Merging upstream version 1.16.0.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 149 insertions, 32 deletions

diff --git a/health/health.c b/health/health.c
index f92a1ba6b..55bd72843 100644
--- a/health/health.c
+++ b/health/health.c
@@ -13,18 +13,74 @@ unsigned int default_health_enabled = 1;
 // ----------------------------------------------------------------------------
 // health initialization
 
+/**
+ * User Config directory
+ *
+ * Get the config directory for health and return it.
+ *
+ * @return a pointer to the user config directory
+ */
 inline char *health_user_config_dir(void) {
     char buffer[FILENAME_MAX + 1];
     snprintfz(buffer, FILENAME_MAX, "%s/health.d", netdata_configured_user_config_dir);
     return config_get(CONFIG_SECTION_HEALTH, "health configuration directory", buffer);
 }
 
+/**
+ * Stock Config Directory
+ *
+ * Get the Stock config directory and return it.
+ *
+ * @return a pointer to the stock config directory.
+ */
 inline char *health_stock_config_dir(void) {
     char buffer[FILENAME_MAX + 1];
     snprintfz(buffer, FILENAME_MAX, "%s/health.d", netdata_configured_stock_config_dir);
     return config_get(CONFIG_SECTION_HEALTH, "stock health configuration directory", buffer);
 }
 
+/**
+ * Silencers init
+ *
+ * Function used to initialize the silencer structure.
+ */
+void health_silencers_init(void) {
+    struct stat statbuf;
+    if (!stat(silencers_filename,&statbuf)) {
+        off_t length = statbuf.st_size;
+        if (length && length < HEALTH_SILENCERS_MAX_FILE_LEN) {
+            FILE *fd = fopen(silencers_filename, "r");
+            if (fd) {
+                char *str = mallocz((length+1)* sizeof(char));
+                if(str) {
+                    size_t copied;
+                    copied = fread(str, sizeof(char), length, fd);
+                    if (copied == (length* sizeof(char))) {
+                        str[length] = 0x00;
+                        json_parse(str, NULL, health_silencers_json_read_callback);
+                        info("Parsed health silencers file %s", silencers_filename);
+                    } else {
+                        error("Cannot read the data from health silencers file %s", silencers_filename);
+                    }
+                    freez(str);
+                }
+                fclose(fd);
+            } else {
+                error("Cannot open the file %s",silencers_filename);
+            }
+        } else {
+            error("Health silencers file %s has the size %ld that is out of range[ 1 , %d ]. Aborting read.", silencers_filename, length, HEALTH_SILENCERS_MAX_FILE_LEN);
+        }
+    } else {
+        error("Cannot open the file %s",silencers_filename);
+    }
+}
+
+/**
+ * Health Init
+ *
+ * Initialize the health thread.
+ */
 void health_init(void) {
     debug(D_HEALTH, "Health configuration initializing");
 
@@ -32,11 +88,20 @@ void health_init(void) {
         debug(D_HEALTH, "Health is disabled.");
         return;
     }
+
+    health_silencers_init();
 }
 
 // ----------------------------------------------------------------------------
 // re-load health configuration
 
+/**
+ * Reload host
+ *
+ * Reload configuration for a specific host.
+ *
+ * @param host the structure of the host that the function will reload the configuration.
+ */
 void health_reload_host(RRDHOST *host) {
     if(unlikely(!host->health_enabled))
         return;
@@ -84,6 +149,11 @@ void health_reload_host(RRDHOST *host) {
     rrdhost_unlock(host);
 }
 
+/**
+ * Reload
+ *
+ * Reload the host configuration for all hosts.
+ */
 void health_reload(void) {
 
     rrd_rdlock();
@@ -255,17 +325,18 @@ static inline void health_alarm_log_process(RRDHOST *host) {
     netdata_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
 
     ALARM_ENTRY *ae;
-    for(ae = host->health_log.alarms; ae && ae->unique_id >= host->health_last_processed_id ; ae = ae->next) {
-        if(unlikely(
-            !(ae->flags & HEALTH_ENTRY_FLAG_PROCESSED) &&
-            !(ae->flags & HEALTH_ENTRY_FLAG_UPDATED)
+    for(ae = host->health_log.alarms; ae && ae->unique_id >= host->health_last_processed_id; ae = ae->next) {
+        if(likely(!alarm_entry_isrepeating(host, ae))) {
+            if(unlikely(
+                    !(ae->flags & HEALTH_ENTRY_FLAG_PROCESSED) &&
+                    !(ae->flags & HEALTH_ENTRY_FLAG_UPDATED)
             )) {
+                if(unlikely(ae->unique_id < first_waiting))
+                    first_waiting = ae->unique_id;
 
-            if(unlikely(ae->unique_id < first_waiting))
-                first_waiting = ae->unique_id;
-
-            if(likely(now >= ae->delay_up_to_timestamp))
-                health_process_notifications(host, ae);
+                if(likely(now >= ae->delay_up_to_timestamp))
+                    health_process_notifications(host, ae);
+            }
         }
     }
 
@@ -294,10 +365,12 @@ static inline void health_alarm_log_process(RRDHOST *host) {
 
         ALARM_ENTRY *t = ae->next;
 
-        health_alarm_log_free_one_nochecks_nounlink(ae);
+        if(likely(!alarm_entry_isrepeating(host, ae))) {
+            health_alarm_log_free_one_nochecks_nounlink(ae);
+            host->health_log.count--;
+        }
 
         ae = t;
-        host->health_log.count--;
     }
 
     netdata_rwlock_unlock(&host->health_log.alarm_log_rwlock);
@@ -411,7 +484,7 @@ SILENCE_TYPE check_silenced(RRDCALC *rc, char* host, SILENCERS *silencers) {
                 debug(D_HEALTH, "Alarm %s matched a silence entry, but no SILENCE or DISABLE command was issued via the command API. The match has no effect.", rc->name);
             } else {
                 debug(D_HEALTH, "Alarm %s via the command API - name:%s context:%s chart:%s host:%s family:%s"
-                		, (silencers->stype==STYPE_DISABLE_ALARMS)?"Disabled":"Silenced"
+                        , (silencers->stype == STYPE_DISABLE_ALARMS)?"Disabled":"Silenced"
                         , rc->name
                         , (rc->rrdset)?rc->rrdset->context:""
                         , rc->chart
@@ -425,6 +498,16 @@ SILENCE_TYPE check_silenced(RRDCALC *rc, char* host, SILENCERS *silencers) {
     return STYPE_NONE;
 }
 
+/**
+ * Update Disabled Silenced
+ *
+ * Update the variable rrdcalc_flags of the structure RRDCALC according with the values of the host structure
+ *
+ * @param host structure that contains information about the host monitored.
+ * @param rc structure with information about the alarm
+ *
+ * @return It returns 1 case rrdcalc_flags is DISABLED or 0 otherwise
+ */
 int update_disabled_silenced(RRDHOST *host, RRDCALC *rc) {
 	uint32_t rrdcalc_flags_old = rc->rrdcalc_flags;
 	// Clear the flags
@@ -454,6 +537,15 @@ int update_disabled_silenced(RRDHOST *host, RRDCALC *rc) {
 		return 0;
 }
 
+/**
+ * Health Main
+ *
+ * The main thread of the health system. In this function all the alarms will be processed.
+ *
+ * @param ptr is a pointer to the netdata_static_thread structure.
+ *
+ * @return It always returns NULL
+ */
 void *health_main(void *ptr) {
     netdata_thread_cleanup_push(health_main_cleanup, ptr);
 
@@ -464,12 +556,6 @@ void *health_main(void *ptr) {
     time_t hibernation_delay  = config_get_number(CONFIG_SECTION_HEALTH, "postpone alarms during hibernation for seconds", 60);
 
     unsigned int loop = 0;
-
-    silencers =  mallocz(sizeof(SILENCERS));
-    silencers->all_alarms=0;
-    silencers->stype=STYPE_NONE;
-    silencers->silencers=NULL;
-
     while(!netdata_exit) {
 		loop++;
 		debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
@@ -756,20 +842,22 @@ void *health_main(void *ptr) {
 						rc->delay_last = delay;
 						rc->delay_up_to_timestamp = now + delay;
 
-						health_alarm_log(
-								host, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id,
-								rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change,
-								rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info,
-								rc->delay_last,
-								(
-								    ((rc->options & RRDCALC_FLAG_NO_CLEAR_NOTIFICATION)? HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION : 0) |
-								    ((rc->rrdcalc_flags & RRDCALC_FLAG_SILENCED)? HEALTH_ENTRY_FLAG_SILENCED : 0)
-								)
-
-						);
-
-						rc->last_status_change = now;
-						rc->status = status;
+                        if(likely(!rrdcalc_isrepeating(rc))) {
+                            ALARM_ENTRY *ae = health_create_alarm_entry(
+                                    host, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id,
+                                    rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change,
+                                    rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info,
+                                    rc->delay_last,
+                                    (
+                                            ((rc->options & RRDCALC_FLAG_NO_CLEAR_NOTIFICATION)? HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION : 0) |
+                                            ((rc->rrdcalc_flags & RRDCALC_FLAG_SILENCED)? HEALTH_ENTRY_FLAG_SILENCED : 0)
+                                    )
+                            );
+                            health_alarm_log(host, ae);
+                        }
+                        rc->last_status_change = now;
+                        rc->old_status = rc->status;
+                        rc->status = status;
 					}
 
 					rc->last_updated = now;
@@ -779,6 +867,35 @@ void *health_main(void *ptr) {
 						next_run = rc->next_update;
 				}
 
+                // process repeating alarms
+                RRDCALC *rc;
+                for(rc = host->alarms; rc ; rc = rc->next) {
+                    int repeat_every = 0;
+                    if(unlikely(rrdcalc_isrepeating(rc))) {
+                        if(unlikely(rc->status == RRDCALC_STATUS_WARNING))
+                            repeat_every = rc->warn_repeat_every;
+                        else if(unlikely(rc->status == RRDCALC_STATUS_CRITICAL))
+                            repeat_every = rc->crit_repeat_every;
+                    }
+                    if(unlikely(repeat_every > 0 && (rc->last_repeat + repeat_every) <= now)) {
+                        rc->last_repeat = now;
+                        ALARM_ENTRY *ae = health_create_alarm_entry(
+                                host, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id,
+                                rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change,
+                                rc->old_value, rc->value, rc->old_status, rc->status, rc->source, rc->units, rc->info,
+                                rc->delay_last,
+                                (
+                                        ((rc->options & RRDCALC_FLAG_NO_CLEAR_NOTIFICATION)? HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION : 0) |
+                                        ((rc->rrdcalc_flags & RRDCALC_FLAG_SILENCED)? HEALTH_ENTRY_FLAG_SILENCED : 0)
+                                )
+                        );
+                        ae->last_repeat = rc->last_repeat;
+                        health_process_notifications(host, ae);
+                        debug(D_HEALTH, "Notification sent for the repeating alarm %u.", ae->alarm_id);
+                        health_alarm_log_free_one_nochecks_nounlink(ae);
+                    }
+                }
+
 				rrdhost_unlock(host);
 			}