc->role; if ((role == RSC_ROLE_STARTED) && is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) { role = RSC_ROLE_SLAVE; } return role; } static const char * native_displayable_state(resource_t *rsc, long options) { const char *rsc_state = NULL; if (options & pe_print_pending) { rsc_state = native_pending_state(rsc); } if (rsc_state == NULL) { rsc_state = role2text(native_displayable_role(rsc)); } return rsc_state; } static void native_print_xml(resource_t * rsc, const char *pre_text, long options, void *print_data) { const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); const char *rsc_state = native_displayable_state(rsc, options); const char *target_role = NULL; /* resource information. */ status_print("%sxml, XML_ATTR_TYPE)); status_print("role=\"%s\" ", rsc_state); if (rsc->meta) { target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } if (target_role) { status_print("target_role=\"%s\" ", target_role); } status_print("active=\"%s\" ", rsc->fns->active(rsc, TRUE) ? "true" : "false"); status_print("orphaned=\"%s\" ", is_set(rsc->flags, pe_rsc_orphan) ? "true" : "false"); status_print("blocked=\"%s\" ", is_set(rsc->flags, pe_rsc_block) ? "true" : "false"); status_print("managed=\"%s\" ", is_set(rsc->flags, pe_rsc_managed) ? "true" : "false"); status_print("failed=\"%s\" ", is_set(rsc->flags, pe_rsc_failed) ? "true" : "false"); status_print("failure_ignored=\"%s\" ", is_set(rsc->flags, pe_rsc_failure_ignored) ? "true" : "false"); status_print("nodes_running_on=\"%d\" ", g_list_length(rsc->running_on)); if (options & pe_print_pending) { const char *pending_task = native_pending_task(rsc); if (pending_task) { status_print("pending=\"%s\" ", pending_task); } } if (options & pe_print_dev) { status_print("provisional=\"%s\" ", is_set(rsc->flags, pe_rsc_provisional) ? "true" : "false"); status_print("runnable=\"%s\" ", is_set(rsc->flags, pe_rsc_runnable) ? "true" : "false"); status_print("priority=\"%f\" ", (double)rsc->priority); status_print("variant=\"%s\" ", crm_element_name(rsc->xml)); } /* print out the nodes this resource is running on */ if (options & pe_print_rsconly) { status_print("/>\n"); /* do nothing */ } else if (g_list_length(rsc->running_on) > 0) { GListPtr gIter = rsc->running_on; status_print(">\n"); for (; gIter != NULL; gIter = gIter->next) { node_t *node = (node_t *) gIter->data; status_print("%s \n", pre_text, node->details->uname, node->details->id, node->details->online ? "false" : "true"); } status_print("%s\n", pre_text); } else { status_print("/>\n"); } } /* making this inline rather than a macro prevents a coverity "unreachable" * warning on the first usage */ static inline const char * comma_if(int i) { return i? ", " : ""; } void common_print(resource_t * rsc, const char *pre_text, const char *name, node_t *node, long options, void *print_data) { const char *desc = NULL; const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE); const char *target_role = NULL; enum rsc_role_e role = native_displayable_role(rsc); int offset = 0; int flagOffset = 0; char buffer[LINE_MAX]; char flagBuffer[LINE_MAX]; CRM_ASSERT(rsc->variant == pe_native); CRM_ASSERT(kind != NULL); if (rsc->meta) { const char *is_internal = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_INTERNAL_RSC); if (crm_is_true(is_internal) && is_not_set(options, pe_print_implicit)) { crm_trace("skipping print of internal resource %s", rsc->id); return; } target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); } if (pre_text == NULL && (options & pe_print_printf)) { pre_text = " "; } if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } if ((options & pe_print_rsconly) || g_list_length(rsc->running_on) > 1) { node = NULL; } if (options & pe_print_html) { if (is_not_set(rsc->flags, pe_rsc_managed)) { status_print(""); } else if (is_set(rsc->flags, pe_rsc_failed)) { status_print(""); } else if (rsc->variant == pe_native && g_list_length(rsc->running_on) == 0) { status_print(""); } else if (g_list_length(rsc->running_on) > 1) { status_print(""); } else if (is_set(rsc->flags, pe_rsc_failure_ignored)) { status_print(""); } else { status_print(""); } } if(pre_text) { offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", pre_text); } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", name); offset += snprintf(buffer + offset, LINE_MAX - offset, "\t(%s", class); if (crm_provider_required(class)) { const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); offset += snprintf(buffer + offset, LINE_MAX - offset, "::%s", prov); } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s):\t", kind); if(is_set(rsc->flags, pe_rsc_orphan)) { offset += snprintf(buffer + offset, LINE_MAX - offset, " ORPHANED "); } if(role > RSC_ROLE_SLAVE && is_set(rsc->flags, pe_rsc_failed)) { offset += snprintf(buffer + offset, LINE_MAX - offset, "FAILED %s", role2text(role)); } else if(is_set(rsc->flags, pe_rsc_failed)) { offset += snprintf(buffer + offset, LINE_MAX - offset, "FAILED"); } else { const char *rsc_state = native_displayable_state(rsc, options); offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", rsc_state); } if(node) { offset += snprintf(buffer + offset, LINE_MAX - offset, " %s", node->details->uname); if (node->details->online == FALSE && node->details->unclean) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%sUNCLEAN", comma_if(flagOffset)); } } if (options & pe_print_pending) { const char *pending_task = native_pending_task(rsc); if (pending_task) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%s%s", comma_if(flagOffset), pending_task); } } if (target_role) { enum rsc_role_e target_role_e = text2role(target_role); /* Ignore target role Started, as it is the default anyways * (and would also allow a Master to be Master). * Show if target role limits our abilities. */ if (target_role_e == RSC_ROLE_STOPPED) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%sdisabled", comma_if(flagOffset)); rsc->cluster->disabled_resources++; } else if (is_set(uber_parent(rsc)->flags, pe_rsc_promotable) && target_role_e == RSC_ROLE_SLAVE) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%starget-role:%s", comma_if(flagOffset), target_role); rsc->cluster->disabled_resources++; } } if (is_set(rsc->flags, pe_rsc_block)) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%sblocked", comma_if(flagOffset)); rsc->cluster->blocked_resources++; } else if (is_not_set(rsc->flags, pe_rsc_managed)) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%sunmanaged", comma_if(flagOffset)); } if(is_set(rsc->flags, pe_rsc_failure_ignored)) { flagOffset += snprintf(flagBuffer + flagOffset, LINE_MAX - flagOffset, "%sfailure ignored", comma_if(flagOffset)); } if ((options & pe_print_rsconly) || g_list_length(rsc->running_on) > 1) { desc = crm_element_value(rsc->xml, XML_ATTR_DESC); } CRM_LOG_ASSERT(offset > 0); if(flagOffset > 0) { status_print("%s (%s)%s%s", buffer, flagBuffer, desc?" ":"", desc?desc:""); } else { status_print("%s%s%s", buffer, desc?" ":"", desc?desc:""); } #if CURSES_ENABLED if ((options & pe_print_rsconly) || g_list_length(rsc->running_on) > 1) { /* Done */ } else if (options & pe_print_ncurses) { /* coverity[negative_returns] False positive */ move(-1, 0); } #endif if (options & pe_print_html) { status_print(" "); } if ((options & pe_print_rsconly)) { } else if (g_list_length(rsc->running_on) > 1) { GListPtr gIter = rsc->running_on; int counter = 0; if (options & pe_print_html) { status_print("
      \n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("["); } for (; gIter != NULL; gIter = gIter->next) { node_t *n = (node_t *) gIter->data; counter++; if (options & pe_print_html) { status_print("
    • \n%s", n->details->uname); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" %s", n->details->uname); } else if ((options & pe_print_log)) { status_print("\t%d : %s", counter, n->details->uname); } else { status_print("%s", n->details->uname); } if (options & pe_print_html) { status_print("
    • \n"); } } if (options & pe_print_html) { status_print("
    \n"); } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print(" ]"); } } if (options & pe_print_html) { status_print("
    \n"); } else if (options & pe_print_suppres_nl) { /* nothing */ } else if ((options & pe_print_printf) || (options & pe_print_ncurses)) { status_print("\n"); } if (options & pe_print_details) { struct print_data_s pdata; pdata.options = options; pdata.print_data = print_data; g_hash_table_foreach(rsc->parameters, native_print_attr, &pdata); } if (options & pe_print_dev) { GHashTableIter iter; node_t *n = NULL; status_print("%s\t(%s%svariant=%s, priority=%f)", pre_text, is_set(rsc->flags, pe_rsc_provisional) ? "provisional, " : "", is_set(rsc->flags, pe_rsc_runnable) ? "" : "non-startable, ", crm_element_name(rsc->xml), (double)rsc->priority); status_print("%s\tAllowed Nodes", pre_text); g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) { status_print("%s\t * %s %d", pre_text, n->details->uname, n->weight); } } if (options & pe_print_max_details) { GHashTableIter iter; node_t *n = NULL; status_print("%s\t=== Allowed Nodes\n", pre_text); g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&n)) { print_node("\t", n, FALSE); } } } void native_print(resource_t * rsc, const char *pre_text, long options, void *print_data) { node_t *node = NULL; CRM_ASSERT(rsc->variant == pe_native); if (options & pe_print_xml) { native_print_xml(rsc, pre_text, options, print_data); return; } if (rsc->running_on != NULL) { node = rsc->running_on->data; } common_print(rsc, pre_text, rsc_printable_id(rsc), node, options, print_data); } void native_free(resource_t * rsc) { pe_rsc_trace(rsc, "Freeing resource action list (not the data)"); common_free(rsc); } enum rsc_role_e native_resource_state(const resource_t * rsc, gboolean current) { enum rsc_role_e role = rsc->next_role; if (current) { role = rsc->role; } pe_rsc_trace(rsc, "%s state: %s", rsc->id, role2text(role)); return role; } node_t * native_location(resource_t * rsc, GListPtr * list, gboolean current) { node_t *one = NULL; GListPtr result = NULL; if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child = (resource_t *) gIter->data; child->fns->location(child, &result, current); } } else if (current && rsc->running_on) { result = g_list_copy(rsc->running_on); } else if (current == FALSE && rsc->allocated_to) { result = g_list_append(NULL, rsc->allocated_to); } if (result && g_list_length(result) == 1) { one = g_list_nth_data(result, 0); } if (list) { GListPtr gIter = result; for (; gIter != NULL; gIter = gIter->next) { node_t *node = (node_t *) gIter->data; if (*list == NULL || pe_find_node_id(*list, node->details->id) == NULL) { *list = g_list_append(*list, node); } } } g_list_free(result); return one; } static void get_rscs_brief(GListPtr rsc_list, GHashTable * rsc_table, GHashTable * active_table) { GListPtr gIter = rsc_list; for (; gIter != NULL; gIter = gIter->next) { resource_t *rsc = (resource_t *) gIter->data; const char *class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); const char *kind = crm_element_value(rsc->xml, XML_ATTR_TYPE); int offset = 0; char buffer[LINE_MAX]; int *rsc_counter = NULL; int *active_counter = NULL; if (rsc->variant != pe_native) { continue; } offset += snprintf(buffer + offset, LINE_MAX - offset, "%s", class); if (crm_provider_required(class)) { const char *prov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER); offset += snprintf(buffer + offset, LINE_MAX - offset, "::%s", prov); } offset += snprintf(buffer + offset, LINE_MAX - offset, ":%s", kind); CRM_LOG_ASSERT(offset > 0); if (rsc_table) { rsc_counter = g_hash_table_lookup(rsc_table, buffer); if (rsc_counter == NULL) { rsc_counter = calloc(1, sizeof(int)); *rsc_counter = 0; g_hash_table_insert(rsc_table, strdup(buffer), rsc_counter); } (*rsc_counter)++; } if (active_table) { GListPtr gIter2 = rsc->running_on; for (; gIter2 != NULL; gIter2 = gIter2->next) { node_t *node = (node_t *) gIter2->data; GHashTable *node_table = NULL; if (node->details->unclean == FALSE && node->details->online == FALSE) { continue; } node_table = g_hash_table_lookup(active_table, node->details->uname); if (node_table == NULL) { node_table = crm_str_table_new(); g_hash_table_insert(active_table, strdup(node->details->uname), node_table); } active_counter = g_hash_table_lookup(node_table, buffer); if (active_counter == NULL) { active_counter = calloc(1, sizeof(int)); *active_counter = 0; g_hash_table_insert(node_table, strdup(buffer), active_counter); } (*active_counter)++; } } } } static void destroy_node_table(gpointer data) { GHashTable *node_table = data; if (node_table) { g_hash_table_destroy(node_table); } } void print_rscs_brief(GListPtr rsc_list, const char *pre_text, long options, void *print_data, gboolean print_all) { GHashTable *rsc_table = crm_str_table_new(); GHashTable *active_table = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_node_table); GHashTableIter hash_iter; char *type = NULL; int *rsc_counter = NULL; get_rscs_brief(rsc_list, rsc_table, active_table); g_hash_table_iter_init(&hash_iter, rsc_table); while (g_hash_table_iter_next(&hash_iter, (gpointer *)&type, (gpointer *)&rsc_counter)) { GHashTableIter hash_iter2; char *node_name = NULL; GHashTable *node_table = NULL; int active_counter_all = 0; g_hash_table_iter_init(&hash_iter2, active_table); while (g_hash_table_iter_next(&hash_iter2, (gpointer *)&node_name, (gpointer *)&node_table)) { int *active_counter = g_hash_table_lookup(node_table, type); if (active_counter == NULL || *active_counter == 0) { continue; } else { active_counter_all += *active_counter; } if (options & pe_print_rsconly) { node_name = NULL; } if (options & pe_print_html) { status_print("
  • \n"); } if (print_all) { status_print("%s%d/%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, rsc_counter ? *rsc_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } else { status_print("%s%d\t(%s):\tActive %s\n", pre_text ? pre_text : "", active_counter ? *active_counter : 0, type, active_counter && (*active_counter > 0) && node_name ? node_name : ""); } if (options & pe_print_html) { status_print("
  • \n"); } } if (print_all && active_counter_all == 0) { if (options & pe_print_html) { status_print("
  • \n"); } status_print("%s%d/%d\t(%s):\tActive\n", pre_text ? pre_text : "", active_counter_all, rsc_counter ? *rsc_counter : 0, type); if (options & pe_print_html) { status_print("
  • \n"); } } } if (rsc_table) { g_hash_table_destroy(rsc_table); rsc_table = NULL; } if (active_table) { g_hash_table_destroy(active_table); active_table = NULL; } } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index 67b1ef97c0..2fe839261b 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,3350 +1,3352 @@ /* * Copyright 2004-2018 Andrew Beekhof * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); #define set_config_flag(data_set, option, flag) do { \ const char *tmp = pe_pref(data_set->config_hash, option); \ if(tmp) { \ if(crm_is_true(tmp)) { \ set_bit(data_set->flags, flag); \ } else { \ clear_bit(data_set->flags, flag); \ } \ } \ } while(0) gboolean unpack_rsc_op(resource_t * rsc, node_t * node, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response *failed, pe_working_set_t * data_set); static gboolean determine_remote_online_status(pe_working_set_t * data_set, node_t * this_node); // Bitmask for warnings we only want to print once uint32_t pe_wo = 0; static gboolean is_dangling_container_remote_node(node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (is_remote_node(node) && node->details->remote_rsc && node->details->remote_rsc->container == NULL && is_set(node->details->remote_rsc->flags, pe_rsc_orphan_container_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] data_set Current working set of cluster * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed */ void pe_fence_node(pe_working_set_t * data_set, node_t * node, const char *reason) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (is_container_remote_node(node)) { resource_t *rsc = node->details->remote_rsc->container; if (is_set(rsc->flags, pe_rsc_failed) == FALSE) { if (!is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", node->details->uname, reason, rsc->id); } else { crm_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", node->details->uname, rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; set_bit(rsc->flags, pe_rsc_failed); } } } else if (is_dangling_container_remote_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", node->details->uname, reason); set_bit(node->details->remote_rsc->flags, pe_rsc_failed); } else if (is_baremetal_remote_node(node)) { resource_t *rsc = node->details->remote_rsc; if (rsc && (!is_set(rsc->flags, pe_rsc_managed))) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", node->details->uname, reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; crm_warn("Remote node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, data_set); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", node->details->uname, pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean", reason); } else { crm_warn("Cluster node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, data_set); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \ "[(@" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_PROVIDES "'" \ "or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \ "and @" XML_NVPAIR_ATTR_VALUE "='unfencing']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \ "//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \ "|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \ "/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(unsigned long long flag, const char *xpath, pe_working_set_t *data_set) { xmlXPathObjectPtr result = NULL; if (is_not_set(data_set->flags, flag)) { result = xpath_search(data_set->input, xpath); if (result && (numXpathResults(result) > 0)) { set_bit(data_set->flags, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode * config, pe_working_set_t * data_set) { const char *value = NULL; GHashTable *config_hash = crm_str_table_new(); data_set->config_hash = config_hash; unpack_instance_attributes(data_set->input, config, XML_CIB_TAG_PROPSET, NULL, config_hash, CIB_OPTIONS_FIRST, FALSE, data_set->now); verify_pe_options(data_set->config_hash); set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes); if(is_not_set(data_set->flags, pe_flag_startup_probes)) { crm_info("Startup probes: disabled (dangerous)"); } value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_notice("Watchdog will be used via SBD if fencing is required"); set_bit(data_set->flags, pe_flag_have_stonith_resource); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set); value = pe_pref(data_set->config_hash, "stonith-timeout"); data_set->stonith_timeout = crm_get_msec(value); crm_debug("STONITH timeout: %d", data_set->stonith_timeout); set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled); crm_debug("STONITH of failed nodes is %s", is_set(data_set->flags, pe_flag_stonith_enabled) ? "enabled" : "disabled"); data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action"); if (!strcmp(data_set->stonith_action, "poweroff")) { pe_warn_once(pe_wo_poweroff, "Support for stonith-action of 'poweroff' is deprecated " "and will be removed in a future release (use 'off' instead)"); data_set->stonith_action = "off"; } crm_trace("STONITH will %s nodes", data_set->stonith_action); set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing); crm_debug("Concurrent fencing is %s", is_set(data_set->flags, pe_flag_concurrent_fencing) ? "enabled" : "disabled"); set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything); crm_debug("Stop all active resources: %s", is_set(data_set->flags, pe_flag_stop_everything) ? "true" : "false"); set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster); if (is_set(data_set->flags, pe_flag_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pe_pref(data_set->config_hash, "no-quorum-policy"); if (safe_str_eq(value, "ignore")) { data_set->no_quorum_policy = no_quorum_ignore; } else if (safe_str_eq(value, "freeze")) { data_set->no_quorum_policy = no_quorum_freeze; } else if (safe_str_eq(value, "suicide")) { if (is_set(data_set->flags, pe_flag_stonith_enabled)) { int do_panic = 0; crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC, &do_panic); if (do_panic || is_set(data_set->flags, pe_flag_have_quorum)) { data_set->no_quorum_policy = no_quorum_suicide; } else { crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum"); data_set->no_quorum_policy = no_quorum_stop; } } else { crm_config_err("Resetting no-quorum-policy to 'stop': stonith is not configured"); data_set->no_quorum_policy = no_quorum_stop; } } else { data_set->no_quorum_policy = no_quorum_stop; } switch (data_set->no_quorum_policy) { case no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case no_quorum_suicide: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans); crm_trace("Orphan resources are %s", is_set(data_set->flags, pe_flag_stop_rsc_orphans) ? "stopped" : "ignored"); set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans); crm_trace("Orphan resource actions are %s", is_set(data_set->flags, pe_flag_stop_action_orphans) ? "stopped" : "ignored"); set_config_flag(data_set, "remove-after-stop", pe_flag_remove_after_stop); crm_trace("Stopped resources are removed from the status section: %s", is_set(data_set->flags, pe_flag_remove_after_stop) ? "true" : "false"); set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode); crm_trace("Maintenance mode: %s", is_set(data_set->flags, pe_flag_maintenance_mode) ? "true" : "false"); set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal); crm_trace("Start failures are %s", is_set(data_set->flags, pe_flag_start_failure_fatal) ? "always fatal" : "handled by failcount"); if (is_set(data_set->flags, pe_flag_stonith_enabled)) { set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing); } if (is_set(data_set->flags, pe_flag_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes"); } node_score_red = char2score(pe_pref(data_set->config_hash, "node-health-red")); node_score_green = char2score(pe_pref(data_set->config_hash, "node-health-green")); node_score_yellow = char2score(pe_pref(data_set->config_hash, "node-health-yellow")); crm_debug("Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s", pe_pref(data_set->config_hash, "node-health-red"), pe_pref(data_set->config_hash, "node-health-yellow"), pe_pref(data_set->config_hash, "node-health-green")); data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy"); crm_trace("Placement strategy: %s", data_set->placement_strategy); return TRUE; } static void destroy_digest_cache(gpointer ptr) { op_digest_cache_t *data = ptr; free_xml(data->params_all); free_xml(data->params_secure); free_xml(data->params_restart); free(data->digest_all_calc); free(data->digest_restart_calc); free(data->digest_secure_calc); free(data); } node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t * data_set) { node_t *new_node = NULL; if (pe_find_node(data_set->nodes, uname) != NULL) { crm_config_warn("Detected multiple node entries with uname=%s" " - this is rarely intended", uname); } new_node = calloc(1, sizeof(node_t)); if (new_node == NULL) { return NULL; } new_node->weight = char2score(score); new_node->fixed = FALSE; new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->type = node_ping; if (safe_str_eq(type, "remote")) { new_node->details->type = node_remote; set_bit(data_set->flags, pe_flag_have_remote_nodes); } else if ((type == NULL) || safe_str_eq(type, "member")) { new_node->details->type = node_member; } new_node->details->attrs = crm_str_table_new(); if (is_remote_node(new_node)) { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("remote")); } else { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("cluster")); } new_node->details->utilization = crm_str_table_new(); new_node->details->digest_cache = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_digest_cache); data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node, sort_node_uname); return new_node; } bool remote_id_conflict(const char *remote_name, pe_working_set_t *data) { bool match = FALSE; #if 1 pe_find_resource(data->resources, remote_name); #else if (data->name_check == NULL) { data->name_check = g_hash_table_new(crm_str_hash, g_str_equal); for (xml_rsc = __xml_first_child(parent); xml_rsc != NULL; xml_rsc = __xml_next_element(xml_rsc)) { const char *id = ID(xml_rsc); /* avoiding heap allocation here because we know the duration of this hashtable allows us to */ g_hash_table_insert(data->name_check, (char *) id, (char *) id); } } if (g_hash_table_lookup(data->name_check, remote_name)) { match = TRUE; } #endif if (match) { crm_err("Invalid remote-node name, a resource called '%s' already exists.", remote_name); return NULL; } return match; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = ID(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *container_managed = NULL; for (attr_set = __xml_first_child(xml_obj); attr_set != NULL; attr_set = __xml_next_element(attr_set)) { if (safe_str_neq((const char *)attr_set->name, XML_TAG_META_SETS)) { continue; } for (attr = __xml_first_child(attr_set); attr != NULL; attr = __xml_next_element(attr)) { const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); if (safe_str_eq(name, XML_RSC_ATTR_REMOTE_NODE)) { remote_name = value; } else if (safe_str_eq(name, "remote-addr")) { remote_server = value; } else if (safe_str_eq(name, "remote-port")) { remote_port = value; } else if (safe_str_eq(name, "remote-connect-timeout")) { connect_timeout = value; } else if (safe_str_eq(name, "remote-allow-migrate")) { remote_allow_migrate=value; } else if (safe_str_eq(name, XML_RSC_ATTR_MANAGED)) { container_managed = value; } } } if (remote_name == NULL) { return NULL; } if (remote_id_conflict(remote_name, data)) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, container_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pe_working_set_t *data_set, node_t *new_node) { if ((new_node->details->type == node_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (is_set(data_set->flags, pe_flag_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode * xml_nodes, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; for (xml_obj = __xml_first_child(xml_nodes); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) { if (crm_str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, TRUE)) { new_node = NULL; id = crm_element_value(xml_obj, XML_ATTR_ID); uname = crm_element_value(xml_obj, XML_ATTR_UNAME); type = crm_element_value(xml_obj, XML_ATTR_TYPE); score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { crm_config_err("Must specify id tag in "); continue; } new_node = pe_create_node(id, uname, type, score, data_set); if (new_node == NULL) { return FALSE; } /* if(data_set->have_quorum == FALSE */ /* && data_set->no_quorum_policy == no_quorum_stop) { */ /* /\* start shutting resources down *\/ */ /* new_node->weight = -INFINITY; */ /* } */ handle_startup_fencing(data_set, new_node); add_node_attrs(xml_obj, new_node, FALSE, data_set); unpack_instance_attributes(data_set->input, xml_obj, XML_TAG_UTILIZATION, NULL, new_node->details->utilization, NULL, FALSE, data_set->now); crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME)); } } if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) { crm_info("Creating a fake local node"); pe_create_node(data_set->localhost, data_set->localhost, NULL, 0, data_set); } return TRUE; } static void setup_container(resource_t * rsc, pe_working_set_t * data_set) { const char *container_id = NULL; if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { resource_t *child_rsc = (resource_t *) gIter->data; setup_container(child_rsc, data_set); } return; } container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER); if (container_id && safe_str_neq(container_id, rsc->id)) { resource_t *container = pe_find_resource(data_set->resources, container_id); if (container) { rsc->container = container; set_bit(container->flags, pe_rsc_is_container); container->fillers = g_list_append(container->fillers, rsc); pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; /* generate remote nodes from resource config before unpacking resources */ for (xml_obj = __xml_first_child(xml_resources); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) { const char *new_node_id = NULL; /* first check if this is a bare metal remote node. Bare metal remote nodes * are defined as a resource primitive only. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = ID(xml_obj); /* The "pe_find_node" check is here to make sure we don't iterate over * an expanded node that has already been added to the node list. */ if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found baremetal remote node %s in container resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Now check for guest remote nodes. * guest remote nodes are defined within a resource primitive. * Example1: a vm resource might be configured as a remote node. * Example2: a vm resource might be configured within a group to be a remote node. * Note: right now we only support guest remote nodes in as a standalone primitive * or a primitive within a group. No cloned primitives can be a guest remote node * right now */ if (crm_str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, TRUE)) { /* expands a metadata defined remote resource into the xml config * as an actual rsc primitive to be unpacked later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest remote node %s in container resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } else if (crm_str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, TRUE)) { xmlNode *xml_obj2 = NULL; /* search through a group to see if any of the primitive contain a remote node. */ for (xml_obj2 = __xml_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = __xml_next_element(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest remote node %s in container resource %s which is in group %s", new_node_id, ID(xml_obj2), ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the PE calculations. */ static void link_rsc2remotenode(pe_working_set_t *data_set, resource_t *new_rsc) { node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (is_set(data_set->flags, pe_flag_quick_location)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } print_resource(LOG_TRACE, "Linking remote-node connection resource, ", new_rsc, FALSE); remote_node = pe_find_node(data_set->nodes, new_rsc->id); CRM_CHECK(remote_node != NULL, return;); remote_node->details->remote_rsc = new_rsc; /* If this is a baremetal remote-node (no container resource * associated with it) then we need to handle startup fencing the same way * as cluster nodes. */ if (new_rsc->container == NULL) { handle_startup_fencing(data_set, remote_node); } else { /* At this point we know if the remote node is a container or baremetal * remote node, update the #kind attribute if a container is involved */ g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); } } static void destroy_tag(gpointer data) { tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] data_set Where to put resource information * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when common_unpack() calls resource_location() */ gboolean unpack_resources(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; GListPtr gIter = NULL; data_set->template_rsc_sets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_tag); for (xml_obj = __xml_first_child(xml_resources); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) { resource_t *new_rsc = NULL; if (crm_str_eq((const char *)xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE, TRUE)) { const char *template_id = ID(xml_obj); if (template_id && g_hash_table_lookup_extended(data_set->template_rsc_sets, template_id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ g_hash_table_insert(data_set->template_rsc_sets, strdup(template_id), NULL); } continue; } crm_trace("Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj), ID(xml_obj)); if (common_unpack(xml_obj, &new_rsc, NULL, data_set)) { data_set->resources = g_list_append(data_set->resources, new_rsc); print_resource(LOG_TRACE, "Added ", new_rsc, FALSE); } else { crm_config_err("Failed unpacking %s %s", crm_element_name(xml_obj), crm_element_value(xml_obj, XML_ATTR_ID)); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } } } for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { resource_t *rsc = (resource_t *) gIter->data; setup_container(rsc, data_set); link_rsc2remotenode(data_set, rsc); } data_set->resources = g_list_sort(data_set->resources, sort_rsc_priority); if (is_set(data_set->flags, pe_flag_quick_location)) { /* Ignore */ } else if (is_set(data_set->flags, pe_flag_stonith_enabled) && is_set(data_set->flags, pe_flag_have_stonith_resource) == FALSE) { crm_config_err("Resource start-up disabled since no STONITH resources have been defined"); crm_config_err("Either configure some or disable STONITH with the stonith-enabled option"); crm_config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode * xml_tags, pe_working_set_t * data_set) { xmlNode *xml_tag = NULL; data_set->tags = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_tag); for (xml_tag = __xml_first_child(xml_tags); xml_tag != NULL; xml_tag = __xml_next_element(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = ID(xml_tag); if (crm_str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, TRUE) == FALSE) { continue; } if (tag_id == NULL) { crm_config_err("Failed unpacking %s: %s should be specified", crm_element_name(xml_tag), XML_ATTR_ID); continue; } for (xml_obj_ref = __xml_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = __xml_next_element(xml_obj_ref)) { const char *obj_ref = ID(xml_obj_ref); if (crm_str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, TRUE) == FALSE) { continue; } if (obj_ref == NULL) { crm_config_err("Failed unpacking %s for tag %s: %s should be specified", crm_element_name(xml_obj_ref), tag_id, XML_ATTR_ID); continue; } if (add_tag_ref(data_set->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode * xml_ticket, pe_working_set_t * data_set) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; ticket_t *ticket = NULL; ticket_id = ID(xml_ticket); if (ticket_id == NULL || strlen(ticket_id) == 0) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, data_set); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_ticket, prop_name); if (crm_str_eq(prop_name, XML_ATTR_ID, TRUE)) { continue; } g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value)); } granted = g_hash_table_lookup(ticket->state, "granted"); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, "last-granted"); if (last_granted) { ticket->last_granted = crm_parse_int(last_granted, 0); } standby = g_hash_table_lookup(ticket->state, "standby"); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode * xml_tickets, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; for (xml_obj = __xml_first_child(xml_tickets); xml_obj != NULL; xml_obj = __xml_next_element(xml_obj)) { if (crm_str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, TRUE) == FALSE) { continue; } unpack_ticket_state(xml_obj, data_set); } return TRUE; } static void unpack_handle_remote_attrs(node_t *this_node, xmlNode *state, pe_working_set_t * data_set) { const char *resource_discovery_enabled = NULL; xmlNode *attrs = NULL; resource_t *rsc = NULL; const char *shutdown = NULL; if (crm_str_eq((const char *)state->name, XML_CIB_TAG_STATE, TRUE) == FALSE) { return; } if ((this_node == NULL) || (is_remote_node(this_node) == FALSE)) { return; } crm_trace("Processing remote node id=%s, uname=%s", this_node->details->id, this_node->details->uname); this_node->details->remote_maintenance = crm_atoi(crm_element_value(state, XML_NODE_IS_MAINTENANCE), "0"); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); shutdown = pe_node_attribute_raw(this_node, XML_CIB_ATTR_SHUTDOWN); if (shutdown != NULL && safe_str_neq("0", shutdown)) { crm_info("Node %s is shutting down", this_node->details->uname); this_node->details->shutdown = TRUE; if (rsc) { rsc->next_role = RSC_ROLE_STOPPED; } } if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) || (rsc && !is_set(rsc->flags, pe_rsc_managed))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { if (is_baremetal_remote_node(this_node) && is_not_set(data_set->flags, pe_flag_stonith_enabled)) { crm_warn("ignoring %s attribute on baremetal remote node %s, disabling resource discovery requires stonith to be enabled.", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } else { /* if we're here, this is either a baremetal node and fencing is enabled, * or this is a container node which we don't care if fencing is enabled * or not on. container nodes are 'fenced' by recovering the container resource * regardless of whether fencing is enabled. */ crm_info("Node %s has resource discovery disabled", this_node->details->uname); this_node->details->rsc_discovery_enabled = FALSE; } } } static bool unpack_node_loop(xmlNode * status, bool fence, pe_working_set_t * data_set) { bool changed = false; xmlNode *lrm_rsc = NULL; for (xmlNode *state = __xml_first_child(status); state != NULL; state = __xml_next_element(state)) { const char *id = NULL; const char *uname = NULL; node_t *this_node = NULL; bool process = FALSE; if (crm_str_eq((const char *)state->name, XML_CIB_TAG_STATE, TRUE) == FALSE) { continue; } id = crm_element_value(state, XML_ATTR_ID); uname = crm_element_value(state, XML_ATTR_UNAME); this_node = pe_find_node_any(data_set->nodes, id, uname); if (this_node == NULL) { crm_info("Node %s is unknown", id); continue; } else if (this_node->details->unpacked) { crm_info("Node %s is already processed", id); continue; } else if (is_remote_node(this_node) == FALSE && is_set(data_set->flags, pe_flag_stonith_enabled)) { // A redundant test, but preserves the order for regression tests process = TRUE; } else if (is_remote_node(this_node)) { bool check = FALSE; resource_t *rsc = this_node->details->remote_rsc; if(fence) { check = TRUE; } else if(rsc == NULL) { /* Not ready yet */ } else if (is_container_remote_node(this_node) && rsc->role == RSC_ROLE_STARTED && rsc->container->role == RSC_ROLE_STARTED) { /* Both the connection and the underlying container * need to be known 'up' before we volunterily process * resources inside it */ check = TRUE; crm_trace("Checking node %s/%s/%s status %d/%d/%d", id, rsc->id, rsc->container->id, fence, rsc->role, RSC_ROLE_STARTED); } else if (is_container_remote_node(this_node) == FALSE && rsc->role == RSC_ROLE_STARTED) { check = TRUE; crm_trace("Checking node %s/%s status %d/%d/%d", id, rsc->id, fence, rsc->role, RSC_ROLE_STARTED); } if (check) { determine_remote_online_status(data_set, this_node); unpack_handle_remote_attrs(this_node, state, data_set); process = TRUE; } } else if (this_node->details->online) { process = TRUE; } else if (fence) { process = TRUE; } if(process) { crm_trace("Processing lrm resource entries on %shealthy%s node: %s", fence?"un":"", is_remote_node(this_node)?" remote":"", this_node->details->uname); changed = TRUE; this_node->details->unpacked = TRUE; lrm_rsc = find_xml_node(state, XML_CIB_TAG_LRM, FALSE); lrm_rsc = find_xml_node(lrm_rsc, XML_LRM_TAG_RESOURCES, FALSE); unpack_lrm_resources(this_node, lrm_rsc, data_set); } } return changed; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode * status, pe_working_set_t * data_set) { const char *id = NULL; const char *uname = NULL; xmlNode *state = NULL; node_t *this_node = NULL; crm_trace("Beginning unpack"); if (data_set->tickets == NULL) { data_set->tickets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_ticket); } for (state = __xml_first_child(status); state != NULL; state = __xml_next_element(state)) { if (crm_str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, TRUE)) { unpack_tickets_state((xmlNode *) state, data_set); } else if (crm_str_eq((const char *)state->name, XML_CIB_TAG_STATE, TRUE)) { xmlNode *attrs = NULL; const char *resource_discovery_enabled = NULL; id = crm_element_value(state, XML_ATTR_ID); uname = crm_element_value(state, XML_ATTR_UNAME); this_node = pe_find_node_any(data_set->nodes, id, uname); if (uname == NULL) { /* error */ continue; } else if (this_node == NULL) { crm_config_warn("Node %s in status section no longer exists", uname); continue; } else if (is_remote_node(this_node)) { /* online state for remote nodes is determined by the * rsc state after all the unpacking is done. we do however * need to mark whether or not the node has been fenced as this plays * a role during unpacking cluster node resource state */ this_node->details->remote_was_fenced = crm_atoi(crm_element_value(state, XML_NODE_IS_FENCED), "0"); continue; } crm_trace("Processing node id=%s, uname=%s", id, uname); /* Mark the node as provisionally clean * - at least we have seen it in the current cluster's lifetime */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance"))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { crm_warn("ignoring %s attribute on node %s, disabling resource discovery is not allowed on cluster nodes", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } crm_trace("determining node state"); determine_online_status(state, this_node, data_set); if (is_not_set(data_set->flags, pe_flag_have_quorum) && this_node->details->online && (data_set->no_quorum_policy == no_quorum_suicide)) { /* Everything else should flow from this automatically * At least until the PE becomes able to migrate off healthy resources */ pe_fence_node(data_set, this_node, "cluster does not have quorum"); } } } while(unpack_node_loop(status, FALSE, data_set)) { crm_trace("Start another loop"); } // Now catch any nodes we didn't see unpack_node_loop(status, is_set(data_set->flags, pe_flag_stonith_enabled), data_set); for (GListPtr gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { node_t *this_node = gIter->data; if (this_node == NULL) { continue; } else if(is_remote_node(this_node) == FALSE) { continue; } else if(this_node->details->unpacked) { continue; } determine_remote_online_status(data_set, this_node); } return TRUE; } static gboolean determine_online_status_no_fencing(pe_working_set_t * data_set, xmlNode * node_state, node_t * this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); if (!crm_is_true(in_cluster)) { crm_trace("Node is down: in_cluster=%s", crm_str(in_cluster)); } else if (safe_str_eq(is_peer, ONLINESTATUS)) { if (safe_str_eq(join, CRMD_JOINSTATE_MEMBER)) { online = TRUE; } else { crm_debug("Node is not ready to run resources: %s", join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Controller is down: in_cluster=%s", crm_str(in_cluster)); crm_trace("\tis_peer=%s, join=%s, expected=%s", crm_str(is_peer), crm_str(join), crm_str(exp_state)); } else { /* mark it unclean */ pe_fence_node(data_set, this_node, "peer is unexpectedly down"); crm_info("\tin_cluster=%s, is_peer=%s, join=%s, expected=%s", crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state)); } return online; } static gboolean determine_online_status_fencing(pe_working_set_t * data_set, xmlNode * node_state, node_t * this_node) { gboolean online = FALSE; gboolean do_terminate = FALSE; bool crmd_online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); const char *terminate = pe_node_attribute_raw(this_node, "terminate"); /* - XML_NODE_IN_CLUSTER ::= true|false - XML_NODE_IS_PEER ::= online|offline - XML_NODE_JOIN_STATE ::= member|down|pending|banned - XML_NODE_EXPECTED ::= member|down */ if (crm_is_true(terminate)) { do_terminate = TRUE; } else if (terminate != NULL && strlen(terminate) > 0) { /* could be a time() value */ char t = terminate[0]; if (t != '0' && isdigit(t)) { do_terminate = TRUE; } } crm_trace("%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate); online = crm_is_true(in_cluster); crmd_online = safe_str_eq(is_peer, ONLINESTATUS); if (exp_state == NULL) { exp_state = CRMD_JOINSTATE_DOWN; } if (this_node->details->shutdown) { crm_debug("%s is shutting down", this_node->details->uname); /* Slightly different criteria since we can't shut down a dead peer */ online = crmd_online; } else if (in_cluster == NULL) { pe_fence_node(data_set, this_node, "peer has not been seen by the cluster"); } else if (safe_str_eq(join, CRMD_JOINSTATE_NACK)) { pe_fence_node(data_set, this_node, "peer failed the pacemaker membership criteria"); } else if (do_terminate == FALSE && safe_str_eq(exp_state, CRMD_JOINSTATE_DOWN)) { if (crm_is_true(in_cluster) || crmd_online) { crm_info("- Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", this_node->details->uname); } } else if (do_terminate && safe_str_eq(join, CRMD_JOINSTATE_DOWN) && crm_is_true(in_cluster) == FALSE && !crmd_online) { crm_info("Node %s was just shot", this_node->details->uname); online = FALSE; } else if (crm_is_true(in_cluster) == FALSE) { pe_fence_node(data_set, this_node, "peer is no longer part of the cluster"); } else if (!crmd_online) { pe_fence_node(data_set, this_node, "peer process is no longer available"); /* Everything is running at this point, now check join state */ } else if (do_terminate) { pe_fence_node(data_set, this_node, "termination was requested"); } else if (safe_str_eq(join, CRMD_JOINSTATE_MEMBER)) { crm_info("Node %s is active", this_node->details->uname); } else if (safe_str_eq(join, CRMD_JOINSTATE_PENDING) || safe_str_eq(join, CRMD_JOINSTATE_DOWN)) { crm_info("Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(data_set, this_node, "peer was in an unknown state"); crm_warn("%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate, this_node->details->shutdown); } return online; } static gboolean determine_remote_online_status(pe_working_set_t * data_set, node_t * this_node) { resource_t *rsc = this_node->details->remote_rsc; resource_t *container = NULL; pe_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; if (container && (g_list_length(rsc->running_on) == 1)) { host = rsc->running_on->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == RSC_ROLE_STARTED) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if (rsc->role == RSC_ROLE_STARTED && rsc->next_role == RSC_ROLE_STOPPED) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && is_set(container->flags, pe_rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if(is_set(rsc->flags, pe_rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if (rsc->role == RSC_ROLE_STOPPED || (container && container->role == RSC_ROLE_STOPPED)) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); return this_node->details->online; } gboolean determine_online_status(xmlNode * node_state, node_t * this_node, pe_working_set_t * data_set) { gboolean online = FALSE; const char *shutdown = NULL; const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); if (this_node == NULL) { crm_config_err("No node to check"); return online; } this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; shutdown = pe_node_attribute_raw(this_node, XML_CIB_ATTR_SHUTDOWN); if (shutdown != NULL && safe_str_neq("0", shutdown)) { this_node->details->shutdown = TRUE; } else if (safe_str_eq(exp_state, CRMD_JOINSTATE_MEMBER)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (is_set(data_set->flags, pe_flag_stonith_enabled) == FALSE) { online = determine_online_status_no_fencing(data_set, node_state, this_node); } else { online = determine_online_status_fencing(data_set, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (this_node->details->type == node_ping) { crm_info("Node %s is not a pacemaker node", this_node->details->uname); } else if (this_node->details->unclean) { pe_proc_warn("Node %s is unclean", this_node->details->uname); } else if (this_node->details->online) { crm_info("Node %s is %s", this_node->details->uname, this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("Node %s is offline", this_node->details->uname); } return online; } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!crm_strlen_zero(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = calloc(base_name_len + 3, sizeof(char)); CRM_ASSERT(zero); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static resource_t * create_fake_resource(const char *rsc_id, xmlNode * rsc_entry, pe_working_set_t * data_set) { resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (!common_unpack(xml_rsc, &rsc, NULL, data_set)) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(data_set->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set); } link_rsc2remotenode(data_set, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); set_bit(rsc->flags, pe_rsc_orphan_container_filler); } set_bit(rsc->flags, pe_rsc_orphan); data_set->resources = g_list_append(data_set->resources, rsc); return rsc; } extern resource_t *create_child_clone(resource_t * rsc, int sub_id, pe_working_set_t * data_set); static resource_t * find_anonymous_clone(pe_working_set_t * data_set, node_t * node, resource_t * parent, const char *rsc_id) { GListPtr rIter = NULL; resource_t *rsc = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pe_rsc_is_clone(parent)); CRM_ASSERT(is_not_set(parent->flags, pe_rsc_unique)); /* Find an instance active (or partially active for grouped clones) on the specified node */ pe_rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, node->details->uname, parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GListPtr nIter = NULL; GListPtr locations = NULL; resource_t *child = rIter->data; child->fns->location(child, &locations, TRUE); if (locations == NULL) { pe_rsc_trace(child, "Resource %s, skip inactive", child->id); continue; } for (nIter = locations; nIter && rsc == NULL; nIter = nIter->next) { node_t *childnode = nIter->data; if (childnode->details == node->details) { /* ->find_rsc() because we might be a cloned group */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); if(rsc) { pe_rsc_trace(rsc, "Resource %s, active", rsc->id); } } /* Keep this block, it means we'll do the right thing if * anyone toggles the unique flag to 'off' */ if (rsc && rsc->running_on) { crm_notice("/Anonymous/ clone %s is already running on %s", parent->id, node->details->uname); skip_inactive = TRUE; rsc = NULL; } } g_list_free(locations); } /* Find an inactive instance */ if (skip_inactive == FALSE) { pe_rsc_trace(parent, "Looking for %s anywhere", rsc_id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GListPtr locations = NULL; resource_t *child = rIter->data; if (is_set(child->flags, pe_rsc_block)) { pe_rsc_trace(child, "Skip: blocked in stopped state"); continue; } child->fns->location(child, &locations, TRUE); if (locations == NULL) { /* ->find_rsc() because we might be a cloned group */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); pe_rsc_trace(parent, "Resource %s, empty slot", rsc->id); } g_list_free(locations); } } if (rsc == NULL) { /* Create an extra orphan */ resource_t *top = create_child_clone(parent, -1, data_set); /* ->find_rsc() because we might be a cloned group */ rsc = top->fns->find_rsc(top, rsc_id, NULL, pe_find_clone); CRM_ASSERT(rsc != NULL); pe_rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, node->details->uname); } return rsc; } static resource_t * unpack_find_resource(pe_working_set_t * data_set, node_t * node, const char *rsc_id, xmlNode * rsc_entry) { resource_t *rsc = NULL; resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when clone-max=0, we create * a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id); if (clone0 && is_not_set(clone0->flags, pe_rsc_unique)) { rsc = clone0; } else { crm_trace("%s is not known as %s either", rsc_id, clone0_id); } /* Grab the parent clone even if this a different unique instance, * so we can remember the clone name, which will be the same. */ parent = uber_parent(clone0); free(clone0_id); crm_trace("%s not found: %s", rsc_id, parent ? parent->id : "orphan"); } else if (rsc->variant > pe_native) { crm_trace("%s is no longer a primitive resource, the lrm_resource entry is obsolete", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pe_rsc_is_anon_clone(parent)) { if (parent && parent->parent) { rsc = find_container_child(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(data_set, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && safe_str_neq(rsc_id, rsc->id) && safe_str_neq(rsc_id, rsc->clone_name)) { free(rsc->clone_name); rsc->clone_name = strdup(rsc_id); pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, node->details->uname, rsc->id, (is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : "")); } return rsc; } static resource_t * process_orphan_resource(xmlNode * rsc_entry, node_t * node, pe_working_set_t * data_set) { resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, node->details->uname); rsc = create_fake_resource(rsc_id, rsc_entry, data_set); if (is_set(data_set->flags, pe_flag_stop_rsc_orphans) == FALSE) { clear_bit(rsc->flags, pe_rsc_managed); } else { print_resource(LOG_TRACE, "Added orphan", rsc, FALSE); CRM_CHECK(rsc != NULL, return NULL); resource_location(rsc, NULL, -INFINITY, "__orphan_dont_run__", data_set); } return rsc; } static void process_rsc_state(resource_t * rsc, node_t * node, enum action_fail_response on_fail, xmlNode * migrate_op, pe_working_set_t * data_set) { node_t *tmpnode = NULL; char *reason = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, role2text(rsc->role), node->details->uname, fail2text(on_fail)); /* process current state */ if (rsc->role != RSC_ROLE_UNKNOWN) { resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { node_t *n = node_copy(node); pe_rsc_trace(rsc, "%s (aka. %s) known on %s", rsc->id, rsc->clone_name, n->details->uname); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (is_set(iter->flags, pe_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (rsc->role > RSC_ROLE_STOPPED && node->details->online == FALSE && node->details->maintenance == FALSE && is_set(rsc->flags, pe_rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (is_container_remote_node(node)) { set_bit(rsc->flags, pe_rsc_failed); should_fence = TRUE; } else if (is_set(data_set->flags, pe_flag_stonith_enabled)) { if (is_baremetal_remote_node(node) && node->details->remote_rsc && is_not_set(node->details->remote_rsc->flags, pe_rsc_failed)) { /* setting unseen = true means that fencing of the remote node will * only occur if the connection resource is not going to start somewhere. * This allows connection resources on a failed cluster-node to move to * another node without requiring the baremetal remote nodes to be fenced * as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(data_set, node, reason); } free(reason); } if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = action_fail_ignore; } switch (on_fail) { case action_fail_ignore: /* nothing to do */ break; case action_fail_fence: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(data_set, node, reason); free(reason); break; case action_fail_standby: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case action_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ clear_bit(rsc->flags, pe_rsc_managed); set_bit(rsc->flags, pe_rsc_block); break; case action_fail_migrate: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -INFINITY, "__action_migration_auto__", data_set); break; case action_fail_stop: rsc->next_role = RSC_ROLE_STOPPED; break; case action_fail_recover: if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { set_bit(rsc->flags, pe_rsc_failed); stop_action(rsc, node, FALSE); } break; case action_fail_restart_container: set_bit(rsc->flags, pe_rsc_failed); if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { stop_action(rsc, node, FALSE); } break; case action_fail_reset_remote: set_bit(rsc->flags, pe_rsc_failed); if (is_set(data_set->flags, pe_flag_stonith_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); } if (tmpnode && is_baremetal_remote_node(tmpnode) && tmpnode->details->remote_was_fenced == 0) { /* connection resource to baremetal resource failed in a way that * should result in fencing the remote-node. */ pe_fence_node(data_set, tmpnode, "remote connection is unrecoverable"); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > RSC_ROLE_STOPPED) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->remote_reconnect_ms) { rsc->next_role = RSC_ROLE_STOPPED; } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (is_set(rsc->flags, pe_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { if (is_set(rsc->flags, pe_rsc_orphan)) { if (is_set(rsc->flags, pe_rsc_managed)) { crm_config_warn("Detected active orphan %s running on %s", rsc->id, node->details->uname); } else { crm_config_warn("Cluster configured not to stop active orphans." " %s must be stopped manually on %s", rsc->id, node->details->uname); } } native_add_running(rsc, node, data_set); if (on_fail != action_fail_ignore) { set_bit(rsc->flags, pe_rsc_failed); } } else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pe_rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id); free(rsc->clone_name); rsc->clone_name = NULL; } else { char *key = stop_key(rsc); GListPtr possible_matches = find_actions(rsc->actions, key, node); GListPtr gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { action_t *stop = (action_t *) gIter->data; stop->flags |= pe_action_optional; } g_list_free(possible_matches); free(key); } } /* create active recurring operations as optional */ static void process_recurring(node_t * node, resource_t * rsc, int start_index, int stop_index, GListPtr sorted_op_list, pe_working_set_t * data_set) { int counter = -1; const char *task = NULL; const char *status = NULL; GListPtr gIter = sorted_op_list; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = ID(rsc_op); const char *interval_ms_s = NULL; counter++; if (node->details->online == FALSE) { pe_rsc_trace(rsc, "Skipping %s/%s: node is offline", rsc->id, node->details->uname); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pe_rsc_trace(rsc, "Skipping %s/%s: resource is not active", id, node->details->uname); continue; } else if (counter < start_index) { pe_rsc_trace(rsc, "Skipping %s/%s: old %d", id, node->details->uname, counter); continue; } interval_ms_s = crm_element_value(rsc_op, XML_LRM_ATTR_INTERVAL_MS); interval_ms = crm_parse_ms(interval_ms_s); if (interval_ms == 0) { pe_rsc_trace(rsc, "Skipping %s/%s: non-recurring", id, node->details->uname); continue; } status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (safe_str_eq(status, "-1")) { pe_rsc_trace(rsc, "Skipping %s/%s: status", id, node->details->uname); continue; } task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); /* create the action */ key = generate_op_key(rsc->id, task, interval_ms); pe_rsc_trace(rsc, "Creating %s/%s", key, node->details->uname); custom_action(rsc, key, task, node, TRUE, TRUE, data_set); } } void calculate_active_ops(GListPtr sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; GListPtr gIter = sorted_op_list; *stop_index = -1; *start_index = -1; for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (safe_str_eq(task, CRMD_ACTION_STOP) && safe_str_eq(status, "0")) { *stop_index = counter; } else if (safe_str_eq(task, CRMD_ACTION_START) || safe_str_eq(task, CRMD_ACTION_MIGRATED)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && safe_str_eq(task, CRMD_ACTION_STATUS)) { const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC); if (safe_str_eq(rc, "0") || safe_str_eq(rc, "8")) { implied_monitor_start = counter; } } else if (safe_str_eq(task, CRMD_ACTION_PROMOTE) || safe_str_eq(task, CRMD_ACTION_DEMOTE)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } static resource_t * unpack_lrm_rsc_state(node_t * node, xmlNode * rsc_entry, pe_working_set_t * data_set) { GListPtr gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = RSC_ROLE_UNKNOWN; const char *task = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); resource_t *rsc = NULL; GListPtr op_list = NULL; GListPtr sorted_op_list = NULL; xmlNode *migrate_op = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = FALSE; enum rsc_role_e saved_role = RSC_ROLE_UNKNOWN; crm_trace("[%s] Processing %s on %s", crm_element_name(rsc_entry), rsc_id, node->details->uname); /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = __xml_first_child(rsc_entry); rsc_op != NULL; rsc_op = __xml_next_element(rsc_op)) { if (crm_str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, TRUE)) { op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } /* find the resource */ rsc = unpack_find_resource(data_set, node, rsc_id, rsc_entry); if (rsc == NULL) { rsc = process_orphan_resource(rsc_entry, node, data_set); } CRM_ASSERT(rsc != NULL); /* process operations */ saved_role = rsc->role; on_fail = action_fail_ignore; rsc->role = RSC_ROLE_UNKNOWN; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); if (safe_str_eq(task, CRMD_ACTION_MIGRATED)) { migrate_op = rsc_op; } unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, data_set); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail, migrate_op, data_set); if (get_target_role(rsc, &req_role)) { if (rsc->next_role == RSC_ROLE_UNKNOWN || req_role < rsc->next_role) { pe_rsc_debug(rsc, "%s: Overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); rsc->next_role = req_role; } else if (req_role > rsc->next_role) { pe_rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(xmlNode * lrm_rsc_list, pe_working_set_t * data_set) { xmlNode *rsc_entry = NULL; for (rsc_entry = __xml_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = __xml_next_element(rsc_entry)) { resource_t *rsc; resource_t *container; const char *rsc_id; const char *container_id; if (safe_str_neq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE)) { continue; } container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER); rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(data_set->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL || is_set(rsc->flags, pe_rsc_orphan_container_filler) == FALSE || rsc->container != NULL) { continue; } pe_rsc_trace(rsc, "Mapped orphaned rsc %s's container to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } gboolean unpack_lrm_resources(node_t * node, xmlNode * lrm_rsc_list, pe_working_set_t * data_set) { xmlNode *rsc_entry = NULL; gboolean found_orphaned_container_filler = FALSE; CRM_CHECK(node != NULL, return FALSE); crm_trace("Unpacking resources on %s", node->details->uname); for (rsc_entry = __xml_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = __xml_next_element(rsc_entry)) { if (crm_str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, TRUE)) { resource_t *rsc = unpack_lrm_rsc_state(node, rsc_entry, data_set); if (!rsc) { continue; } if (is_set(rsc->flags, pe_rsc_orphan_container_filler)) { found_orphaned_container_filler = TRUE; } } } /* now that all the resource state has been unpacked for this node * we have to go back and map any orphaned container fillers to their * container resource */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(lrm_rsc_list, data_set); } return TRUE; } static void set_active(resource_t * rsc) { resource_t *top = uber_parent(rsc); if (top && is_set(top->flags, pe_rsc_promotable)) { rsc->role = RSC_ROLE_SLAVE; } else { rsc->role = RSC_ROLE_STARTED; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { node_t *node = value; int *score = user_data; node->weight = *score; } #define STATUS_PATH_MAX 1024 static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, pe_working_set_t * data_set) { int offset = 0; char xpath[STATUS_PATH_MAX]; offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//node_state[@uname='%s']", node); offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//" XML_LRM_TAG_RESOURCE "[@id='%s']", resource); /* Need to check against transition_magic too? */ if (source && safe_str_eq(op, CRMD_ACTION_MIGRATE)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_target='%s']", op, source); } else if (source && safe_str_eq(op, CRMD_ACTION_MIGRATED)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_source='%s']", op, source); } else { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s']", op); } CRM_LOG_ASSERT(offset > 0); return get_xpath_object(xpath, data_set->input, LOG_DEBUG); } static void unpack_rsc_migration(resource_t *rsc, node_t *node, xmlNode *xml_op, pe_working_set_t * data_set) { /* * The normal sequence is (now): migrate_to(Src) -> migrate_from(Tgt) -> stop(Src) * * So if a migrate_to is followed by a stop, then we don't need to care what * happened on the target node * * Without the stop, we need to look for a successful migrate_from. * This would also imply we're no longer running on the source * * Without the stop, and without a migrate_from op we make sure the resource * gets stopped on both source and target (assuming the target is up) * */ int stop_id = 0; int task_id = 0; xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, node->details->id, NULL, data_set); if (stop_op) { crm_element_value_int(stop_op, XML_LRM_ATTR_CALLID, &stop_id); } crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id); if (stop_op == NULL || stop_id < task_id) { int from_rc = 0, from_status = 0; const char *migrate_source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); node_t *target = pe_find_node(data_set->nodes, migrate_target); node_t *source = pe_find_node(data_set->nodes, migrate_source); xmlNode *migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, migrate_target, migrate_source, data_set); rsc->role = RSC_ROLE_STARTED; /* can be master? */ if (migrate_from) { crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc); crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS, &from_status); pe_rsc_trace(rsc, "%s op on %s exited with status=%d, rc=%d", ID(migrate_from), migrate_target, from_status, from_rc); } if (migrate_from && from_rc == PCMK_OCF_OK && from_status == PCMK_LRM_OP_DONE) { pe_rsc_trace(rsc, "Detected dangling migration op: %s on %s", ID(xml_op), migrate_source); /* all good * just need to arrange for the stop action to get sent * but _without_ affecting the target somehow */ rsc->role = RSC_ROLE_STOPPED; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } else if (migrate_from) { /* Failed */ if (target && target->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", migrate_target, target, target->details->online); native_add_running(rsc, target, data_set); } } else { /* Pending or complete but erased */ if (target && target->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", migrate_target, target, target->details->online); native_add_running(rsc, target, data_set); if (source && source->details->online) { /* If we make it here we have a partial migration. The migrate_to * has completed but the migrate_from on the target has not. Hold on * to the target and source on the resource. Later on if we detect that * the resource is still going to run on that target, we may continue * the migration */ rsc->partial_migration_target = target; rsc->partial_migration_source = source; } } else { /* Consider it failed here - forces a restart, prevents migration */ set_bit(rsc->flags, pe_rsc_failed); clear_bit(rsc->flags, pe_rsc_allow_migrate); } } } } static void unpack_rsc_migration_failure(resource_t *rsc, node_t *node, xmlNode *xml_op, pe_working_set_t * data_set) { const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); CRM_ASSERT(rsc); if (safe_str_eq(task, CRMD_ACTION_MIGRATED)) { int stop_id = 0; int migrate_id = 0; const char *migrate_source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, migrate_source, NULL, data_set); xmlNode *migrate_op = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATE, migrate_source, migrate_target, data_set); if (stop_op) { crm_element_value_int(stop_op, XML_LRM_ATTR_CALLID, &stop_id); } if (migrate_op) { crm_element_value_int(migrate_op, XML_LRM_ATTR_CALLID, &migrate_id); } /* Get our state right */ rsc->role = RSC_ROLE_STARTED; /* can be master? */ if (stop_op == NULL || stop_id < migrate_id) { node_t *source = pe_find_node(data_set->nodes, migrate_source); if (source && source->details->online) { native_add_running(rsc, source, data_set); } } } else if (safe_str_eq(task, CRMD_ACTION_MIGRATE)) { int stop_id = 0; int migrate_id = 0; const char *migrate_source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, migrate_target, NULL, data_set); xmlNode *migrate_op = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, migrate_target, migrate_source, data_set); if (stop_op) { crm_element_value_int(stop_op, XML_LRM_ATTR_CALLID, &stop_id); } if (migrate_op) { crm_element_value_int(migrate_op, XML_LRM_ATTR_CALLID, &migrate_id); } /* Get our state right */ rsc->role = RSC_ROLE_STARTED; /* can be master? */ if (stop_op == NULL || stop_id < migrate_id) { node_t *target = pe_find_node(data_set->nodes, migrate_target); pe_rsc_trace(rsc, "Stop: %p %d, Migrated: %p %d", stop_op, stop_id, migrate_op, migrate_id); if (target && target->details->online) { native_add_running(rsc, target, data_set); } } else if (migrate_op == NULL) { /* Make sure it gets cleaned up, the stop may pre-date the migrate_from */ rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } } } static void record_failed_op(xmlNode *op, node_t* node, resource_t *rsc, pe_working_set_t * data_set) { xmlNode *xIter = NULL; const char *op_key = crm_element_value(op, XML_LRM_ATTR_TASK_KEY); if (node->details->online == FALSE) { return; } for (xIter = data_set->failed->children; xIter; xIter = xIter->next) { const char *key = crm_element_value(xIter, XML_LRM_ATTR_TASK_KEY); const char *uname = crm_element_value(xIter, XML_ATTR_UNAME); if(safe_str_eq(op_key, key) && safe_str_eq(uname, node->details->uname)) { crm_trace("Skipping duplicate entry %s on %s", op_key, node->details->uname); return; } } crm_trace("Adding entry %s on %s", op_key, node->details->uname); crm_xml_add(op, XML_ATTR_UNAME, node->details->uname); crm_xml_add(op, XML_LRM_ATTR_RSCID, rsc->id); add_node_copy(data_set->failed, op); } static const char *get_op_key(xmlNode *xml_op) { const char *key = crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY); if(key == NULL) { key = ID(xml_op); } return key; } static void unpack_rsc_op_failure(resource_t * rsc, node_t * node, int rc, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = FALSE; action_t *action = NULL; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); CRM_ASSERT(rsc); *last_failure = xml_op; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && safe_str_eq(task, CRMD_ACTION_STATUS)) { is_probe = TRUE; pe_rsc_trace(rsc, "is a probe: %s", key); } if (rc != PCMK_OCF_NOT_INSTALLED || is_set(data_set->flags, pe_flag_symmetric_cluster)) { crm_warn("Processing failed op %s for %s on %s: %s (%d)", task, rsc->id, node->details->uname, services_ocf_exitcode_str(rc), rc); record_failed_op(xml_op, node, rsc, data_set); } else { crm_trace("Processing failed op %s for %s on %s: %s (%d)", task, rsc->id, node->details->uname, services_ocf_exitcode_str(rc), rc); } action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); if ((action->on_fail <= action_fail_fence && *on_fail < action->on_fail) || (action->on_fail == action_fail_reset_remote && *on_fail <= action_fail_recover) || (action->on_fail == action_fail_restart_container && *on_fail <= action_fail_recover) || (*on_fail == action_fail_restart_container && action->on_fail >= action_fail_migrate)) { pe_rsc_trace(rsc, "on-fail %s -> %s for %s (%s)", fail2text(*on_fail), fail2text(action->on_fail), action->uuid, key); *on_fail = action->on_fail; } if (safe_str_eq(task, CRMD_ACTION_STOP)) { resource_location(rsc, node, -INFINITY, "__stop_fail__", data_set); } else if (safe_str_eq(task, CRMD_ACTION_MIGRATE) || safe_str_eq(task, CRMD_ACTION_MIGRATED)) { unpack_rsc_migration_failure(rsc, node, xml_op, data_set); } else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_MASTER; } else if (safe_str_eq(task, CRMD_ACTION_DEMOTE)) { if (action->on_fail == action_fail_block) { rsc->role = RSC_ROLE_MASTER; rsc->next_role = RSC_ROLE_STOPPED; } else if(rc == PCMK_OCF_NOT_RUNNING) { rsc->role = RSC_ROLE_STOPPED; } else { /* * Staying in master role would put the PE/TE into a loop. Setting * slave role is not dangerous because the resource will be stopped * as part of recovery, and any master promotion will be ordered * after that stop. */ rsc->role = RSC_ROLE_SLAVE; } } if(is_probe && rc == PCMK_OCF_NOT_INSTALLED) { /* leave stopped */ pe_rsc_trace(rsc, "Leaving %s stopped", rsc->id); rsc->role = RSC_ROLE_STOPPED; } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "Setting %s active", rsc->id); set_active(rsc); } pe_rsc_trace(rsc, "Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s", rsc->id, role2text(rsc->role), node->details->unclean ? "true" : "false", fail2text(action->on_fail), role2text(action->fail_role)); if (action->fail_role != RSC_ROLE_STARTED && rsc->next_role < action->fail_role) { rsc->next_role = action->fail_role; } if (action->fail_role == RSC_ROLE_STOPPED) { int score = -INFINITY; resource_t *fail_rsc = rsc; if (fail_rsc->parent) { resource_t *parent = uber_parent(fail_rsc); if (pe_rsc_is_clone(parent) && is_not_set(parent->flags, pe_rsc_unique)) { /* For clone resources, if a child fails on an operation * with on-fail = stop, all the resources fail. Do this by preventing * the parent from coming up again. */ fail_rsc = parent; } } crm_warn("Making sure %s doesn't come up again", fail_rsc->id); /* make sure it doesn't come up again */ - g_hash_table_destroy(fail_rsc->allowed_nodes); + if (fail_rsc->allowed_nodes != NULL) { + g_hash_table_destroy(fail_rsc->allowed_nodes); + } fail_rsc->allowed_nodes = node_hash_from_list(data_set->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } pe_free_action(action); } static int determine_op_status( resource_t *rsc, int rc, int target_rc, node_t * node, xmlNode * xml_op, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; int result = PCMK_LRM_OP_DONE; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); bool is_probe = FALSE; CRM_ASSERT(rsc); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && safe_str_eq(task, CRMD_ACTION_STATUS)) { is_probe = TRUE; } if (target_rc >= 0 && target_rc != rc) { result = PCMK_LRM_OP_ERROR; pe_rsc_debug(rsc, "%s on %s returned '%s' (%d) instead of the expected value: '%s' (%d)", key, node->details->uname, services_ocf_exitcode_str(rc), rc, services_ocf_exitcode_str(target_rc), target_rc); } /* we could clean this up significantly except for old LRMs and CRMs that * didn't include target_rc and liked to remap status */ switch (rc) { case PCMK_OCF_OK: if (is_probe && target_rc == 7) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Operation %s found resource %s active on %s", task, rsc->id, node->details->uname); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || target_rc == rc || is_not_set(rsc->flags, pe_rsc_managed)) { result = PCMK_LRM_OP_DONE; rsc->role = RSC_ROLE_STOPPED; /* clear any previous failure actions */ *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; } else if (safe_str_neq(task, CRMD_ACTION_STOP)) { result = PCMK_LRM_OP_ERROR; } break; case PCMK_OCF_RUNNING_MASTER: if (is_probe) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Operation %s found resource %s active in master mode on %s", task, rsc->id, node->details->uname); } else if (target_rc == rc) { /* nothing to do */ } else if (target_rc >= 0) { result = PCMK_LRM_OP_ERROR; } rsc->role = RSC_ROLE_MASTER; break; case PCMK_OCF_DEGRADED_MASTER: case PCMK_OCF_FAILED_MASTER: rsc->role = RSC_ROLE_MASTER; result = PCMK_LRM_OP_ERROR; break; case PCMK_OCF_NOT_CONFIGURED: result = PCMK_LRM_OP_ERROR_FATAL; break; case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: case PCMK_OCF_UNIMPLEMENT_FEATURE: if (rc == PCMK_OCF_UNIMPLEMENT_FEATURE && (interval_ms > 0)) { result = PCMK_LRM_OP_NOTSUPPORTED; break; } else if (pe_can_fence(data_set, node) == FALSE && safe_str_eq(task, CRMD_ACTION_STOP)) { /* If a stop fails and we can't fence, there's nothing else we can do */ pe_proc_err("No further recovery can be attempted for %s: %s action failed with '%s' (%d)", rsc->id, task, services_ocf_exitcode_str(rc), rc); clear_bit(rsc->flags, pe_rsc_managed); set_bit(rsc->flags, pe_rsc_block); } result = PCMK_LRM_OP_ERROR_HARD; break; default: if (result == PCMK_LRM_OP_DONE) { crm_info("Treating %s (rc=%d) on %s as an ERROR", key, rc, node->details->uname); result = PCMK_LRM_OP_ERROR; } } return result; } static bool check_operation_expiry(resource_t *rsc, node_t *node, int rc, xmlNode *xml_op, pe_working_set_t * data_set) { bool expired = FALSE; time_t last_failure = 0; guint interval_ms = 0; int failure_timeout = rsc->failure_timeout; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *clear_reason = NULL; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); /* clearing recurring monitor operation failures automatically * needs to be carefully considered */ if ((interval_ms != 0) && safe_str_eq(task, "monitor")) { /* TODO, in the future we should consider not clearing recurring monitor * op failures unless the last action for a resource was a "stop" action. * otherwise it is possible that clearing the monitor failure will result * in the resource being in an undeterministic state. * * For now we handle this potential undeterministic condition for remote * node connection resources by not clearing a recurring monitor op failure * until after the node has been fenced. */ if (is_set(data_set->flags, pe_flag_stonith_enabled) && rsc->remote_reconnect_ms) { node_t *remote_node = pe_find_node(data_set->nodes, rsc->id); if (remote_node && remote_node->details->remote_was_fenced == 0) { if (strstr(ID(xml_op), "last_failure")) { crm_info("Waiting to clear monitor failure for remote node %s until fencing has occurred", rsc->id); } /* disabling failure timeout for this operation because we believe * fencing of the remote node should occur first. */ failure_timeout = 0; } } } if (failure_timeout > 0) { int last_run = 0; if (crm_element_value_int(xml_op, XML_RSC_OP_LAST_CHANGE, &last_run) == 0) { time_t now = get_effective_time(data_set); if (now > (last_run + failure_timeout)) { expired = TRUE; } } } if (expired) { if (failure_timeout > 0) { if (pe_get_failcount(node, rsc, &last_failure, pe_fc_default, xml_op, data_set)) { if (pe_get_failcount(node, rsc, &last_failure, pe_fc_effective, xml_op, data_set) == 0) { clear_reason = "it expired"; } else { expired = FALSE; } } else if (rsc->remote_reconnect_ms && strstr(ID(xml_op), "last_failure")) { /* always clear last failure when reconnect interval is set */ clear_reason = "reconnect interval is set"; } } } else if (strstr(ID(xml_op), "last_failure") && ((strcmp(task, "start") == 0) || (strcmp(task, "monitor") == 0))) { op_digest_cache_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, data_set); if (digest_data->rc == RSC_DIGEST_UNKNOWN) { crm_trace("rsc op %s/%s on node %s does not have a op digest to compare against", rsc->id, key, node->details->id); } else if(container_fix_remote_addr(rsc) && digest_data->rc != RSC_DIGEST_MATCH) { // We can't sanely check the changing 'addr' attribute. Yet crm_trace("Ignoring rsc op %s/%s on node %s", rsc->id, key, node->details->id); } else if (digest_data->rc != RSC_DIGEST_MATCH) { clear_reason = "resource parameters have changed"; } } if (clear_reason != NULL) { node_t *remote_node = pe_find_node(data_set->nodes, rsc->id); char *key = generate_op_key(rsc->id, CRM_OP_CLEAR_FAILCOUNT, 0); action_t *clear_op = custom_action(rsc, key, CRM_OP_CLEAR_FAILCOUNT, node, FALSE, TRUE, data_set); add_hash_param(clear_op->meta, XML_ATTR_TE_NOWAIT, XML_BOOLEAN_TRUE); crm_notice("Clearing failure of %s on %s because %s " CRM_XS " %s", rsc->id, node->details->uname, clear_reason, clear_op->uuid); if (is_set(data_set->flags, pe_flag_stonith_enabled) && rsc->remote_reconnect_ms && remote_node && remote_node->details->unclean) { action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, data_set); crm_notice("Waiting for %s to complete before clearing %s failure for remote node %s", fence?fence->uuid:"nil", task, rsc->id); order_actions(fence, clear_op, pe_order_implies_then); } } if (expired && (interval_ms == 0) && safe_str_eq(task, CRMD_ACTION_STATUS)) { switch(rc) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_MASTER: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_MASTER: /* Don't expire probes that return these values */ expired = FALSE; break; } } return expired; } int get_target_rc(xmlNode *xml_op) { int dummy = 0; int target_rc = 0; char *dummy_string = NULL; const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, &dummy_string, &dummy, &dummy, &target_rc); free(dummy_string); return target_rc; } static enum action_fail_response get_action_on_fail(resource_t *rsc, const char *key, const char *task, pe_working_set_t * data_set) { int result = action_fail_recover; action_t *action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); result = action->on_fail; pe_free_action(action); return result; } static void update_resource_state(resource_t * rsc, node_t * node, xmlNode * xml_op, const char * task, int rc, xmlNode * last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { gboolean clear_past_failure = FALSE; CRM_ASSERT(rsc); CRM_ASSERT(xml_op); if (rc == PCMK_OCF_NOT_RUNNING) { clear_past_failure = TRUE; } else if (rc == PCMK_OCF_NOT_INSTALLED) { rsc->role = RSC_ROLE_STOPPED; } else if (safe_str_eq(task, CRMD_ACTION_STATUS)) { if (last_failure) { const char *op_key = get_op_key(xml_op); const char *last_failure_key = get_op_key(last_failure); if (safe_str_eq(op_key, last_failure_key)) { clear_past_failure = TRUE; } } if (rsc->role < RSC_ROLE_STARTED) { set_active(rsc); } } else if (safe_str_eq(task, CRMD_ACTION_START)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (safe_str_eq(task, CRMD_ACTION_STOP)) { rsc->role = RSC_ROLE_STOPPED; clear_past_failure = TRUE; } else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_MASTER; clear_past_failure = TRUE; } else if (safe_str_eq(task, CRMD_ACTION_DEMOTE)) { /* Demote from Master does not clear an error */ rsc->role = RSC_ROLE_SLAVE; } else if (safe_str_eq(task, CRMD_ACTION_MIGRATED)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (safe_str_eq(task, CRMD_ACTION_MIGRATE)) { unpack_rsc_migration(rsc, node, xml_op, data_set); } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "%s active on %s", rsc->id, node->details->uname); set_active(rsc); } /* clear any previous failure actions */ if (clear_past_failure) { switch (*on_fail) { case action_fail_stop: case action_fail_fence: case action_fail_migrate: case action_fail_standby: pe_rsc_trace(rsc, "%s.%s is not cleared by a completed stop", rsc->id, fail2text(*on_fail)); break; case action_fail_block: case action_fail_ignore: case action_fail_recover: case action_fail_restart_container: *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; break; case action_fail_reset_remote: if (rsc->remote_reconnect_ms == 0) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; } break; } } } gboolean unpack_rsc_op(resource_t * rsc, node_t * node, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { int task_id = 0; const char *key = NULL; const char *task = NULL; const char *task_key = NULL; int rc = 0; int status = PCMK_LRM_OP_UNKNOWN; int target_rc = get_target_rc(xml_op); guint interval_ms = 0; gboolean expired = FALSE; resource_t *parent = rsc; enum action_fail_response failure_strategy = action_fail_recover; CRM_CHECK(rsc != NULL, return FALSE); CRM_CHECK(node != NULL, return FALSE); CRM_CHECK(xml_op != NULL, return FALSE); task_key = get_op_key(xml_op); task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY); crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id); crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); CRM_CHECK(task != NULL, return FALSE); CRM_CHECK(status <= PCMK_LRM_OP_NOT_INSTALLED, return FALSE); CRM_CHECK(status >= PCMK_LRM_OP_PENDING, return FALSE); if (safe_str_eq(task, CRMD_ACTION_NOTIFY) || safe_str_eq(task, CRMD_ACTION_METADATA)) { /* safe to ignore these */ return TRUE; } if (is_not_set(rsc->flags, pe_rsc_unique)) { parent = uber_parent(rsc); } pe_rsc_trace(rsc, "Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)", task_key, task, task_id, status, rc, node->details->uname, role2text(rsc->role)); if (node->details->unclean) { pe_rsc_trace(rsc, "Node %s (where %s is running) is unclean." " Further action depends on the value of the stop's on-fail attribute", node->details->uname, rsc->id); } if (status == PCMK_LRM_OP_ERROR) { /* Older versions set this if rc != 0 but it's up to us to decide */ status = PCMK_LRM_OP_DONE; } if(status != PCMK_LRM_OP_NOT_INSTALLED) { expired = check_operation_expiry(rsc, node, rc, xml_op, data_set); } /* Degraded results are informational only, re-map them to their error-free equivalents */ if (rc == PCMK_OCF_DEGRADED && safe_str_eq(task, CRMD_ACTION_STATUS)) { rc = PCMK_OCF_OK; /* Add them to the failed list to highlight them for the user */ if ((node->details->shutdown == FALSE) || (node->details->online == TRUE)) { crm_trace("Remapping %d to %d", PCMK_OCF_DEGRADED, PCMK_OCF_OK); record_failed_op(xml_op, node, rsc, data_set); } } else if (rc == PCMK_OCF_DEGRADED_MASTER && safe_str_eq(task, CRMD_ACTION_STATUS)) { rc = PCMK_OCF_RUNNING_MASTER; /* Add them to the failed list to highlight them for the user */ if ((node->details->shutdown == FALSE) || (node->details->online == TRUE)) { crm_trace("Remapping %d to %d", PCMK_OCF_DEGRADED_MASTER, PCMK_OCF_RUNNING_MASTER); record_failed_op(xml_op, node, rsc, data_set); } } if (expired && target_rc != rc) { const char *magic = crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC); pe_rsc_debug(rsc, "Expired operation '%s' on %s returned '%s' (%d) instead of the expected value: '%s' (%d)", key, node->details->uname, services_ocf_exitcode_str(rc), rc, services_ocf_exitcode_str(target_rc), target_rc); if (interval_ms == 0) { crm_notice("Ignoring expired calculated failure %s (rc=%d, magic=%s) on %s", task_key, rc, magic, node->details->uname); goto done; } else if(node->details->online && node->details->unclean == FALSE) { crm_notice("Re-initiated expired calculated failure %s (rc=%d, magic=%s) on %s", task_key, rc, magic, node->details->uname); /* This is SO horrible, but we don't have access to CancelXmlOp() yet */ crm_xml_add(xml_op, XML_LRM_ATTR_RESTART_DIGEST, "calculated-failure-timeout"); goto done; } } if(status == PCMK_LRM_OP_DONE || status == PCMK_LRM_OP_ERROR) { status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set); } pe_rsc_trace(rsc, "Handling status: %d", status); switch (status) { case PCMK_LRM_OP_CANCELLED: /* do nothing?? */ pe_err("Don't know what to do for cancelled ops yet"); break; case PCMK_LRM_OP_PENDING: if (safe_str_eq(task, CRMD_ACTION_START)) { set_bit(rsc->flags, pe_rsc_start_pending); set_active(rsc); } else if (safe_str_eq(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_MASTER; } else if (safe_str_eq(task, CRMD_ACTION_MIGRATE) && node->details->unclean) { /* If a pending migrate_to action is out on a unclean node, * we have to force the stop action on the target. */ const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); node_t *target = pe_find_node(data_set->nodes, migrate_target); if (target) { stop_action(rsc, target, FALSE); } } if (rsc->pending_task == NULL) { if (safe_str_eq(task, CRMD_ACTION_STATUS) && (interval_ms == 0)) { /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, uncomment this and the corresponding part of * native.c:native_pending_task(). */ /*rsc->pending_task = strdup("probe");*/ } else { rsc->pending_task = strdup(task); } } break; case PCMK_LRM_OP_DONE: pe_rsc_trace(rsc, "%s/%s completed on %s", rsc->id, task, node->details->uname); update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set); break; case PCMK_LRM_OP_NOT_INSTALLED: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if (failure_strategy == action_fail_ignore) { crm_warn("Cannot ignore failed %s (status=%d, rc=%d) on %s: " "Resource agent doesn't exist", task_key, status, rc, node->details->uname); /* Also for printing it as "FAILED" by marking it as pe_rsc_failed later */ *on_fail = action_fail_migrate; } resource_location(parent, node, -INFINITY, "hard-error", data_set); unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); break; case PCMK_LRM_OP_ERROR: case PCMK_LRM_OP_ERROR_HARD: case PCMK_LRM_OP_ERROR_FATAL: case PCMK_LRM_OP_TIMEOUT: case PCMK_LRM_OP_NOTSUPPORTED: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if ((failure_strategy == action_fail_ignore) || (failure_strategy == action_fail_restart_container && safe_str_eq(task, CRMD_ACTION_STOP))) { crm_warn("Pretending the failure of %s (rc=%d) on %s succeeded", task_key, rc, node->details->uname); update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set); crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname); set_bit(rsc->flags, pe_rsc_failure_ignored); record_failed_op(xml_op, node, rsc, data_set); if (failure_strategy == action_fail_restart_container && *on_fail <= action_fail_recover) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); if(status == PCMK_LRM_OP_ERROR_HARD) { do_crm_log(rc != PCMK_OCF_NOT_INSTALLED?LOG_ERR:LOG_NOTICE, "Preventing %s from re-starting on %s: operation %s failed '%s' (%d)", parent->id, node->details->uname, task, services_ocf_exitcode_str(rc), rc); resource_location(parent, node, -INFINITY, "hard-error", data_set); } else if(status == PCMK_LRM_OP_ERROR_FATAL) { crm_err("Preventing %s from re-starting anywhere: operation %s failed '%s' (%d)", parent->id, task, services_ocf_exitcode_str(rc), rc); resource_location(parent, NULL, -INFINITY, "fatal-error", data_set); } } break; } done: pe_rsc_trace(rsc, "Resource %s after %s: role=%s, next=%s", rsc->id, task, role2text(rsc->role), role2text(rsc->next_role)); return TRUE; } gboolean add_node_attrs(xmlNode * xml_obj, node_t * node, gboolean overwrite, pe_working_set_t * data_set) { const char *cluster_name = NULL; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_UNAME), strdup(node->details->uname)); g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID), strdup(node->details->id)); if (safe_str_eq(node->details->id, data_set->dc_uuid)) { data_set->dc_node = node; node->details->is_dc = TRUE; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE)); } else { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE)); } cluster_name = g_hash_table_lookup(data_set->config_hash, "cluster-name"); if (cluster_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME), strdup(cluster_name)); } unpack_instance_attributes(data_set->input, xml_obj, XML_TAG_ATTR_SETS, NULL, node->details->attrs, NULL, overwrite, data_set->now); if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) { const char *site_name = pe_node_attribute_raw(node, "site-name"); if (site_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(site_name)); } else if (cluster_name) { /* Default to cluster-name if unset */ g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(cluster_name)); } } return TRUE; } static GListPtr extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GListPtr gIter = NULL; GListPtr op_list = NULL; GListPtr sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = __xml_first_child(rsc_entry); rsc_op != NULL; rsc_op = __xml_next_element(rsc_op)) { if (crm_str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, TRUE)) { crm_xml_add(rsc_op, "resource", rsc); crm_xml_add(rsc_op, XML_ATTR_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", ID(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", ID(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GListPtr find_operations(const char *rsc, const char *node, gboolean active_filter, pe_working_set_t * data_set) { GListPtr output = NULL; GListPtr intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE); node_t *this_node = NULL; xmlNode *node_state = NULL; for (node_state = __xml_first_child(status); node_state != NULL; node_state = __xml_next_element(node_state)) { if (crm_str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, TRUE)) { const char *uname = crm_element_value(node_state, XML_ATTR_UNAME); if (node != NULL && safe_str_neq(uname, node)) { continue; } this_node = pe_find_node(data_set->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (is_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); } else { determine_online_status(node_state, this_node, data_set); } if (this_node->details->online || is_set(data_set->flags, pe_flag_stonith_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE); tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE); for (lrm_rsc = __xml_first_child(tmp); lrm_rsc != NULL; lrm_rsc = __xml_next_element(lrm_rsc)) { if (crm_str_eq((const char *)lrm_rsc->name, XML_LRM_TAG_RESOURCE, TRUE)) { const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_ID); if (rsc != NULL && safe_str_neq(rsc_id, rsc)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; }