Adding upstream version 1.44.3.upstream/1.44.3

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

1 files changed, 125 insertions, 59 deletions

diff --git a/daemon/service.c b/daemon/service.c
index f7fe86e0..8a65de66 100644
--- a/daemon/service.c
+++ b/daemon/service.c
@@ -76,33 +76,48 @@ static void svc_rrddim_obsolete_to_archive(RRDDIM *rd) {
     rrddim_free(st, rd);
 }
 
-static bool svc_rrdset_archive_obsolete_dimensions(RRDSET *st, bool all_dimensions) {
+static inline bool svc_rrdset_archive_obsolete_dimensions(RRDSET *st, bool all_dimensions) {
+    if(!all_dimensions && !rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS))
+        return true;
+
     worker_is_busy(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS);
 
+    rrdset_flag_clear(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
+
     RRDDIM *rd;
     time_t now = now_realtime_sec();
 
-    bool done_all_dimensions = true;
+    size_t dim_candidates = 0;
+    size_t dim_archives = 0;
 
     dfe_start_write(st->rrddim_root_index, rd) {
-        if(unlikely(
-                all_dimensions ||
-                (rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE) && (rd->collector.last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now))
-                    )) {
-
-            if(dictionary_acquired_item_references(rd_dfe.item) == 1) {
-                netdata_log_info("Removing obsolete dimension '%s' (%s) of '%s' (%s).", rrddim_name(rd), rrddim_id(rd), rrdset_name(st), rrdset_id(st));
-                svc_rrddim_obsolete_to_archive(rd);
+        bool candidate = (all_dimensions || rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE));
+
+        if(candidate) {
+            dim_candidates++;
+
+            if(rd->collector.last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now) {
+                size_t references = dictionary_acquired_item_references(rd_dfe.item);
+                if(references == 1) {
+//                    netdata_log_info("Removing obsolete dimension 'host:%s/chart:%s/dim:%s'",
+//                                     rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
+                    svc_rrddim_obsolete_to_archive(rd);
+                    dim_archives++;
+                }
+//                else
+//                    netdata_log_info("Cannot remove obsolete dimension 'host:%s/chart:%s/dim:%s'",
+//                            rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
             }
-            else
-                done_all_dimensions = false;
         }
-        else
-            done_all_dimensions = false;
     }
     dfe_done(rd);
 
-    return done_all_dimensions;
+    if(dim_archives != dim_candidates) {
+        rrdset_flag_set(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
+        return false;
+    }
+
+    return true;
 }
 
 static void svc_rrdset_obsolete_to_free(RRDSET *st) {
@@ -132,50 +147,78 @@ static void svc_rrdset_obsolete_to_free(RRDSET *st) {
     rrdset_free(st);
 }
 
-static void svc_rrdhost_cleanup_obsolete_charts(RRDHOST *host) {
+static inline void svc_rrdhost_cleanup_charts_marked_obsolete(RRDHOST *host) {
+    if(!rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS))
+        return;
+
     worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS);
 
+    rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
+
+    size_t full_candidates = 0;
+    size_t full_archives = 0;
+    size_t partial_candidates = 0;
+    size_t partial_archives = 0;
+
     time_t now = now_realtime_sec();
     RRDSET *st;
     rrdset_foreach_reentrant(st, host) {
         if(rrdset_is_replicating(st))
             continue;
 
-        if(unlikely(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE)
-                     && st->last_accessed_time_s + rrdset_free_obsolete_time_s < now
-                     && st->last_updated.tv_sec + rrdset_free_obsolete_time_s < now
-                     && st->last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now
-                     )) {
-            svc_rrdset_obsolete_to_free(st);
-        }
-        else if(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS)) {
-            rrdset_flag_clear(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
-            svc_rrdset_archive_obsolete_dimensions(st, false);
+        RRDSET_FLAGS flags = rrdset_flag_get(st);
+        bool obsolete_chart = flags & RRDSET_FLAG_OBSOLETE;
+        bool obsolete_dims = flags & RRDSET_FLAG_OBSOLETE_DIMENSIONS;
+
+        if(obsolete_dims) {
+            partial_candidates++;
+
+            if(svc_rrdset_archive_obsolete_dimensions(st, false))
+                partial_archives++;
         }
-        else if (unlikely(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE))) {
-            rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS);
+
+        if(obsolete_chart) {
+            full_candidates++;
+
+            if(unlikely(   st->last_accessed_time_s + rrdset_free_obsolete_time_s < now
+                        && st->last_updated.tv_sec + rrdset_free_obsolete_time_s < now
+                        && st->last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now
+                       )) {
+                svc_rrdset_obsolete_to_free(st);
+                full_archives++;
+            }
         }
     }
     rrdset_foreach_done(st);
+
+    if(partial_archives != partial_candidates)
+        rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
+
+    if(full_archives != full_candidates)
+        rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS);
 }
 
-static void svc_rrdset_check_obsoletion(RRDHOST *host) {
+static void svc_rrdhost_detect_obsolete_charts(RRDHOST *host) {
     worker_is_busy(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK);
 
     time_t now = now_realtime_sec();
     time_t last_entry_t;
     RRDSET *st;
+
+    time_t child_connect_time = host->child_connect_time;
+
     rrdset_foreach_read(st, host) {
         if(rrdset_is_replicating(st))
             continue;
 
         last_entry_t = rrdset_last_entry_s(st);
 
-        if(last_entry_t && last_entry_t < host->child_connect_time &&
-           host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT + ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT * st->update_every
-             < now)
+        if (last_entry_t && last_entry_t < child_connect_time &&
+            child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT +
+                    (ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT * st->update_every) <
+                now)
 
-            rrdset_is_obsolete(st);
+            rrdset_is_obsolete___safe_from_collector_thread(st);
     }
     rrdset_foreach_done(st);
 }
@@ -190,24 +233,24 @@ static void svc_rrd_cleanup_obsolete_charts_from_all_hosts() {
         if(rrdhost_receiver_replicating_charts(host) || rrdhost_sender_replicating_charts(host))
             continue;
 
-        if(rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS)) {
-            rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
-            svc_rrdhost_cleanup_obsolete_charts(host);
-        }
+        svc_rrdhost_cleanup_charts_marked_obsolete(host);
 
-        if(host != localhost
-            && host->trigger_chart_obsoletion_check
-            && (
-                   (
-                    host->child_last_chart_command
-                 && host->child_last_chart_command + host->health.health_delay_up_to < now_realtime_sec()
-                   )
-                || (host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT < now_realtime_sec())
-                )
-            ) {
-            svc_rrdset_check_obsoletion(host);
+        if (host == localhost)
+            continue;
+
+        netdata_mutex_lock(&host->receiver_lock);
+
+        time_t now = now_realtime_sec();
+
+        if (host->trigger_chart_obsoletion_check &&
+            ((host->child_last_chart_command &&
+              host->child_last_chart_command + host->health.health_delay_up_to < now) ||
+             (host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT < now))) {
+            svc_rrdhost_detect_obsolete_charts(host);
             host->trigger_chart_obsoletion_check = 0;
         }
+
+        netdata_mutex_unlock(&host->receiver_lock);
     }
 
     rrd_unlock();
@@ -226,22 +269,45 @@ restart_after_removal:
         if(!rrdhost_should_be_removed(host, protected_host, now))
             continue;
 
-        netdata_log_info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", rrdhost_hostname(host), host->machine_guid);
+        bool is_archived = rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED);
+        if (!is_archived) {
+            netdata_log_info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", rrdhost_hostname(host), host->machine_guid);
 
-        if (rrdhost_option_check(host, RRDHOST_OPTION_DELETE_ORPHAN_HOST)
-            /* don't delete multi-host DB host files */
-            && !(host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE && is_storage_engine_shared(host->db[0].instance))
-        ) {
-            worker_is_busy(WORKER_JOB_DELETE_HOST_CHARTS);
-            rrdhost_delete_charts(host);
+            if (rrdhost_option_check(host, RRDHOST_OPTION_DELETE_ORPHAN_HOST)
+                /* don't delete multi-host DB host files */
+                && !(host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE && is_storage_engine_shared(host->db[0].instance))
+            ) {
+                worker_is_busy(WORKER_JOB_DELETE_HOST_CHARTS);
+                rrdhost_delete_charts(host);
+            }
+            else {
+                worker_is_busy(WORKER_JOB_SAVE_HOST_CHARTS);
+                rrdhost_save_charts(host);
+            }
         }
-        else {
-            worker_is_busy(WORKER_JOB_SAVE_HOST_CHARTS);
-            rrdhost_save_charts(host);
+
+        bool force = false;
+
+        if (rrdhost_option_check(host, RRDHOST_OPTION_EPHEMERAL_HOST) && now - host->last_connected > rrdhost_free_ephemeral_time_s)
+            force = true;
+
+        if (!force && is_archived)
+            continue;
+
+       if (force) {
+            netdata_log_info("Host '%s' with machine guid '%s' is archived, ephemeral clean up.", rrdhost_hostname(host), host->machine_guid);
         }
 
         worker_is_busy(WORKER_JOB_FREE_HOST);
-        rrdhost_free___while_having_rrd_wrlock(host, false);
+#ifdef ENABLE_ACLK
+        // in case we have cloud connection we inform cloud
+        // a child disconnected
+        if (netdata_cloud_enabled && force) {
+            aclk_host_state_update(host, 0, 0);
+            unregister_node(host->machine_guid);
+        }
+#endif
+        rrdhost_free___while_having_rrd_wrlock(host, force);
         goto restart_after_removal;
     }