Re: logd vs. Polygraph

Robert Collins wrote:

> On 21 Aug 2001 22:52:05 -0500, Joe Cooper wrote:
>
>>Interesting results:
>>
>>Squid+logd access.log:
>>[root@zope squid-logd]# wc -l /var/log/squid/access.log
>> 331763 /var/log/squid/access.log
>>
>>cachemgr info:
>>Number of HTTP requests received: 698836
>>
>>Polygraph:
>>215.24| got 646166 xactions and 0 errors
>>
>>This strikes me a bit odd... I won't say logd is losing entries, but
>>I'm perplexed if it isn't. Note that the higher number in mgr output is
>>due to me not restarting Squid between the time I realized I had
>>forgotten to zero the access log--so I zeroed the log and restarted the
>>test without restarting Squid. But it does look like about half, but
>>not quite, half of the requests are disappearing.
>>
>
> Ouch. I wonder if squid is swallowing them itself (it has to keep the
> log pipe open to replace the log helper seamlessly.) I need to look into
> the select/poll stuff in more detail - unless Henrik or Adrian have any
> insight. (We open the fds using pipe(), then don't do anything with them
> other than pass them to the log helper. One of the ends gets marked as
> ipc.)

Ok, at 20 reqs/sec, it appears to keep up with every request.

I noticed that Squid was falling behind when running at 50 (remember
this is a completely stock compile without aufs, so not very fast and 50
does eventually overload it). So perhaps as long as Squid keeps it
together logd will too. I'll run a longer test at this rate to see if
anything degrades with time.
                                   --
                      Joe Cooper <joe@swelltech.com>
                  Affordable Web Caching Proxy Appliances
                         http://www.swelltech.com
Received on Tue Aug 21 2001 - 23:03:50 MDT