mirror of
https://github.com/hanzoai/datastore.git
synced 2026-09-26 13:55:07 +00:00
The suite verified return-to-baseline only after a successful query (reap path) or an externally killed pool worker. This covers the failure path: the script allocates ~128 MiB and hangs, `command_read_timeout` fails the query, and the error-path teardown must SIGTERM the stuck process and release its registry entry, returning both metrics to baseline. A leaked wrapper on the error path would keep the gauge stuck high; no other test would catch that.