On Sun, Jun 22, 2025 at 12:45:31PM -0400, Noah Meyerhans wrote: > > It seems likely that the sa-comple postinst and spamd postinst are > > racing with each other here. I'm really surprised that this hasn't come > > up before... > > A dpkg trigger allowing spamd's own postinst to restart the process, > rather than doing it directly from sa-compile's postinst, may resolve > the race condition. Something like > https://salsa.debian.org/noahm/spamassassin/-/commit/991ad7c86f0bccb4d95dcc2de6e0eb42116b6f50
That commit was wrong; this one is better: https://salsa.debian.org/noahm/spamassassin/-/commit/2b3a32a070c2a739e8ade02e2d3f21bcfb20dc69 I had considered the possibility that even triggers wouldn't be necessary, and that we could just rely on spamd's postinst to restart the service (both spamd and sa-compile have versioned dependencies on spamassassin that will ensure that they're always upgraded together). However, because they don't have a direct dependency relationship, the order in which they're configured is nondeterministic. So I do believe that sa-compile's postinst needs to ensure spamd is restarted, one way or another. noah
