Bug #9683
closed
Silent failures when root filesystem is full
0%
Description
I recently saw this issue on a production 2.4.4-RELEASE-p2
Root filesystem was full due to not ticking a box for log rotation.
The main dashboard was an empty screen. Squid and Suricata had both died silently. Firewall was still passing traffic without AV or IPS.
Not only that but their configs were lost and had to be restored from backup.
I realise that certain configurations could have prevented this, or at least alerted that their was an issue. But this is a pretty terrible thing to happen out-of-the-box.
Updated by Jim Pingle about 6 years ago
- Status changed from New to Not a Bug
That's a problem with a package, not the base system. That wouldn't happen with the base system alone.
Probably also less likely to happen when #8350 is implemented.
Aside from that, if the filesystem is full, it's already too late for the system to do much of anything useful. There are ways, such as SNMP, to monitor and alert about disk usage.
Updated by Neil McKenzie about 6 years ago
I'm surprised this was so hastily marked as 'Not a Bug' considering the consequences.
Yes, filling up the filesystem was due to package configuration. However the dashboard dying is the base system right? And what about the config files being lost? It happened to both Suricata and Squid. Are they both problematic or is this something that can be prevented with changes in the base system?
Can the OS be configured (by default) to reserve space so it doesn't behave so badly when it fills up? Can changes to the base system prevent packages from filling up the filesystem?
This is very bad thing to happen in production!