[AndrewSav]

Description

Note

TLDR: another small fix for race in tests that sometimes makes tests fail for no good reason.

• What is the current behavior (you can also link to an open issue here) ?

  The test restarts clamd before running the reload:

  s6-svc -d /services/clamd; while pgrep -x clamd > /dev/null; do sleep 0.2; done; s6-svc -u /services/clamd

  setup.sh gives every service a finish script that runs s6-svc -d when the service exits with code 0, so that a service stopped on purpose stays down. That script runs after the clamd process has exited, so the loop above can finish before it does. When the test's -u lands first, the finish script's -d wins and clamd stays down. The reload's ping (clamdscan --ping 300:1) then gives up after 300 seconds and the test fails.

  It is rare: one failure in a day of local suite runs, and none in 21 loops of the exact sequence. In those loops the finish script logged 32 to 196 ms before the test's -u. The failed run ended with clamd down (exitcode 0) and the finish script's stopping clamd process in the log.

  The image itself does not hit this. It brings clamd up from freshclam's run script and reloads it with SIGUSR2; nothing in it stops clamd and starts it again.

• What is the new behavior (if this is a feature change) ?

  s6-svc -wD -d /services/clamd; s6-svc -u /services/clamd

  -wD returns only once the service is down and its finish script has exited, so the -u can no longer be undone. The image's own .s6-svscan/finish waits the same way. No other test stops and restarts a service.

Type of change

• Test (adding missing tests or correcting existing ones)

Status

• Ready

Todo List

• Tests

How has this been tested ?

• With a 1-second delay added to clamd's finish script before its -d, the old sequence left clamd down 3 times out of 3, the same end state as the failed run. With the same delay in place, the new sequence brought clamd up 3 times out of 3.
• The test still catches the bug it was written for: straight after the new restart, a reload that does not wait for clamd fails with rc 2 (socket missing).
• The default suite passed 136/136 with this change, on 9d2d46a. The branch has since been rebased onto bd09d3b, which only touches the workflows; the change itself is identical.