You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Overlay-base database size cap (7,500 MB) disables incremental PR analysis for large JavaScript/TypeScript repos #4327
Incremental (overlay) analysis never kicks in for our pull requests, because the default branch's
javascript-typescript overlay-base database is larger than the hard-coded upload cap in src/overlay/caching.ts (OVERLAY_BASE_DATABASE_MAX_UPLOAD_SIZE_MB = 7500). Every default-branch
run logs:
Database size (11832 MB) exceeds maximum upload size (7500 MB). Skip uploading overlay-base database to cache.
so every PR run finds no overlay-base database and falls back to a full analysis.
A large pnpm monorepo: about 22,000 JS/TS files extracted after paths-ignore (tests, specs,
stories, e2e, tooling) is already applied through a config file.
Numbers
Overlay-base database on the default branch: about 11.7 GB uncompressed, measured by the action.
The same file set built locally as a plain database (codeql database create, no overlay mode):
about 1.6 GB. So the overlay-base form is roughly 7x the plain database for us.
PR analysis time: about 17 minutes for javascript-typescript on 16 cores, all of it a full
re-analysis. Before the repo grew past the cap, PR runs took 2 to 7 minutes.
Trimming more paths cannot get us under the cap without excluding production code.
Ask
One of:
Make the cap configurable (an input or environment variable) for repos whose cache budget allows it.
Apply the cap to the compressed size the cache actually stores (the code comment estimates
about 3 GB compressed at 2.5:1).
Reduce the overlay-base database size before upload, if some of it is not needed to restore the
base for PR analysis.
Small update: the 7x size difference is unexpected, although it is somewhat on the higher side of what we see (average is something like a 4x size difference).
Activity
jketema commentedon Nov 16, 2026
Hi @hernan82arg,
We'll have a look to see whether we can make the cap configurable.
jketema commentedon Nov 18, 2026
Small update: the 7x size difference is unexpected, although it is somewhat on the higher side of what we see (average is something like a 4x size difference).