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
This commit was created on GitHub.com and signed with GitHub’s verified signature.
Fixed
In-memory result storage works. A job configured with ResultStorage::Memory had
its result silently discarded: workers only stored ResultStorage::Database results.
Workers now store them in an in-memory store owned by the JobQueue and shared by its
clones, before the job is marked Completed, and get_job_result, delete_job_result, cleanup_expired_results and delete_job use it. Results expire at their TTL (checked
on read and pruned as results are stored), and the store holds at most 10,000 results
by default (queue::DEFAULT_MEMORY_RESULT_CAPACITY), evicting the one that expires
soonest, or the oldest; set the limit with the new JobQueue::with_memory_result_capacity. In-memory results are per process: other
processes, including cargo hammerwork and the web dashboard, don't see them, and
they are lost on restart. See Job Results. TestQueue::delete_job now also deletes the job's result.