Re: Squid connection retry problem -- fd leak?

From: WWW server manager <webadm@dont-contact.us>
Date: Mon, 15 Sep 1997 09:36:58 +0100 (BST)

Apiset Tananchai wrote:
> I already backout connection-retry patch and only running 1.NOVM.14 with
> fixed for icp.c and unlinkd.c and it seems to work fine for the moment.
>
> Does anyone using connection retry patch with 1.NOVM.14 got similar
> problem with me? And do you have any suggestion to fix this problem?

I had the "original" file descriptor leak problem with Squid 1.NOVM.14 (and
earlier versions, I think) on Solaris 2.5, and with some extra diagnostic
logging here and Duane searching for the problem based on the resulting
evidence, the changes in 1.NOVM.15 do seem to have fixed that.

However ... adding the connection retry patch seems to cause "the same
problem" (as far as symptoms are concerned) to reappear when added to
1.NOVM.15 or 1.NOVM.16 (both of which work fine without it).

The symptoms I saw (for both problems) were that even when load was low
overnight, some high-numbered file descriptors remained in use (as shown by
the cachemgr display), initially as pairs of descriptors (one reading, one
writing) open to a zero-length cache file with an "old" (maybe days old)
modification date (and maybe also a socket FD sometimes, not sure about
that). After a SIGHUP, the "reading" sockets disappeared from the cachemgr
list, but the writing ones remained until I restarted the server. The stuck
descriptors would appear and accumulate gradually over time - days,
typically, rather than hours to become significant, though one time it
managed to get up to 700+ descriptors supposedly in use over a couple of
hours. Another symptom was that the cachemgr display also showed a list of
VM objects corresponding to the stuck descriptors, when the VM object list
would normally be empty with NOVM.

Duane's analysis of the original problem was that a connection problem
occurred at a stage in connection setup when Squid didn't have a timeout set
to allow it to clean up after problems, leaving things in a mess. It looks
as if the retry patch may introduce a similar problem.

                                John Line

-- 
University of Cambridge WWW manager account (usually John Line)
Send general WWW-related enquiries to webmaster@ucs.cam.ac.uk
Received on Mon Sep 15 1997 - 01:50:16 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:05 MST