We observed two crashes of the Dovecot master during administrative reloads
on
18 November 2026 UTC. The test VPS runs Debian 13 (amd64) and uses local
ext4
for Maildir. The Dovecot CE vendor packages, including dovecot-core, are
2:2.4.5-2+debian13. OpenSSL is 3.5.7-1~deb13u3 and libc6 is
2.41-12+deb13u4. The effective shutdown_clients setting is yes.
The packaged dovecot.service uses /usr/bin/doveadm reload for
ExecReload.
These are the two observed cases.
- During a TLS configuration cutover, systemd initiated a reload. The
Dovecot master logged
SIGHUP received - reloading configuration. systemd then loggedMain process exited, code=killed, status=11/SEGV. The reload control process exited with75/TEMPFAIL. - During an IMAP certificate renewal rehearsal with Certbot 4.0.0, the
ACME staging renewal succeeded and the deploy hook ran the configuration
check followed by
systemctl reload dovecot.service. Certbot's dry run used the current IMAP certificate for the hook. This timeExecReloadexited with status0and systemd recorded anotherSIGSEGVin the main process. Starting Dovecot again with the same configuration restored IMAPS, and new connections verified the expected certificate.
We expected the master to remain running after a successful reload. Another TLS reload completed successfully between these two crashes, so the operation does not fail consistently on this host.
For the first crash, the recorded instruction address resolves to
__strcmp_avx2 in the installed libc. We have no caller frames or usable
core dump. One bounded GDB capture attempt did not reproduce the crash.
The trigger and the caller of the libc function remain undetermined.
Is this a known master reload issue? Which configuration details or diagnostic output would be most useful for triage?
We observed two crashes of the Dovecot master during administrative
reloads on
18 November 2026 UTC. The test VPS runs Debian 13 (amd64) and uses local
ext4
for Maildir. The Dovecot CE vendor packages, including dovecot-core, are
2:2.4.5-2+debian13. OpenSSL is 3.5.7-1~deb13u3 and libc6 is
2.41-12+deb13u4. The effective shutdown_clients setting is yes.
The packaged dovecot.service uses /usr/bin/doveadm reload for
ExecReload.
These are the two observed cases.
- During a TLS configuration cutover, systemd initiated a reload. The
Dovecot master logged
SIGHUP received - reloading configuration. systemd then loggedMain process exited, code=killed, status=11/SEGV. The reload control process exited with75/TEMPFAIL. - During an IMAP certificate renewal rehearsal with Certbot 4.0.0, the
ACME staging renewal succeeded and the deploy hook ran the
configuration
check followed by
systemctl reload dovecot.service. Certbot's dry run used the current IMAP certificate for the hook. This timeExecReloadexited with status0and systemd recorded anotherSIGSEGVin the main process. Starting Dovecot again with the same configuration restored IMAPS, and new connections verified the expected certificate.
We expected the master to remain running after a successful reload. Another TLS reload completed successfully between these two crashes, so the operation does not fail consistently on this host.
For the first crash, the recorded instruction address resolves to
__strcmp_avx2 in the installed libc. We have no caller frames or usable
core dump. One bounded GDB capture attempt did not reproduce the crash.
The trigger and the caller of the libc function remain undetermined.
Is this a known master reload issue? Which configuration details or diagnostic output would be most useful for triage?