infra/stacks/monitoring/modules/monitoring/alert_digest.py
Viktor Barzin 97dcf49b8e
Some checks failed
ci/woodpecker/push/build-cli Pipeline was successful
ci/woodpecker/push/default Pipeline was canceled
monitoring: reduce Slack alert noise (alert-on-change + daily digest)
Reviewed the last 24h of Slack alerts after the midday node-pressure blip:
the volume came far less from the outage than from (a) alerts re-pinging
every few hours while nothing changed and (b) a pod cascade that fired
uninhibited. This hardens the alerting *system* so recurrences are quiet,
rather than just clearing today's broken services.

Changes (all in the monitoring module):

* Alert-on-change routing. warning/info repeat_interval -> 8760h (notify
  once, then only on a membership change or resolve); critical 1h -> 6h
  (a slow nag, not an hourly drip). send_resolved stays on. The bulk of
  the 24h volume was these re-pings (RpiSofiaUndervoltage alone fired
  continuously for ~24h, re-notifying every 4h).

* Daily digest CronJob (alert_digest.tf + alert_digest.py) -> #alerts at
  08:00 Europe/London: the full current board grouped by severity + what
  resolved in the last 24h. This is the standing-state safety net for the
  alert-on-change model. Stock python:3.12-alpine, pure-stdlib script
  (no pip/apk at runtime -> none of the per-run disk-write footprint that
  disabled status-page-pusher). Reuses the existing Alertmanager Slack
  webhook via a namespaced Secret; reads Alertmanager v2 + Prometheus.

* Cascade inhibition. NodeConditionBad/NodeDiskPressure now suppress the
  downstream pod-churn alerts (PodCrashLooping, PodImagePullBackOff,
  PodsStuckContainerCreating, ScrapeTargetDown, *ReplicasMismatch, ...).
  The midday DiskPressure event on 4 nodes fired 25 PodCrashLooping + 14
  PodImagePullBackOff uninhibited because only NodeDown was a source.

* T3 probe de-duplication. T3ProbeLegDown now inhibits T3ProbeDropBurst
  for the same leg — two alerts described one condition and were the #1
  noise source (~3,400 alert-minutes over 24h).

* ScrapeTargetDown false positives. Scrape only Ready endpoints, so
  completed CronJob pods that linger in EndpointSlices as NotReady
  addresses stop firing phantom "down" alerts (tts/tripit/beads). A Ready
  pod with a genuinely broken metrics endpoint still fires.

* for: 0m -> 5m on the flappy backup-status flags (LVM/Weekly/Offsite/
  NfsMirror/Vzdump *Failing) and DNS spike detectors, so a single
  transient Pushgateway/scrape blip no longer fires-and-resolves.

* Added an Alertmanager scrape target: it carried no prometheus.io/scrape
  annotation, so notification volume was unmeasurable — now we can verify
  this change worked (alertmanager_notifications_total et al.).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-12 20:35:56 +00:00

230 lines
8.2 KiB
Python

#!/usr/bin/env python3
"""Daily alert digest -> Slack.
Posts a once-a-day "state of the lab" summary to the #alerts Slack channel:
the full current board of firing alerts grouped by severity, plus a one-line
list of what fired-and-cleared in the last 24h.
This is the safety net for the "alert on change" routing model (warnings/info
no longer re-notify while firing; criticals re-ping slowly). The digest is the
recurring reminder of everything still firing, reviewed once each morning.
Pure stdlib (urllib + json) on purpose: the CronJob runs stock python:alpine
with NO pip/apk install at runtime, so it has none of the per-run disk-write
footprint that got status-page-pusher disabled (infra memory id=559).
Sources:
* Current board -> Alertmanager v2 (/api/v2/alerts): active, not silenced,
not inhibited == exactly what a human would otherwise be paged about, with
the human-readable `summary` annotation. Falls back to Prometheus ALERTS
if Alertmanager is unreachable.
* Resolved-in-24h -> Prometheus (alertnames seen firing in the last 24h that
are not firing now). Best-effort; skipped silently if Prometheus errors.
Env (all have in-cluster defaults):
ALERTMANAGER_URL default http://prometheus-alertmanager.monitoring.svc.cluster.local:9093
PROMETHEUS_URL default http://prometheus-server.monitoring.svc.cluster.local:80
SLACK_WEBHOOK_URL Slack incoming-webhook URL. If empty (or DRY_RUN set),
the payload is printed to stdout instead of posted.
SLACK_CHANNEL default "#alerts"
DRY_RUN if set (any value), print instead of posting.
"""
import datetime
import json
import os
import sys
import urllib.parse
import urllib.request
ALERTMANAGER_URL = os.environ.get(
"ALERTMANAGER_URL",
"http://prometheus-alertmanager.monitoring.svc.cluster.local:9093",
).rstrip("/")
PROMETHEUS_URL = os.environ.get(
"PROMETHEUS_URL",
"http://prometheus-server.monitoring.svc.cluster.local:80",
).rstrip("/")
SLACK_WEBHOOK_URL = os.environ.get("SLACK_WEBHOOK_URL", "").strip()
SLACK_CHANNEL = os.environ.get("SLACK_CHANNEL", "#alerts")
DRY_RUN = bool(os.environ.get("DRY_RUN", "")) or not SLACK_WEBHOOK_URL
SEV_ORDER = ["critical", "warning", "info"]
SEV_EMOJI = {"critical": ":red_circle:", "warning": ":large_yellow_circle:", "info": ":large_blue_circle:"}
def _get_json(url, timeout=30):
req = urllib.request.Request(url, headers={"Accept": "application/json"})
with urllib.request.urlopen(req, timeout=timeout) as resp:
return json.load(resp)
def _humanize(seconds):
seconds = int(max(seconds, 0))
d, rem = divmod(seconds, 86400)
h, rem = divmod(rem, 3600)
m, _ = divmod(rem, 60)
if d:
return "%dd%dh" % (d, h)
if h:
return "%dh%dm" % (h, m)
if m:
return "%dm" % m
return "<1m"
def _now_utc():
return datetime.datetime.now(datetime.timezone.utc)
def _age(starts_at):
if not starts_at:
return ""
try:
ts = starts_at.replace("Z", "+00:00")
# Trim sub-second precision beyond microseconds if present.
started = datetime.datetime.fromisoformat(ts)
return _humanize((_now_utc() - started).total_seconds())
except ValueError:
return ""
def fetch_current_from_alertmanager():
"""Active, non-silenced, non-inhibited alerts with their summaries."""
q = urllib.parse.urlencode(
{"active": "true", "silenced": "false", "inhibited": "false", "unprocessed": "false"}
)
data = _get_json("%s/api/v2/alerts?%s" % (ALERTMANAGER_URL, q))
alerts = []
for a in data:
if a.get("status", {}).get("state") != "active":
continue
labels = a.get("labels", {})
ann = a.get("annotations", {})
alerts.append(
{
"alertname": labels.get("alertname", "?"),
"severity": (labels.get("severity") or "info").lower(),
"lane": labels.get("lane", ""),
"summary": ann.get("summary", ""),
"age": _age(a.get("startsAt", "")),
}
)
return alerts
def fetch_current_from_prometheus():
"""Fallback: firing alerts from Prometheus (no summaries, includes inhibited)."""
url = "%s/api/v1/query?%s" % (
PROMETHEUS_URL,
urllib.parse.urlencode({"query": 'ALERTS{alertstate="firing"}'}),
)
data = _get_json(url)
alerts = []
for s in data.get("data", {}).get("result", []):
m = s.get("metric", {})
alerts.append(
{
"alertname": m.get("alertname", "?"),
"severity": (m.get("severity") or "info").lower(),
"lane": m.get("lane", ""),
"summary": "",
"age": "",
}
)
return alerts
def fetch_resolved_last_24h(active_names):
"""Alertnames that fired in the last 24h but are not firing now."""
try:
url = "%s/api/v1/query?%s" % (
PROMETHEUS_URL,
urllib.parse.urlencode(
{"query": 'count by (alertname) (max_over_time(ALERTS{alertstate="firing"}[24h]))'}
),
)
data = _get_json(url)
seen = {s["metric"].get("alertname", "?") for s in data.get("data", {}).get("result", [])}
return sorted(seen - active_names)
except Exception:
return []
def build_message(alerts, resolved):
today = _now_utc().strftime("%a %d %b %Y")
by_sev = {s: [] for s in SEV_ORDER}
for a in alerts:
by_sev.setdefault(a["severity"], []).append(a)
n = len(alerts)
counts = " ".join("%s %d" % (s, len(by_sev.get(s, []))) for s in SEV_ORDER if by_sev.get(s))
if n == 0:
header = ":white_check_mark: *Daily alert digest* — %s\nAll clear: nothing firing." % today
else:
header = ":bar_chart: *Daily alert digest* — %s\nFiring now: *%d*%s" % (
today,
n,
(" (" + counts + ")") if counts else "",
)
lines = [header]
for sev in SEV_ORDER:
items = sorted(by_sev.get(sev, []), key=lambda a: a["alertname"])
if not items:
continue
lines.append("")
lines.append("%s *%s (%d)*" % (SEV_EMOJI.get(sev, ""), sev.capitalize(), len(items)))
for a in items:
lock = ":lock: " if a["lane"] == "security" else ""
age = (" _(%s)_" % a["age"]) if a["age"] else ""
summary = ("%s" % a["summary"]) if a["summary"] else ""
lines.append("%s*%s*%s%s" % (lock, a["alertname"], summary, age))
# Any non-standard severities (defensive — shouldn't happen).
extra = [s for s in by_sev if s not in SEV_ORDER and by_sev[s]]
for sev in sorted(extra):
lines.append("")
lines.append("*%s (%d)*" % (sev, len(by_sev[sev])))
for a in sorted(by_sev[sev], key=lambda a: a["alertname"]):
lines.append("• *%s* — %s" % (a["alertname"], a["summary"]))
if resolved:
lines.append("")
lines.append(":white_check_mark: Resolved in last 24h (%d): %s" % (len(resolved), ", ".join(resolved)))
return "\n".join(lines)
def post_to_slack(text):
payload = {"channel": SLACK_CHANNEL, "text": text}
if DRY_RUN:
print("[DRY_RUN] would POST to Slack channel %s:\n%s" % (SLACK_CHANNEL, text))
return
body = json.dumps(payload).encode("utf-8")
req = urllib.request.Request(
SLACK_WEBHOOK_URL, data=body, headers={"Content-Type": "application/json"}
)
with urllib.request.urlopen(req, timeout=30) as resp:
if resp.status >= 300:
raise RuntimeError("Slack POST failed: HTTP %d" % resp.status)
def main():
try:
alerts = fetch_current_from_alertmanager()
source = "alertmanager"
except Exception as e:
sys.stderr.write("alertmanager fetch failed (%s); falling back to prometheus\n" % e)
alerts = fetch_current_from_prometheus()
source = "prometheus-fallback"
active_names = {a["alertname"] for a in alerts}
resolved = fetch_resolved_last_24h(active_names)
text = build_message(alerts, resolved)
sys.stderr.write("digest: %d firing (source=%s), %d resolved-24h\n" % (len(alerts), source, len(resolved)))
post_to_slack(text)
if __name__ == "__main__":
main()