#422345 intermittent full filesystem breaks squid until restart

Package:
squid3
Source:
squid
Submitter:
martin f krafft
Date:
2025-07-28 19:45:04 UTC
Severity:
important
#422345#3
Date:
2007-05-05 09:30:03 UTC
From:
To:
My /var just filled up, but some scripts immediately took action and
archived logs. However, while at 100% full, someone was accessing
the squid instance and it barfed:

  squid[4432]: diskHandleWrite: FD 110: disk write error: (28) No space left on device
  squid[4432]: storeAufsWriteDone: got failure (-6)
  squid[4432]: storeSwapOutFileClosed: dirno 0, swapfile 0000270E, errflag=-6   (28) No space left on device
  squid[4432]: WARNING: Shrinking cache_dir #0 to 898356 KB
  squid[4432]: commSetEvents: epoll_ctl(EPOLL_CTL_ADD): failed on fd=73: (1) Operation not permitted

Two seconds later, /var had space again, but any attempt to connect
to a webpage via squid got us a "connection refused" and squid kept
writing

  squid[4432]: diskHandleWrite: FD 16: disk write error: (28) No space left on device
  squid[4432]: storeDirWriteCleanLogs: Starting...
  squid[4432]: WARNING: Closing open FD   16

until I restarted the process.

squid should really be able to recover from this without having to
be restarted.

And ideally, it should respond to requests with a message alerting
to the error, rather than a "connection refused".

This is reproducible at will.