Re: [Savannah-hackers-public] Out of memory issues on Savannah?

[Simon Josefsson]

Bob Proulx <bob@proulx.com> writes:

> Hi Simon,
>
> Simon Josefsson wrote:
>> Not really a solution to the savannah malloc failures (I've been seeing
>> them too when pulling gnulib 10x in parallel from savannah during
>> continous integration pipeline jobs) but a note that I am (as
>> libidn/libidn2 maintainer) considering to not use a gnulib submodule but
>> rather rely on ./bootstrap to pull in gnulib through some other
>> mechanism, which would allow you to use a gnulib mirror and the
>> --gnulib-srcdir or --gnulib-refdir options to ./bootstrap, which would
>> reduce savannah gnulib clone usage.
>
> I am just going to ask what URL you are using to obtain the gnulib
> clone?  This one?
>
>     git clone https://git.savannah.gnu.org/git/gnulib.git

Yes, it is the default in gnulib's own bootstrap-funclib.sh:

default_gnulib_url=https://git.savannah.gnu.org/git/gnulib.git

>> Are there any savannah usage graphs per-project?  I wonder if gnulib is
>> a top consumer of resources, or if some other project (guix?) is a large
>> consumer that needs optimizing.
>
> Nothing per project because all of the git services are running on one
> VM virtual machine.  (It's currently on vcs2 running on kvmhost3.)
> We have system level metrics trend graphs.
>
>     https://git.savannah.gnu.org/
>
> The memory trend graph shows that earlier today shows a significant
> memory use for several hours and then it stopped.
>
> For the most part our method has been to monitor general system
> activity and to block abuse using fail2ban rules.  That's a
> reactionary stance and fails when we are all busy with the conference
> but works most of time.  Meanwhile I was talking just today with the
> FSF sysadmins about available resources as a general thing.

Okay.  It would be nice to somehow understand if my CI/CD setup causes a
real resource drain on savannah, or if this usage is just noise in the
overall resources.  Having some kind of top list of git repositories
using resources would help with that.  GNU InetUtils CI/CD pulled 5+
gnulib clones at the same time, but since there is little activity for
InetUtils it only happens once every other month or so.  I did change it
to use a GitLab.com git mirror instead, and that resolve the spurious
errors we saw, but to be honest I kind of doubt that this was the
trigger.  But without some data it is just guessing.

/Simon

signature.asc
Description: PGP signature