Jest generating thousands of new transform cache files on every CI run and and provides no performance improvement.

[tejasbagal1]

Hello everyone,
We are running a large Angular application (~1337 test suites, ~14k tests) in a Tekton/Kubernetes CI pipeline using Jest 29, ts-jest, jest-preset-angular, Babel (babelConfig: true), and coverage enabled (jest --coverage).

We configured a persistent Jest cache: cacheDirectory": "<rootDir>/.jest-cache"

At the end of each pipeline run, the .jest-cache directory is archived and stored in object storage. Before the next run, it is restored to the same location.

The cache is successfully restored and existing cache files are being reused. However, on every pipeline run:

• Cache size grows significantly (e.g. 212 MB → 579 MB)
• File count grows from ~5,900 to ~15,600 files
• Thousands of new transform cache files are generated
• Cache never seems to stabilize
• Test execution time does not improve as expected from a warm cache

Notable thing is workspace path remains unchanged, Node.js, Yarn, Jest, and ts-jest versions remain unchanged, The Jest transform cache directory hash remains the same across runs and Each CI execution runs in a new Tekton/Kubernetes pod

For comparison, the same project on my local machine macOS and a RHEL VM behaves as expected: the cache stabilizes after the first run and subsequent test runs are significantly faster.

Has anyone seen Jest/ts-jest caches continuously grow across Kubernetes/Tekton runs despite stable paths and tool versions?

Are there known factors that influence cache keys and could cause partial cache misses in ephemeral container environments?

[shreechaturvedi]

Set isolatedModules: true on the jest-preset-angular transformer and wipe the restored .jest-cache once. You'll lose type-checking in the transform.

transform: {
  '^.+\\.(ts|js|mjs|html|svg)$': [
    'jest-preset-angular',
    { isolatedModules: true, babelConfig: true },
  ],
},

With isolated modules off, ts-jest hashes statSync(dep).mtimeMs for every resolved import into the transform cache key. A new Tekton pod checks out the tree again, so the cache keys change even when the files didn't. Jest never drops the old files, so the directory only grows.

Your Mac/RHEL tree keeps those mtimes, so the same cache can hit there.