Observability
Avrea tracks failure and flake rates across your team, captures every step duration, and keeps every log line searchable. Spot regressions from the analytics view, drill into a specific run, and read step-level logs.
Analytics
Section titled “Analytics”The Analytics page surfaces workflow- and job-level metrics over the last 30 days:
- Flake rate: jobs that failed where the same step has succeeded in other runs, surfacing flaky tests across your team.
- Failure rate: fraction of completed jobs that failed.
- Duration: median (p50) and p95 of successful runs.
- Workflow runs: total count, broken down by success and failure.
Each metric includes a 30-day trend chart so you can see whether things are getting better or worse.
Run history
Section titled “Run history”The run history view lists every workflow run across your organization, filtered by repository, branch, workflow, or status. Click any run to see its jobs; click a job to see its steps and logs.
Each run and job also has a firewall panel listing the destinations it connected to and anything that was blocked. See Egress firewall.
Step-level logs
Section titled “Step-level logs”Each job detail page lists every step in the workflow with its own collapsible log section. The log viewer supports:
- Live tail: new lines append and auto-scroll while the job is running. Auto-scroll pauses when you scroll up so you can read back without losing your place.
- Search and filter: full-text search with match highlighting, plus a filter mode that hides non-matching lines.
- Log groups: GitHub Actions
::group::/::endgroup::markers are rendered as collapsible sections. Groups containing search matches auto-expand. - ANSI rendering: 256-color and truecolor escape sequences render as styled output, including cursor controls.
- Copy: copy the visible (or filtered) lines to the clipboard, with or without timestamps.
Logs are available during the run and kept after the job completes.
Export VM logs to your own collector
Section titled “Export VM logs to your own collector”Organization admins can forward the logs of runner VMs to an OpenTelemetry collector of their own under Settings → OTel export. Enter the OTLP/HTTP endpoint and, if the collector needs one, the name and value of an authentication header. The header value is stored as a credential: it is not shown again, and the form keeps it until you replace or remove it.
The organization setting is the default for every repository. A repository's own OTel export settings override it.
Slack notifications
Section titled “Slack notifications”Connect a Slack workspace under Settings → Integrations to get notifications in your channels: failing runs with a button to rerun them, commit feeds for repositories with a Git mirror, and organization notices.
Channel routes choose which events land in which channel. A run route covers the repository's default branch unless you widen it with a branch pattern, and a route can be muted without removing it. A channel marked invite bot needs the Avrea app added before messages can be delivered.
Runner usage
Section titled “Runner usage”Settings > Usage > Runner shows how many CI minutes your organization spent and where they ran. Use it to find what is behind a change in usage, or to track a migration to Avrea.
Switch between four views:
- Runner type: minutes on Avrea, GitHub-hosted and other runners, broken down by runner label.
- Workflows: the workflows that used the most run time, each with its repository.
- Jobs: the same, per job.
- Runtime OS: minutes per operating system.
Group the chart by day, week or month, pick the current month, an earlier month or a rolling window, and export any view to CSV.
The page does not list repositories per runner. As a starting point for
finding where a runner is used, export the Workflows view to CSV, which
names each workflow's repository, and check runs-on in those workflow files.
runs-on can be an expression or come from a reusable workflow, so the label
may not appear in the file itself.
A few things to know when reading the numbers:
- The page is not a billing total. Billed usage also includes customer VMs and builder sessions, which are not shown here.
- GitHub's own usage report lists Avrea runners as self-hosted.
- Minutes are job run time from start to finish, for finished jobs in repositories you can access where the Avrea GitHub App is installed.
- GitHub-hosted covers GitHub's standard runner labels. Larger GitHub-hosted runners use labels you choose, so they appear under Other runners together with self-hosted and third-party runners.
- The workflow and job views list up to 200 rows, and the runner type view up to 200 labels each for Avrea, GitHub-hosted and other runners, most run time first. Totals still count every job.
- CodeBuild usage is grouped by project and overrides across runs.
- Avrea rows keep the labels your workflows request. Where the invoice names the runner differently, the row says what it is billed as.
- Runtime OS comes from the runner image for Avrea jobs, and from the requested labels for other runners.
- Every organization member can open the page. Amounts are on the Billing page.
Right-sizing
Section titled “Right-sizing”Avrea also suggests a smaller or larger runner for each job from the CPU, memory and disk its recent runs used. See Runner right-sizing.
Debugging a live job
Section titled “Debugging a live job”For inspecting VM state, running ad-hoc commands, or confirming a fix in-place, SSH directly into the running job. See Debugging.