Environment:
Cloudron 10.0.5, Ubuntu 26.04.1, MySQL 8.4.11 (also seen before
the OS upgrade on 22.04 / MySQL 8.0.46), ~100 apps, mail enabled, eventlog
table ~38k rows.
Symptom
The box service exits through the uncaughtException handler and is restarted
by systemd. 3 times in 23 minutes today ( NRestarts=3 ), and 3 times on
2026-09-24:
From uncaughtException handler.
BoxError: Lock wait timeout exceeded; try restarting transaction
at Object.transaction (box/src/database.js:58:15)
at upsertLoginEvent (box/src/eventlog.js:57)
While box restarts, the dashboard, LDAP/OIDC and mail authentication are
briefly unavailable.
What I found
The stack points at the UPDATE in upsertLoginEvent() . Because of
DATE(creationTime) in the WHERE clause, MySQL doesn't use
creationTime_index and scans the whole table:
EXPLAIN UPDATE eventlog SET dataJson='{}' WHERE action='user.login' AND sourceJson LIKE '%x%' AND DATE(creationTime)=CURDATE();
type: index | possible_keys: NULL | key: PRIMARY | rows: 32747 | Extra: Using where
On a test MySQL I could reproduce the lock wait: with that UPDATE open in
one transaction, a write to an unrelated eventlog row from a second
session times out.
Some callers (the mail auth paths in app-bridge.js and ldapserver.js )
call upsertLoginEvent() without await , which would explain why the error
ends up in the uncaughtException handler.
Happy to provide more logs if useful.
