From d4f3ca33af3e40d31eb5479f6dbcb152cd9f984a Mon Sep 17 00:00:00 2001
From: "Costa Tsaousis (ktsaou)" <costa@tsaousis.gr>
Date: Sun, 20 Nov 2016 17:03:16 +0200
Subject: [PATCH] added backed alarms

---
 conf.d/Makefile.am           |  1 +
 conf.d/health.d/backend.conf | 44 ++++++++++++++++++++++++++++++++++++
 src/backends.c               |  6 ++---
 3 files changed, 48 insertions(+), 3 deletions(-)
 create mode 100644 conf.d/health.d/backend.conf

diff --git a/conf.d/Makefile.am b/conf.d/Makefile.am
index 690b5e7c..529dedbd 100644
--- a/conf.d/Makefile.am
+++ b/conf.d/Makefile.am
@@ -52,6 +52,7 @@ dist_pythonconfig_DATA = \
 healthconfigdir=$(configdir)/health.d
 dist_healthconfig_DATA = \
     health.d/apache.conf \
+    health.d/backend.conf \
     health.d/cpu.conf \
     health.d/disks.conf \
     health.d/entropy.conf \
diff --git a/conf.d/health.d/backend.conf b/conf.d/health.d/backend.conf
new file mode 100644
index 00000000..ba40a5f1
--- /dev/null
+++ b/conf.d/health.d/backend.conf
@@ -0,0 +1,44 @@
+
+# make sure we are sending data to backend
+
+   alarm: backend_last_buffering
+      on: netdata.backend_metrics
+    calc: $now - $last_collected_t
+   units: seconds ago
+   every: 10s
+    warn: $this > (($status >= $WARNING)  ? ($update_every) : ( 5 * $update_every))
+    crit: $this > (($status == $CRITICAL) ? ($update_every) : (60 * $update_every))
+   delay: down 5m multiplier 1.5 max 1h
+    info: number of seconds since the last successful buffering of backend data
+      to: dba
+
+   alarm: backend_metrics_sent
+      on: netdata.backend_metrics
+   units: %
+    calc: $sent * 100 / $buffered
+   every: 10s
+    warn: $this != 100
+   delay: down 5m multiplier 1.5 max 1h
+    info: percentage of metrics sent to the backend server
+      to: dba
+
+   alarm: backend_metrics_lost
+      on: netdata.backend_metrics
+   units: metrics
+    calc: $lost
+   every: 10s
+    crit: $this != 0
+   delay: down 5m multiplier 1.5 max 1h
+    info: number of metrics lost due to repeating failures to contact the backend server
+      to: dba
+
+   alarm: backend_slow
+      on: netdata.backend_latency
+   units: %
+    calc: $latency * 100 / ($update_every * 1000)
+   every: 10s
+    warn: $this > 50
+    crit: $this > 100
+   delay: down 5m multiplier 1.5 max 1h
+    info: the percentage of time between iterations needed by the backend time to process the data sent by netdata
+      to: dba
diff --git a/src/backends.c b/src/backends.c
index 75395973..5f330342 100644
--- a/src/backends.c
+++ b/src/backends.c
@@ -164,7 +164,7 @@ static inline int connect_to_one(const char *definition, int default_port, struc
     return fd;
 }
 
-static inline calculated_number backend_duration_average(RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
+static inline calculated_number backend_calculate_value_from_stored_data(RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
     time_t first_t = rrdset_first_entry_t(st);
     time_t last_t = rrdset_last_entry_t(st);
 
@@ -227,7 +227,7 @@ static inline int format_dimension_collected_graphite_plaintext(BUFFER *b, const
 
 static inline int format_dimension_stored_graphite_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
     (void)host;
-    calculated_number value = backend_duration_average(st, rd, after, before, options);
+    calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
     if(!isnan(value)) {
         buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, hostname, st->id, rd->id, value, (uint32_t) before);
         return 1;
@@ -246,7 +246,7 @@ static inline int format_dimension_collected_opentsdb_telnet(BUFFER *b, const ch
 
 static inline int format_dimension_stored_opentsdb_telnet(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
     (void)host;
-    calculated_number value = backend_duration_average(st, rd, after, before, options);
+    calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
     if(!isnan(value)) {
         buffer_sprintf(b, "put %s.%s.%s %u " CALCULATED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t) before, value, hostname);
         return 1;
-- 
2.39.2