[gold] remove old alerts
These should already have been ported to the k8s style.
Change-Id: I93b5b258f7e32cabe5af48c7c1ef9026b711ced9
Reviewed-on: https://skia-review.googlesource.com/c/buildbot/+/234911
Reviewed-by: Kevin Lubick <kjlubick@google.com>
diff --git a/prometheus/sys/alert.rules b/prometheus/sys/alert.rules
index a436c22..d580238 100644
--- a/prometheus/sys/alert.rules
+++ b/prometheus/sys/alert.rules
@@ -171,66 +171,6 @@
description = "Issue tracker ingestion has failed to run in at least 30 minutes."
}
-# Gold
-
-ALERT GoldIgnoreMonitoring
- IF liveness_gold_expired_ignore_rules_monitoring_s{instance="skia-gold-prod:20001"} > 200
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "At least two rounds of monitoring for expired ignore rules have failed back to back. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#goldignoremonitoring",
- }
-
-ALERT GoldErrorRate
- IF rate(num_log_lines{level="ERROR",job=~"skiacorrectness-.*"}[2m]) > 1
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "The error rate for Gold {{ $labels.instance }} is too high. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#golderrorrate https://console.cloud.google.com/logs/viewer?project=google.com:skia-buildbots&resource=logging_log%2Fname%2F{{ $labels.log_group }}&logName=projects%2Fgoogle.com:skia-buildbots%2Flogs%2F{{ $labels.log_source }}",
- }
-
-ALERT GoldDiffServerErrorRate
- IF rate(num_log_lines{level="ERROR", instance="skia-diffserver-prod:20000"}[2m]) > 1
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "The error rate for Gold Diffserver {{ $labels.instance }} is too high. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#golddiffservererrorrate https://console.cloud.google.com/logs/viewer?project=google.com:skia-buildbots&resource=logging_log%2Fname%2F{{ $labels.log_group }}&logName=projects%2Fgoogle.com:skia-buildbots%2Flogs%2F{{ $labels.log_source }}",
- }
-
-ALERT GoldIngestionStalled
- IF liveness_gold_s{metric="since-last-run",source="poll"} > 750
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "At least two rounds of gold ingestion have failed back to back for {{ $labels.corpus }}. See https://mon.skia.org/dashboard/db/gold-panel",
- }
-
-ALERT GoldIngestionErrorRate
- IF rate(num_log_lines{level="ERROR",job=~".*_ingestion"}[2m]) > 1
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "The error rate for Gold Ingestion {{ $labels.corpus }} is too high. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#goldingestionerrorrate https://console.cloud.google.com/logs/viewer?project=google.com:skia-buildbots&resource=logging_log%2Fname%2F{{ $labels.log_group }}&logName=projects%2Fgoogle.com:skia-buildbots%2Flogs%2F{{ $labels.log_source }}",
- }
-
-ALERT GoldCommitTooOldWallTime
- IF gold_last_commit_age_s{type="wall_time",job!="skiacorrectness-pdfium"} > 24*60*60
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- abbr = "{{ $labels.job }}",
- description = "It has been at least 24 hours since the last commit made it into Gold for {{ $labels.job }}. Some process might have hung, or perhaps that repo simply has not seen a commit in that period. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#goldcommittoooldwalltime",
- }
-
-ALERT GoldCommitTooOldNewerCommit
- IF gold_last_commit_age_s{type="with_new_commit"} > 1*60*60
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- abbr = "{{ $labels.job }}",
- description = "It has been at least 1 hour since a new commit landed, and Gold for {{ $labels.job }} still hasn't picked it up. Some process might have hung. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#goldcommittoooldnewercommit",
- }
-
-ALERT GoldStatusStalled
- IF liveness_gold_status_monitoring_s > 10*60
- LABELS { category = "infra", severity = "critical", owner = "kjlubick@google.com"}
- ANNOTATIONS {
- description = "It has been at least 10 minutes since the Gold's status was re-computed. Some process might have hung. https://skia.googlesource.com/buildbot/+/refs/heads/master/golden/docs/PROD.md#goldstatusstalled",
- }
-
# Skolo
ALERT BackupNotDone