1 # other rules to consider:
2 # filesystem, network, ntp rules:
3 # https://github.com/cloudalchemy/ansible-prometheus defaults/main.yml
4 # on my system, the interpolated values are in /a/opt/ansible-prometheus/rules.yml
12 # ## uncomment for testing an alert firing
13 # - alert: test-alert4
19 # description: "always-firing alert VALUE = {{ $value }}"
23 ###### BEGIN MISC NOTES ######
26 # other interesting exporters
27 # https://github.com/prometheus-community/node-exporter-textfile-collector-scripts
30 # interesting post: https://www.metricfire.com/blog/top-5-prometheus-alertmanager-gotchas/
32 # interesting promql query that could be useful later.
33 # changes(ALERTS_FOR_STATE[24h])
37 # alert flap strategy.
38 # https://roidelapluie.be/blog/2019/02/21/prometheus-last/
40 # Another idea generally is to make an alert that fires for 24 hours and
41 # inhibits another alert for the same thing, which we want at most
42 # 1 alert per 24 hours.
44 ###### END MISC NOTES ######
46 # various queries only look at increases, so invert the up metric so we
47 # can better query on down.
52 # alerting on missing metrics:
53 # https://www.robustperception.io/absent-alerting-for-scraped-metrics
54 # that doesnt work if we want to alert across multiple hosts, eg
55 # up{job="node"} == 1 unless node_systemd_unit_state{name="systemstatus.service",state="active",job="node"}
56 # however, google lead me to a solution here
57 # https://www.linkedin.com/pulse/prometheus-alert-missing-metrics-labels-nirav-shah
58 # there is also the absent() function, but i didnt see a way to make that work
59 - alert: mysers_units_missing
61 count(up{job="node"} == 1) by (instance) * 3 unless
62 count(node_systemd_unit_state{name=~"(systemstatus|btrfsmaintstop|dynamicipupdate).service",state="active"}) by (instance)
67 - alert: epanicclean_not_active
69 node_systemd_unit_state{name="epanicclean.service",state="active"} != 1
74 - alert: epanicclean_missing
76 count(up{job=~"node|tlsnode"} == 1) by (instance) unless
77 count(node_systemd_unit_state{job=~"node|tlsnode",name="epanicclean.service",state="active"}) by (instance)
82 - alert: mysers_not_active
84 node_systemd_unit_state{name=~"(systemstatus|btrfsmaintstop|dynamicipupdate).service",state="active"} != 1
89 - alert: sysd_result_fail
90 # not sure 30m is really needed, it prevents the alert from flapping
93 rate(node_systemd_unit_result_fail_count[30m]) > 0
97 - alert: exim_paniclog
103 - alert: check_crypttab
109 # 17 minutes: if we reboot causing 1 send to fail, thats 10 minutes. we
110 # test this every 5 minutes, so thats 15 minutes at most.
111 - alert: mailtest_check_vps
113 time() - mailtest_check_last_usec{job="tlsnode"} >= 60 * 17
117 summary: '17 minutes down'
119 - alert: mailtest_check_unexpected_spamd_vps
121 mailtest_check_unexpected_spamd_results >= 1
125 summary: 'jr -u mailtest-check -e'
127 - alert: mailtest_check_mailhost
129 time() - max by (folder,from) (mailtest_check_last_usec{job="node"}) >= 60 * 17
133 summary: '17 minutes down'
135 # 20 minutes. just allow for more due to prod alert.
136 - alert: mailtest_check_gnu_mailhost
138 time() - max by (folder,from) (mailtest_check_last_usec{folder="/m/md/l/testignore", from="iank@gnu.org"}) >= 60 * 20
142 summary: '20 minutes down'
146 expr: hour() == 17 and minute() < 5
151 summary: Prometheus daily test alert
154 #### Inhibit notes ####
155 ## Example of expressions to detect if the target_down alert
156 # fired in the last 24 hours. Initially, I thought his could
157 # be an alert which inhibits up_resets, but eventually I figured
158 # that doesn't make much sense, and the idea of using an alert
159 # that is not an indication of something wrong, only inhibits another
160 # alert, I think works better to integrate directly into the
161 # alert it would inhibit, this may mean a recording rule. That avoids
162 # an alert we have to ignore or filter out.
164 # Alternate expression, to calculate if the alert would have fired is:
165 # min_over_time(sum_over_time(up[30m])[1d:]) == 0
166 # where 30m matches the for: time in target_down
168 # Note: for graphing, surround in the expression in sum_over_time()
169 # ALERTS{alertname="target_down",alertstate="firing"}[1d]
170 #### end Inhibit notes ####
173 # For targets where we alert only on long downtimes, we
174 # still want to know if it is going down many times for short times over
175 # a long period of time. But ignore reboots.
177 ## Another way would be to detect an overall downtime:
178 # avg_over_time(node_systemd_unit_state{name="dynamicipupdate.service",state="active"}[1d]) < .95
181 resets(up[1d]) - changes(node_boot_time_seconds[1d]) > 12
185 summary: "Target has gone down {{ $value }} times in 1 day, > 12"
189 # https://awesome-prometheus-alerts.grep.to/rules
191 # todo, we should probably group the prometheus alerts that indicate a
192 # host-local problem.
193 # eg, set a label alert-group: local-prom, then make a receiver that
194 # groups by it when the alert-group is local-prom.
196 - name: awesome prometheus alerts
199 - alert: PrometheusJobMissing
200 expr: absent(up{job="prometheus"})
205 summary: Prometheus job missing (instance {{ $labels.instance }})
206 description: "A Prometheus job has disappeared\n VALUE = {{ $value }}"
208 # TODO: some hosts, notably li and MAIL_HOST, we want to alert sooner than 30m,
209 # and severity to day. mail host is tricky since it roams, but I think the
210 # right way to do it is to check for absence of this metric:
211 # mailtest_check_last_usec{folder="/m/md/l/testignore",from="ian@iankelling.org"}
218 summary: Target down for 30m
221 # todo: this should group with the above alert
222 - alert: PrometheusAllTargetsMissing
223 expr: count by (job) (up) == 0
227 # alert-group: local-prom
229 description: "A Prometheus job does not have living target anymore.\n VALUE = {{ $value }}"
231 - alert: PrometheusConfigurationReloadFailure
232 expr: prometheus_config_last_reload_successful != 1
237 description: "Prometheus configuration reload error\n VALUE = {{ $value }}"
239 - alert: PrometheusTooManyRestarts
240 expr: changes(process_start_time_seconds{job=~"prometheus|pushgateway|alertmanager"}[30m]) > 10
245 description: "Prometheus has restarted more than ten times in the last 30 minutes. It might be crashlooping.\n VALUE = {{ $value }}"
247 - alert: PrometheusAlertmanagerJobMissing
248 expr: absent(up{job="alertmanager"})
253 description: "A Prometheus AlertManager job has disappeared\n VALUE = {{ $value }}"
255 - alert: PrometheusAlertmanagerConfigurationReloadFailure
256 expr: alertmanager_config_last_reload_successful != 1
261 description: "AlertManager configuration reload error\n VALUE = {{ $value }}"
263 - alert: PrometheusNotConnectedToAlertmanager
264 expr: prometheus_notifications_alertmanagers_discovered < 1
269 description: "Prometheus cannot connect the alertmanager\n VALUE = {{ $value }}"
271 - alert: PrometheusRuleEvaluationFailures
272 expr: increase(prometheus_rule_evaluation_failures_total[3m]) > 0
277 description: "Prometheus encountered {{ $value }} rule evaluation failures, leading to potentially ignored alerts.\n VALUE = {{ $value }}"
279 - alert: PrometheusTemplateTextExpansionFailures
280 expr: increase(prometheus_template_text_expansion_failures_total[3m]) > 0
285 description: "Prometheus encountered {{ $value }} template text expansion failures\n VALUE = {{ $value }}"
287 - alert: PrometheusRuleEvaluationSlow
288 expr: prometheus_rule_group_last_duration_seconds > prometheus_rule_group_interval_seconds
293 description: "Prometheus rule evaluation took more time than the scheduled interval. It indicates a slower storage backend access or too complex query.\n VALUE = {{ $value }}"
295 - alert: PrometheusNotificationsBacklog
296 expr: min_over_time(prometheus_notifications_queue_length[30m]) > 0
301 description: "The Prometheus notification queue has not been empty for 10 minutes\n VALUE = {{ $value }}"
303 - alert: PrometheusAlertmanagerNotificationFailing
304 expr: rate(alertmanager_notifications_failed_total[1m]) > 0
309 description: "Alertmanager is failing sending notifications\n VALUE = {{ $value }}"
311 # file_sd doesnt count as service discovery, so 0 is expected.
312 # - alert: PrometheusTargetEmpty
313 # expr: prometheus_sd_discovered_targets == 0
318 # description: "Prometheus has no target in service discovery\n VALUE = {{ $value }}"
320 - alert: PrometheusTargetScrapingSlow
321 expr: prometheus_target_interval_length_seconds{quantile="0.9"} > 90
326 description: "Prometheus is scraping exporters slowly\n VALUE = {{ $value }}"
328 - alert: PrometheusLargeScrape
329 expr: increase(prometheus_target_scrapes_exceeded_sample_limit_total[10m]) > 10
334 description: "Prometheus has many scrapes that exceed the sample limit\n VALUE = {{ $value }}"
336 - alert: PrometheusTargetScrapeDuplicate
337 expr: increase(prometheus_target_scrapes_sample_duplicate_timestamp_total[5m]) > 0
342 description: "Prometheus has many samples rejected due to duplicate timestamps but different values\n VALUE = {{ $value }}"
344 - alert: PrometheusTsdbCheckpointCreationFailures
345 expr: increase(prometheus_tsdb_checkpoint_creations_failed_total[1m]) > 0
350 description: "Prometheus encountered {{ $value }} checkpoint creation failures\n VALUE = {{ $value }}"
352 - alert: PrometheusTsdbCheckpointDeletionFailures
353 expr: increase(prometheus_tsdb_checkpoint_deletions_failed_total[1m]) > 0
358 description: "Prometheus encountered {{ $value }} checkpoint deletion failures\n VALUE = {{ $value }}"
360 - alert: PrometheusTsdbCompactionsFailed
361 expr: increase(prometheus_tsdb_compactions_failed_total[1m]) > 0
366 description: "Prometheus encountered {{ $value }} TSDB compactions failures\n VALUE = {{ $value }}"
368 - alert: PrometheusTsdbHeadTruncationsFailed
369 expr: increase(prometheus_tsdb_head_truncations_failed_total[1m]) > 0
374 description: "Prometheus encountered {{ $value }} TSDB head truncation failures\n VALUE = {{ $value }}"
376 - alert: PrometheusTsdbReloadFailures
377 expr: increase(prometheus_tsdb_reloads_failures_total[1m]) > 0
382 description: "Prometheus encountered {{ $value }} TSDB reload failures\n VALUE = {{ $value }}"
384 - alert: PrometheusTsdbWalCorruptions
385 expr: increase(prometheus_tsdb_wal_corruptions_total[1m]) > 0
390 description: "Prometheus encountered {{ $value }} TSDB WAL corruptions\n VALUE = {{ $value }}"
392 - alert: PrometheusTsdbWalTruncationsFailed
393 expr: increase(prometheus_tsdb_wal_truncations_failed_total[1m]) > 0
398 description: "Prometheus encountered {{ $value }} TSDB WAL truncation failures\n VALUE = {{ $value }}"