Interpreting sar output on a Varnish server.

[maillists0 at gmail.com]

I have a fairly busy 2.1 box that stopped serving traffic a couple of times
recently. Since it's actively serving our sites, it ended up getting
bounced before we could look at Varnish to see what was going on. But
restarting Varnish did get things back to working.

In the process of trying to troubleshoot this, I came across something I
don't understand. If I run 'sar -n ALL', I get this for open sockets in the
period leading up to the crisis:


                            totsck    tcpsck    udpsck    rawsck
ip-frag    tcp-tw

    11:45:01 AM     14921     13657         0         0         0       118
    11:55:01 AM     23930     20664         0         0         0       152
    12:05:02 PM     32092     25969         0         0         0       113
    12:15:01 PM     39461     30043         0         0         0       113
    12:25:02 PM     46668     33715         0         0         0        94
    12:35:01 PM     54069     36689         0         0         0       107
    12:45:01 PM     61508     39127         0         0         0        90
    12:55:03 PM     68697     40981         0         0         0       101
    01:05:01 PM     75922     42634         0         0         0        93
    01:15:02 PM     82843     43848         0         0         0        98
    01:25:01 PM     89889     45174         0         0         0       103
    01:35:01 PM     97185     46296         0         0         0        94
    01:45:01 PM     100184     39404         0         0         0        85

Varnish became unresponsive near the 100,000 sockets point.

My question is this: if totsck is total sockets open, how is it that the
numbers in the other columns don't sum to equal totsck? What kind of socket
might be missing from this output? Does this even tell me anything useful
at all?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://www.varnish-cache.org/lists/pipermail/varnish-misc/attachments/20130816/887ab951/attachment.html>