Skip to Content.
Sympa Menu

en - Re: [sympa-users] Profiling wwsympa bad performances and our findings

Subject: The mailing list for listmasters using Sympa

List archive

Chronological Thread  
  • From: Matt Taggart <address@concealed>
  • To: address@concealed
  • Subject: Re: [sympa-users] Profiling wwsympa bad performances and our findings
  • Date: Mon, 4 May 2020 10:55:26 -0700

On 5/4/20 2:33 AM, Olivier Salaün wrote:
Hello,

I wish to summarize here the results of our investigation to understand the origin of very bad performances of wwsympa in our production server (Sympa 6.2.48).

We are running a ML service with : 8 robots, 5.200 lists (total), 2.500 family lists, 960.000 subscribers (subscriber_table), 36.000 admins (admin_table).

You may know that on Sympa's web interface the critical operations are the ones that require wwsympa.fcgi to load all the mailing lists. This includes the */lists* and */lists_categories* URLs. On our main robot (1.500 ML) /lists processing takes: *3 minutes* if the wwsympa.fcgi process is not running ; *16 seconds* otherwise. These response times make Sympa's web interface almost unusable. That's why we decided to investigate...

We had to disable those years ago for the same reasons. We created a cronjob that runs once a day at 3am to generate a hierarchy of static pages for the topics. Having the directory be up to 24h out of date was worth getting rid of the overhead.

To find out what's going on we used a Perl Profiler (https://metacpan.org/pod/Devel::NYTProf). Once enabled it keeps track of all subroutine calls and draws a picture of the time spent in each subroutine, how many times it was called and other fine grained details.
[snip]

This is great work! Good job!

Thanks,

--
Matt Taggart
address@concealed



Archive powered by MHonArc 2.6.19+.

Top of Page