mirror of
https://github.com/coder/coder.git
synced 2026-09-21 20:51:01 +08:00
Michael Suchacz
6e5f02bfbe
## Summary Reduce peak memory during linting so cold-cache runs do not exhaust memory on high-core development hosts or leave CI runners with little headroom. ## Problem `golangci-lint` v1 sizes its default concurrency from available CPUs, not the cgroup memory limit. On a 96-CPU development host capped at 32 GiB, a cold-cache full lint run peaked at about 33.9 GB and reached the cgroup limit. The CI workflow also used bare `make -j`, which allows every lint target to run concurrently. ## Fix Make the supported `make lint/go` path cap `golangci-lint` at the lower of detected CPUs and eight workers, preserving lower concurrency on smaller hosts. Preserve an inherited `GOMEMLIMIT`, use a configurable 8 GiB fallback when it is unset, and bound CI's parallel lint targets to the runner CPU count. In the same high-core cold-cache reproduction, peak memory for the full lint target fell to about 10.1 GB. Raw `golangci-lint run` invocations remain unchanged. A `golangci-lint` v2 migration remains separate because it does not size concurrency from the memory limit and requires broader configuration migration work. > Mux prepared and updated this PR on Mike's behalf.