การตรวจจับการทดสอบไม่แน่นอนและการทำงานแบบขนานใน CI

As an automated suite grows, two CI concerns dominate: keeping runs fast through parallelism, and keeping them trustworthy by catching flaky tests. This lesson covers both.

Running tests sequentially scales linearly with suite size. Splitting work across multiple workers shrinks wall-clock time and shortens the feedback loop developers depend on.

CI systems shard tests across parallel jobs. Each job runs a slice, identified by an index and total count.

Parallel runs only work if tests are independent. Shared databases, ports, or files cause collisions. Isolate state per worker, e.g. a separate schema per shard.

Naive splitting can leave one worker with all the slow tests. Use timing data so each shard takes roughly equal time.

A flaky test passes and fails without code changes. CI can detect them by re-running failures: a test that fails then passes on retry is flagged as flaky, not broken.

Many runners support an automatic retry count for failed tests. Keep it low so genuine failures still surface.

Store test results across builds to compute a flakiness rate per test. Dashboards highlight the worst offenders for the team to fix.

Decide whether to stop on first failure (fast feedback) or run the whole suite (complete picture). Many teams run all in CI but fail fast locally.

Cache dependencies and build outputs between runs so workers spend time on tests, not setup. This complements parallelism.

Parallel, flaky-aware pipelines give fast and believable signals, so developers act on red builds instead of re-running until green.

What is a prerequisite for safely running tests in parallel?

You learned to scale CI testing:

• Shard tests across workers, balanced by timing
• Isolation is required for safe parallelism
• Detect flaky tests via limited retries
• Track flakiness rates and fix the worst offenders