trafficserver-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From geo...@free.fr
Subject Re: cache inspector alternative
Date Wed, 09 Jul 2014 17:39:15 GMT

> >>>>> There is a Jira ticket somewhere to expose the cache inspector
> >>>>> features as a set of command-line tools, but Jira being what it
> >>>>> is,
> >>>>> I can't find it right now.
> >>> 
> >>> http://nwk1-gnscapacity-sv11.apple.com:4242/cgi-bin/heatmap.cgi?num_hours=2&precision=0&stats_system=Epic&summarize=no&map_name=jptyo2_cdn_heat
> >> 
> >> The link is not reachable from my point view
> > 
> > sorry I meant https://issues.apache.org/jira/browse/TS-2194

thanks, but as you said, it's still unsolved.

> >>>>>> I was thinking of querying Traffic Server for non
> >>>>>> stale objects to retrieve all valid URLs stored but it can be
> >>>>>> a
> >>>>>> big IO consumer...
> >> 
> >> The same question : currently is it possible to query Traffic
> >> Server with existing tools and plugins for non stale objects to
> >> retrieve all valid URLs stored ?
> > 
> > I think the cache inspector UI can do this.
> 
> 
> To traverse the entire cache, it'd be incredibly expensive.

yes it does not end at all, just making Traffic Server performance down a lot.

Today one stress test runs a massive curl+purge of 1.9 million of objects took 10 minutes
while adding 6.4 millions of objects 10 minutes before and during 30 minutes. Note that this
stress test was done 4 times today, trafficserver process eats all the given RAM (30GB) and
more that causes system to swap, even after a restart of Traffic Server before a new test.
Server is HP DL 360p gen8 with 1 Intel Xeon E5-2667 / 32GB of RAM / 4x4TB Seagate Constellation
drives in a big RAID5. ATS is version 4.1.2 under FreeBSD 10... I will give a try with version
4.2.2 I patched earlier for FreeBSD 10 to try to eliminate the swap problem.

The platform I use contains 2 millions of objects for one category, the site has more than
25 languages translation available reaching more than 50 million of objects. So if I want
to purge this category, even using CURL+PURGE method, it will take a long long time, ie 250
minutes so more than 4 hours! My boss does not accept this. For me it looks reasonnable except
swapping.
I'd looked at the "cache-key-genid" (https://github.com/godaddy/ats-plugin-cache-key-genid)
that can partially meet my needs but as it is based on "host" and not "path".

> Take a look at the regex cache invalidation plugin that Comcast
> donated. It might do what you want for purging?

Nice, do you have any pointer on that ? I tried to find something about it on google with
no success.

Thanks again!
--
Denis

> -- Leif
> > 
> > J
> 

Mime
View raw message