Add a configuration command to reset all statistics

ldms/python/ldmsd/ldmsd_communicator.py

@@ -55,6 +55,7 @@
 import time
 import json
 import errno
+from pickle import NONE
 
 #:Dictionary contains the cmd_id, required attribute list
 #:and optional attribute list of each ldmsd commands. For example,
@@ -160,6 +161,7 @@
                       'metric_sets_default_authz': {'req_attr':[], 'opt_attr': ['uid', 'gid', 'perm']},
                       'set_sec_mod' : {'req_attr': ['regex'], 'opt_attr': ['uid', 'gid', 'perm']},
                       'log_status' : {'req_attr' : [], 'opt_attr' : ['name']},
+                      'stats_reset' : {'req_attr' : [], 'opt_attr' : ['list']},
                       ##### Failover. #####
                       'failover_config': {
                                 'req_attr': [
@@ -529,6 +531,7 @@ class LDMSD_Request(object):
     SET_DEFAULT_AUTHZ = 0x600 + 17
     SET_SEC_MOD = 0x600 + 19
     LOG_STATUS = 0x600 + 20
+    STATS_RESET = 0x600 + 21
 
     FAILOVER_CONFIG        = 0x700
     FAILOVER_PEERCFG_START = 0x700  +  1
@@ -1774,6 +1777,24 @@ def log_level(self, level, name = None, regex = None):
             self.close()
             return errno.ENOTCONN, str(e)
 
+    def stats_reset(self, s = None):
+        """
+        Reset the statistics counters
+        """
+        if s is not None and len(s) > 0:
+            attr_list = [LDMSD_Req_Attr(attr_id = LDMSD_Req_Attr.STRING, value = s)]
+        else:
+            attr_list = []
+        req = LDMSD_Request(command_id = LDMSD_Request.STATS_RESET, attrs = attr_list)
+
+        try:
+            req.send(self)
+            resp = req.receive(self)
+            return resp['errcode'], resp['msg']
+        except Exception as e:
+            self.close()
+            return errno.ENOTCONN, str(e)
+
     def getCfgCntr(self):
         req = LDMSD_Request(command_id=LDMSD_Request.CFG_CNTR)
         try:

ldms/python/ldmsd/ldmsd_controller

@@ -2728,6 +2728,26 @@ class LdmsdCmdParser(cmd.Cmd):
               "log level as the default logger (ldmsd). When the default log "
               "level changes, their log levels change accordingly.")
 
+    def complete_stats_reset(self, text, line, begidx, endidx):
+        return self.__complete_attr_list('stats_reset', text)
+
+    def do_stats_reset(self, arg):
+        """
+        Reset the statistics counters
+
+        Parameters:
+           [list=]   A comma-seprated list of statistics to be reset
+              thread - reset the thread statistics.
+              xprt   - reset the transport statistics.
+              update - reset the update time statistics and skipped and over-sampled counters.
+              store  - reset the store time statistics.
+              stream - reset the stream and stream client statistics
+        """
+        arg = self.handle_args('stats_reset', arg)
+        rc, msg = self.comm.stats_reset(s = arg['list'])
+        if rc:
+            print(f"Failed to reset the statistics")
+
     def do_option(self, arg):
         """
         ONLY SUPPORTED IN CONFIGURATION FILES

ldms/src/core/ldms.h

@@ -1339,7 +1339,8 @@ char *ldms_stream_stats_tq_to_str(struct ldms_stream_stats_tq_s *tq);
  *
  * \param  match    The stream name or a regular expression.
  * \param  is_regex 1 if \c match is a regular expression; otherwise, 0.
- * \param  is_reset 1 means to reset the streams' statistics
+ * \param  is_reset 0 means not to reset the statistics.
+ *                  A non-zero value means to reset the statistics.
  *
  * \retval str The string describing the stats.
  *
@@ -1388,6 +1389,11 @@ char *ldms_stream_client_stats_tq_to_str(struct ldms_stream_client_stats_tq_s *t
  */
 char *ldms_stream_client_stats_str(int is_reset);
 
+/**
+ * \brief Reset the statistics of streams and their clients
+ */
+void ldms_stream_n_client_stats_reset();
+
 /** \} */
 
 /**

ldms/src/core/ldms_stream.c

@@ -1739,6 +1739,62 @@ char *ldms_stream_stats_tq_to_str(struct ldms_stream_stats_tq_s *tq)
 	return ret;
 }
 
+void ldms_stream_n_client_stats_reset()
+{
+	struct rbn *rbn, *srbn;
+	struct ldms_stream_s *s;
+	struct ldms_stream_client_entry_s *sce;
+	struct ldms_stream_src_stats_s *src;
+	ldms_stream_client_t cli;
+
+	/*
+	 * There is a possibility of racing because the readlock is used.
+	 * However, the reset logic does not change the tree or list's structures.
+	 */
+	__STREAM_RDLOCK();
+
+	/* Reset regex clients first */
+	TAILQ_FOREACH(cli, &__regex_client_tq, entry) {
+		pthread_rwlock_rdlock(&cli->rwlock);
+		LDMS_STREAM_COUNTERS_INIT(&cli->tx);
+		LDMS_STREAM_COUNTERS_INIT(&cli->drops);
+		TAILQ_FOREACH(sce, &cli->stream_tq, client_stream_entry) {
+			LDMS_STREAM_COUNTERS_INIT(&sce->tx);
+			LDMS_STREAM_COUNTERS_INIT(&sce->drops);
+		}
+		pthread_rwlock_unlock(&cli->rwlock);
+	}
+
+	RBT_FOREACH(rbn, &__stream_rbt) {
+		s = container_of(rbn, struct ldms_stream_s, rbn);
+		pthread_rwlock_rdlock(&s->rwlock);
+
+		RBT_FOREACH(srbn, &s->src_stats_rbt) {
+			src = container_of(srbn, struct ldms_stream_src_stats_s, rbn);
+			LDMS_STREAM_COUNTERS_INIT(&src->rx);
+		}
+
+		TAILQ_FOREACH(sce, &s->client_tq, stream_client_entry) {
+			/* reset client's stats */
+			cli = sce->client;
+			if (!cli)
+				continue;
+			if (cli->is_regex)
+				continue; /* Already reset above */
+
+			LDMS_STREAM_COUNTERS_INIT(&sce->tx);
+			LDMS_STREAM_COUNTERS_INIT(&sce->drops);
+			LDMS_STREAM_COUNTERS_INIT(&cli->tx);
+			LDMS_STREAM_COUNTERS_INIT(&cli->drops);
+		}
+		LDMS_STREAM_COUNTERS_INIT(&s->rx);
+		pthread_rwlock_unlock(&s->rwlock);
+	}
+
+	__STREAM_UNLOCK();
+	return;
+}
+
 char *ldms_stream_stats_str(const char *match, int is_regex, int is_reset)
 {
 	struct ldms_stream_stats_tq_s *tq = NULL;

ldms/src/core/ldms_xprt.c

@@ -220,13 +220,15 @@ void ldms_xprt_rate_data(struct ldms_xprt_rate_data *data, int reset)
 	struct timespec now;
 	double dur_s;
 	(void)clock_gettime(CLOCK_REALTIME, &now);
-	dur_s = ldms_timespec_diff_s(&xprt_start, &now);
-	data->connect_rate_s = (double)xprt_connect_count / dur_s;
-	data->connect_request_rate_s = (double)xprt_connect_request_count / dur_s;
-	data->disconnect_rate_s = (double)xprt_disconnect_count / dur_s;
-	data->reject_rate_s = (double)xprt_reject_count / dur_s;
-	data->auth_fail_rate_s = (double)xprt_auth_fail_count / dur_s;
-	data->duration = dur_s;
+	if (data) {
+		dur_s = ldms_timespec_diff_s(&xprt_start, &now);
+		data->connect_rate_s = (double)xprt_connect_count / dur_s;
+		data->connect_request_rate_s = (double)xprt_connect_request_count / dur_s;
+		data->disconnect_rate_s = (double)xprt_disconnect_count / dur_s;
+		data->reject_rate_s = (double)xprt_reject_count / dur_s;
+		data->auth_fail_rate_s = (double)xprt_auth_fail_count / dur_s;
+		data->duration = dur_s;
+	}
 	if (reset) {
 		struct ldms_xprt *x;
 		pthread_mutex_lock(&xprt_list_lock);

ldms/src/ldmsd/ldmsd_request.c

@@ -262,6 +262,7 @@ static int prdcr_hint_tree_status_handler(ldmsd_req_ctxt_t reqc);
 static int update_time_stats_handler(ldmsd_req_ctxt_t reqc);
 static int set_sec_mod_handler(ldmsd_req_ctxt_t reqc);
 static int log_status_handler(ldmsd_req_ctxt_t reqc);
+static int stats_reset_handler(ldmsd_req_ctxt_t reqc);
 
 /* these are implemented in ldmsd_failover.c */
 int failover_config_handler(ldmsd_req_ctxt_t req_ctxt);
@@ -510,6 +511,9 @@ static struct request_handler_entry request_handler[] = {
 	[LDMSD_LOG_STATUS_REQ] = {
 		LDMSD_LOG_STATUS_REQ, log_status_handler, XUG
 	},
+	[LDMSD_STATS_RESET_REQ] = {
+		LDMSD_STATS_RESET_REQ, stats_reset_handler, XALL
+	},
 
 	/* Transport Stats Request */
 	[LDMSD_XPRT_STATS_REQ] = {
@@ -8788,3 +8792,84 @@ static int store_time_stats_handler(ldmsd_req_ctxt_t reqc)
 	json_entity_free(strgp_dict);
 	return rc;
 }
+
+static void __prdset_stats_reset(struct timespec *now, int is_update, int is_store)
+{
+	ldmsd_prdcr_t prdcr;
+	ldmsd_prdcr_set_t prdset;
+	struct rbn *rbn;
+
+	for (prdcr = ldmsd_prdcr_first(); prdcr; prdcr = ldmsd_prdcr_next(prdcr)) {
+		ldmsd_prdcr_lock(prdcr);
+		RBT_FOREACH(rbn, &prdcr->set_tree) {
+			prdset = container_of(rbn, struct ldmsd_prdcr_set, rbn);
+			if (is_update) {
+				memset(&prdset->updt_stat, 0, sizeof(struct ldmsd_stat));
+				prdset->updt_stat.start = prdset->store_stat.start = *now;
+				prdset->oversampled_cnt = prdset->skipped_upd_cnt = 0;
+			}
+			if (is_store)
+				memset(&prdset->store_stat, 0, sizeof(struct ldmsd_stat));
+		}
+		ldmsd_prdcr_unlock(prdcr);
+	}
+}
+
+static int stats_reset_handler(ldmsd_req_ctxt_t reqc)
+{
+	struct timespec now;
+	int rc = 0;
+	char *s;
+	char *tmp, *tok, *ptr;
+	int is_update;
+	int is_store;
+	int is_thread;
+	int is_xprt;
+	int is_stream;
+	is_update = is_store = is_thread = is_xprt = is_stream = 0;
+
+	s = ldmsd_req_attr_str_value_get_by_id(reqc, LDMSD_ATTR_STRING);
+	if (s) {
+		tmp = strdup(s);
+		if (!tmp) {
+			ovis_log(config_log, OVIS_LCRIT, "Memory allocation failure\n");
+			(void) Snprintf(&reqc->line_buf, &reqc->line_len, "Memory allocation failed.");
+			rc = ENOMEM;
+			goto out;
+		}
+
+		tok = strtok_r(tmp, ",", &ptr);
+		while (tok) {
+			if (0 == strcasecmp(tok, "update"))
+				is_update = 1;
+			else if (0 == strcasecmp(tok, "store"))
+				is_store = 1;
+			else if (0 == strcasecmp(tok, "thread"))
+				is_thread = 1;
+			else if (0 == strcasecmp(tok, "xprt"))
+				is_xprt = 1;
+			else if (0 == strcasecmp(tok, "stream"))
+				is_stream = 1;
+			tok = strtok_r(NULL, ",", &ptr);
+		}
+
+	} else {
+		is_update = is_store = is_thread = is_xprt = is_stream = 1;
+	}
+
+	clock_gettime(CLOCK_REALTIME, &now);
+	if (is_thread)
+		zap_thrstat_reset_all();
+
+	if (is_xprt)
+		ldms_xprt_rate_data(NULL, 1);
+
+	__prdset_stats_reset(&now, is_update, is_store);
+
+	if (is_stream)
+		ldms_stream_n_client_stats_reset();
+out:
+	free(s);
+	ldmsd_send_req_response(reqc, reqc->line_buf);
+	return rc;
+}

ldms/src/ldmsd/ldmsd_request.h

@@ -145,6 +145,7 @@ enum ldmsd_request {
 	LDMSD_CMDLINE_OPTIONS_SET_REQ,
 	LDMSD_SET_SEC_MOD_REQ,
 	LDMSD_LOG_STATUS_REQ,
+	LDMSD_STATS_RESET_REQ,
 
 	/* failover requests by user */
 	LDMSD_FAILOVER_CONFIG_REQ = 0x700, /* "failover_config" user command */