Pipeline run diagnostics and maintenance
MS Learn module: Diagnose and maintain pipeline runs · Learning objectives: control run retries and timeouts, read and debug rundowns, and keep pipelines healthy over time so CI stays trustworthy.
Overview
A pipeline that rarely fails is easy to trust — until it does. This module is the operational layer: how a run is scheduled, constrained (retry/back-off/timeouts), and debugged, and how to maintain pipelines (aging, feature-dependence, log hygiene) so a sporadically flaky job does not become a silent blocker. Everything here exists to keep the CI signal clear.
Units / lessons covered
- Run model: queueing, job retries, max attempts, back-off, timeouts per job/stage
- Debugging: logs, STDOUT capture, system data, download logs, rerunning,
--force/ redeploy - Pipeline maintenance: node/hardware drift, credential rotation, purged runs and retention
- Monitoring: run status API, health dashboards, alerts on failed/slow runs
- GitHub Actions:
retries,timeout-minutes,concurrency, Rerun failed jobs
Concepts introduced (link to term notes)
Key terms / commands
retryCountOnTaskFailure·timeoutInMinutes·maxWait·rerun failed·concurrency- GitHub:
retries:·timeout-minutes:·concurrency:· Rerun failed jobs · run status badges
Hands-on / what to try
Force a step to fail, inspect the log to find the error, add a retry with a small time-out so a transient failure passes a suite run; then set a 30-day run-retention policy and confirm old runs are purged automatically.
Exam focus
- Retries help transient failures; they do not fix logic bugs — over-retry can hide real defects.
- Timeouts bound runaway jobs; set them per step/job not only at pipeline scope.
- Maintenance is operational: it’s distinct from authoring and covers lifecycle (create/update/delete/retention).
Related
Path MOC · pipeline-components-and-stage · advanced-yaml-pipelines · pipeline-maintenance
📘 Source: Pipeline triggers and lifecycle · Troubleshoot pipelines
⚠ verify: include the standard pipeline artifacts — internal FAQ points to PublishSummary + Runtime logs; retention Purging docs pending.