Download tasks/afx-e2-s02/environment/workspace/monitoring/README.md from openenvforge/3ambench: direct link, hf CLI and curl.
- Browser
- Download file 6.87 kB
-
https://huggingface.co/datasets/openenvforge/3ambench/resolve/main/tasks/afx-e2-s02/environment/workspace/monitoring/README.md
- Command line
-
hf download hf://datasets/openenvforge/3ambench/tasks/afx-e2-s02/environment/workspace/monitoring/README.md
-
curl -L -o README.md https://huggingface.co/datasets/openenvforge/3ambench/resolve/main/tasks/afx-e2-s02/environment/workspace/monitoring/README.md
monitoring
This is where Linnea Vård's alerting lives: the rules our Prometheus evaluates and the Alertmanager config every Prometheus in the company sends to. Observability owns it; teams ask for changes with OPS- tickets and review whatever touches their own rules. The platform and data Prometheus servers use the same Alertmanager, which is why many team subtrees route alerts this repo never fires.
Where things live
| path | contents | source of truth for |
|---|---|---|
rules/ |
every alerting and recording rule | |
alertmanager/alertmanager.yml |
routes, receivers, inhibitions | |
slo/services.yaml |
the SLO catalog | SLO targets and windows, SLI family per service, traffic floors |
teams/ownership.yaml |
owners | which team owns each service; beats any runbook |
docs/adr/ |
decisions | burn-rate policy, scrape intervals, disk alerts |
runbooks/*.md |
one page per alert or service | what the alert means, how quickly it must page |
incidents/ |
packets from past incidents | PagerDuty exports and postmortems are the authoritative times; Slack threads are local time and can be a minute off. |
tests/visible/ |
the smoke suite | nothing; it is not a spec |
CI is bin/af-check: promtool check, the smoke tests, amtool check-config. To replay an incident against a branch, feed one of the packet's query_range pulls to bin/range2test and run the promtool test it writes.
Some files are older than the conventions here. Leave their formatting alone unless you are changing them for a reason. Every group runs at the global evaluation interval of 1m: no interval:, query_offset: or limit: on a group. No rule reads ALERTS or ALERTS_FOR_STATE, except an alert that keeps its own firing series up. If one alert should silence another, write an inhibition rule in Alertmanager so on-call can see it.
No symlinks anywhere in the repo: the deploy job rejects them. Copy the file instead.
Scrape targets
We scrape the app pods and the ingress controller. Relabelling puts the Kubernetes Service name in service and the pod name in pod on every app target, so a service's targets are up{service="…"}. The kubelet and kube-state-metrics are scraped by platform's Prometheus and we don't alert on them; platform federates an allowlist of cAdvisor series to us (container_memory_*, container_spec_*, container_cpu_*), none from kube-state-metrics. The uptime vendor's synthetic checks cover the public homepage only. That leaves up and each service's own request counters as our liveness signals.
ADR 0004 has the scrape interval of every job. A rate window should cover about four scrape intervals.
Paging policy
page: wakes someone up.critical: the old name forpage. Routed and paged exactly like it; writepagein new rules.warning: pages nobody.ticket: lands in the owning team's queue.
Some generation-1 alerts still exist twice, as a warning and a critical copy. The rule for paging: users are affected now or will be within the hour, and the person who gets paged can act. If not, it's a ticket.
Alert classes
- Service alerts (
<ServiceCamel>HighErrorRatio,<ServiceCamel>Down, burn alerts) page the owning team. Service error alerts are per service, never per pod; one page per service per outage. They use a 5m window andfor: 10m. Errors are 5xx; 4xx are the caller's problem. - Capacity alerts (
ContainerMemoryNearLimit, disk) open a ticket for the owning team, except where an ADR says a level pages (disk: ADR 0015).ContainerMemoryNearLimitopens a ticket when a container keeps reaching 90% of its memory limit for 10 minutes. One busy 5-minute stretch isn't that. We want the ticket inside 20 minutes of the first time it reaches 90%, which leaves time before the OOM. - Fleet alerts (node, database, cache, broker exporters) are
infra's,severity: warning, and page nobody. - Any other alert keeps the
severityandteamit has unless a ticket says otherwise. Ticket alerts carryservicelike pages do.
Every alert has labels severity and team. Service, capacity and ticket alerts also carry service on the
alert (from the expression or as a static label). Annotations: a summary that says what is affected, and
runbook_url.
Service health
<ServiceCamel>Down (for checkout-api: CheckoutApiDown) is the only Down alert a service has. Lose the last healthy target, or lose the targets altogether, and the owning team is paged within 10 minutes; a restart that takes targets away for at most 2 minutes doesn't page. Labels severity: page, team, service.
SLO alerting
Error SLIs are recorded as slo:sli_error:ratio_rate<window> (windows 5m, 30m, 1h, 6h): the fraction of bad requests by (service), one record per window for every service with the family's metrics. Don't add a service filter to them. Latency SLIs are recorded per service as slo:sli_latency:ratio_rate<window>, since each service has its own objective. A burn alert reads its own service's SLI records and nothing else; it is called <ServiceCamel>ErrorBudgetBurnFast (severity: page, for: 2m) or <ServiceCamel>ErrorBudgetBurnSlow (severity: ticket, for: 15m) and is labelled severity, team and slo: <service>-availability. Factors: ADR 0008; window: the catalog.
Routing
The team label on the alert is the routing key and has to match teams/ownership.yaml. Each team has a subtree matching its team:
| alert severity | delivered to |
|---|---|
page or critical |
pagerduty-<team> + slack-<team> + slack-sre-fyi |
ticket |
jira-<team> + slack-<team> |
warning |
slack-<team> only |
SRE watches every page in #sre-fyi (slack-sre-fyi); nobody is paged from there. The root receiver slack-alerts-default is the catch-all for alerts with no team; nobody is on it. Rule evaluation failures go to slack-prometheus-meta. The paging unit is the service: one page per service per outage; two services failing together are two pages, even for the same team. Don't change receivers you don't own, global, or add time intervals.
Inhibition rules
- service alerts holding back other service alerts: always
equal: [service], so no outage silences another service. - critical copy over warning copy of the same alert for the same service:
equal: [alertname, service].
Runbooks
Runbook links point at https://runbooks.linneavard.example/alerts/<AlertName>. That site is generated from runbooks/*.md here; a page's front matter lists the alerts it covers.
Changelog
- 2025-11: burn alerts moved to the SLI records (ADR 0008).
- 2026-02:
slack-sre-fyiadded; SRE stopped being paged for every team. - 2026-05: disk alerts per ADR 0015.
- 2026-08: kafka-exporter scraped every 1m instead of 15s (ADR 0004).