Disk cleanup¶
Workers keep ephemeral scratch directories under configurable workdirs and delete them when each job finishes (success or failure).
| Workdir env | Used by | Pattern |
|---|---|---|
UPLOAD_WORKDIR |
upload-processor | <base>/upload-<random>/ |
ATOMIZE_WORKDIR |
atomize-processor | scratch under base (or system tmp) |
Contract covered by unit tests (tests/test_upload_scratch.py and atomize equivalents): dirs are removed after success and after failure.
Ops notes¶
- ECS instance root volumes should be large enough for concurrent probe runs (upstream deploy guide suggests large EBS, e.g. ≥ 100 GB) plus shared
WORKDIR_ROOTfrom user-data. - Durable artifacts belong in S3, not on the instance. If disk fills, check stuck containers / incomplete cleanup, then recycle the ECS instance or clear orphaned workdirs under the configured roots.
- Legacy poll mode used
REPOS_BASE_PATHfor on-disk clones; queue mode prefers S3 + scratch.
Unknowns
There is no separate long-running “disk cleanup daemon” documented in the atomizer README. Host-level janitor scripts, if any, live outside this repo.