Repository navigation
Home
github-actions[bot] edited this page Nov 18, 2026
·
2 revisions
How Lapilli judges alert rules, and how to run it: once over a period with
lapilli verdict, or every hour with lapilli serve. The same pages are at
lapilli.dev/docs and in the
wiki, a copy of
docs/ written again on each change; the README says what
Lapilli is, and how to install it.
-
Getting started: install it, write down an incident
or two, and read a first report on your own alerts; then keep judging with
lapilli serve - Words: stones, firings, incidents, the ledger, and the rest of the words these pages use
- Questions: what it needs, what it writes, and what it sends anywhere
- How it judges: where firings and incidents come from, when a firing catches an incident, noisy rules and what to change, what a rule adds, when firings came, near misses, rules that changed
-
Use
lapilli verdict: the commands, the reports it writes, the data sources it reads and how fast, and when it fails - The numbers: what each number is, and how to count it again from the report, without Lapilli
- A cloud's own alarms: judging what CloudWatch, Azure Monitor, Google Cloud and OCI alarms fired
- Rules on logs: Loki: LogQL rules replayed against Loki
-
Is the replay right?:
--check-replay, the replay beside what really fired -
What an alert says:
--check-content, what each rule says when it fires, and how it is written
-
Before a change is merged:
--compare, the rules proposed beside the rules in place, and what a rule is expected to do (--expect) - Pull request backtests in CI: the backtest as a comment on each pull request that changes alert rules
- Incidents kept elsewhere: the incident log copied from where it is kept
-
Incidents nobody wrote down:
--impact, incidents found in a signal of user impact
- Every week: the Helm chart's weekly report, to Slack and a Pushgateway
-
Showing a report outside:
--redact, a report with every name given an alias
- Keep judging: the ledger and its dashboard, teams, tasks and goals; checking it before it is deployed, the ledger in Postgres, several replicas, firings recorded elsewhere, tasks kept in step with a tracker
- Notifications as they were sent: the notifications kept beside the firings
- Changing rules: proposals, approval, and the change made where the rule is kept
-
Ask it from an agent: MCP at
/mcp
-
Commands: every option of
lapilli verdict,serve,preflightandledger, as-hprints it -
The HTTP API: every endpoint
lapilli serveanswers on, and the role each needs -
Metrics: every metric
lapilli serveexposes at/metrics - Upgrading: which version changed the ledger, and going back
- When something goes wrong: what an error or an odd number means, and what to do
- What it doesn't do yet
- The Helm chart: charts/lapilli
- Scripts that copy in incidents, firings, tasks and earlier versions of the rules, open a pull request for an approved change, and run backtests in CI: recipes
- How each part was designed: design
This page is copied from docs/README.md, where it is changed: a pull request there changes it here.
Start here
Judging a period
- How it judges
- Use
lapilli verdict - The numbers
- A cloud's own alarms
- Rules on logs: Loki
- Is the replay right?
- What an alert says
Before a change
Incidents
Reports
Every hour: lapilli serve
Reference
And
Also at lapilli.dev/docs.