A test helper races with the process it watches and fails under load #40

Priority

low

Components

• app/nullius (backend)
• app/web (routes)
• templates and static files
• docs and the site
• catalogue (benchmarks, agents, capabilities)
• the standard (ports, kits, relay, runs)
• CI, workflows and skills

Evidence

tests.test_kit_check_isolation.CommandAgentGroup.test_a_normal_end_ends_what_it_left_running failed twice in a row in the docker tier on a machine busy with other containers, and passed three times run alone:

File "app/tests/test_kit_check_isolation.py", line 212, in alive
    if Path(f"/proc/{pid}/stat").exists() and Path(f"/proc/{pid}/stat").read_text().split()[2] == "Z":
FileNotFoundError: [Errno 2] No such file or directory: '/proc/1450917/stat'

The helper checks that the file exists and then reads it; the process can end between the two. The run it tests did what it should: the process was gone.

Screenshot

No response

Reproduction

1. Load the machine (other containers running).
2. python3 app/tests/parallel.py --docker, a few times.

Expected

The helper reads /proc/<pid>/stat once and takes a file that is gone as the process having ended.

Acceptance

• alive treats a /proc/<pid>/stat that disappears as not alive, instead of raising.

Dependencies and related

none