My new best friend


#1

my new best friend is /perl-status (mod_perl’s interactive status and
info tool). If you do any mod_perl coding and haven’t touched /perl-status
yet, go read the mod_perl guide and get up to speed on it. right now.
really. I mean it.

I believe that the horrible horrible memory I saw leaking was in fact due
to some upsetting interactions between Apache::Session and
DBIx::SearchBuilder::Record::Cachable.

While I was poking around, I also finally got the full picture of just
how upsettingly heavy Date::Manip is… By itself, it was eating almost
as much memory as all of RT. I’ve moved RT over to Date::Parse from
Graham Barr. It’s much smaller (236,606 bytes compared to 1,610,254 bytes)
It doesn’t have as many bells and whistles, but it’s less likely to take you
on a tour of your server’s swap partition.

In the 72 hours, The Beta 3 bug list has dropped from 28 issues to 13 issues.
We’re starting to get there.

    -j

jesse reed vincent – root@eruditorum.orgjesse@fsck.com
70EBAC90: 2A07 FC22 7DB4 42C1 9D71 0108 41A3 3FB3 70EB AC90

Gur SOV jnagf gb znxr guvf fvt vyyrtny.


#2

Jesse / Ivan,

Jesse wrote:

I believe that the horrible horrible memory I saw leaking was in fact due
to some upsetting interactions between Apache::Session and
DBIx::SearchBuilder::Record::Cachable.

While I was poking around, I also finally got the full picture of just
how upsettingly heavy Date::Manip is… By itself, it was eating almost
as much memory as all of RT. I’ve moved RT over to Date::Parse from
Graham Barr. It’s much smaller (236,606 bytes compared to 1,610,254 bytes)
It doesn’t have as many bells and whistles, but it’s less likely to take you
on a tour of your server’s swap partition.

This explains a lot. Especially the ‘slowing down’ - it was swapped out but
that wasn’t entirely down to RT. One of my esteemed colleages had whipped
384M of my 512M from the server. (Sounds of beating and kicking followed
by howls of agony). It’s back now! As the memory leak progressed, it
started shoveling things into swap and making things slow down considerably.

This still doesn’t explain the segv’s though as it shouldn’t do that,
swap or
no swap. I still suspect a race condition due to delays introduced by
swapping,
but it’s going to be a lot harder to prove now unless I knobble the memory
again which I’m obviously not keen to do!

In the 72 hours, The Beta 3 bug list has dropped from 28 issues to 13 issues.
We’re starting to get there.

Linear or exponential decay? Well done!

Mark

Mark Vevers. mark@ifl.net / mvevers@rm.com
Internet Backbone Engineering Team
Internet for Learning, Research Machines Plc
Tel: +44 1235 823380, Fax: +44 1235 823424