decentraland / worlds-content-server / 30386332626

92%

main: 92%

Pull #517

github

fix: cancel content uploads when deployment processing aborts

Deployment requests were observed stuck in flight for hours: after the
processing deadline aborted a deployment, a wedged S3 upload could keep the
storage stage pending indefinitely, pinning the deployments-in-flight
gauges, the multipart upload lease (in-flight byte budget), the temp files,
and the request itself until a restart.

- Thread the deployment abort signal into storage.storeStream so an abort
  tears down the in-flight S3 request itself.
- Stop waiting for active uploads on abort in the content upload pool:
  every source stream is already destroyed by then, so waiting only tied
  the request to a transport that may never settle. Stragglers remain
  observed inside mapWithConcurrency.
- Configure explicit S3 client timeouts: the SDK v3 Node handler defaults
  both the connection and request timeouts to 0 (no limit).
- Migrate to @dcl/catalyst-storage v5 (AWS SDK v3), dropping the
  end-of-support aws-sdk v2 dependency and replacing deep dist imports
  with package exports.

1749 of 2022 branches covered (86.5%)

Branch coverage included in aggregate %.

5 of 5 new or added lines in 3 files covered. (100.0%)

3672 of 3886 relevant lines covered (94.49%)

114.24 hits per line