[Debian-med-packaging] Bug#1120170: Bug#1120170: bbmap autopkgtest uses insane amount of memory

Étienne Mollier emollier at debian.org
Thu Nov 6 20:23:28 GMT 2025


Hi Paul,

Paul Gevers, on 2025-11-06:
> I looked at the results of the autopkgtest of your package because I was
> investigating memory issues on our ci.d.n infrastructure. I noticed that
> your package uses an insane amount of memory.
>
> On our amd64 worker, I saw it peaking at 51% of 250 GB. On our loong64
> worker I saw it peaking at 83% of 127 GB. Now there's an argument to make
> that the infra should protect itself against this kind of misbehavior, but
> currently it doesn't (we're working on that). I would appreciate it if you
> limit the test on memory usage.

My apologies, the software preallocates memory for its Java
runtime environment, calculated on what it estimates being
available on the machine as a whole.  I understand this does not
play well on shared systems; that being said, I do see some code
that tries to accomodate the memory allocation when running on
slurm-wlm clusters.  I don't have a good idea how to address the
issue of large memory consumption of the software, but would it
help if I implement a cap of memory preallocation in autopkgtest
context?

If so, I consider implementing a check based on the definition
of the AUTOPKGTEST_TMP environment variable.  That sounds very
hackish, but would avoid hammering on ci.d.n, while having low
risk of introducing possible performance regressions for end
users.

> In the past I already added a reject list entry on riscv64 as the test times
> out. It also seems to take much more time on loong64 than on other
> architectures, so I'm also going to add it to the reject list for loong64
> too. You don't need to investigate this as riscv64 has some under powered
> workers so I consider that an infrastructure problem and loong64 isn't a
> release architecture yet, but solutions for both architectures are welcome.

Yes, that might have been 2021 or so, I rediscovered a patch of
mine to cap the memory usage on 32-bit platforms at 2000MiB.

Thank you for raising that issue!

Have a nice day,  :)
-- 
  .''`.  Étienne Mollier <emollier at debian.org>
 : :' :  pgp: 8f91 b227 c7d6 f2b1 948c  8236 793c f67e 8f0d 11da
 `. `'   sent from /dev/pts/3, please excuse my verbosity
   `-    on air: Wobbler - Naiad Dreams
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <http://alioth-lists.debian.net/pipermail/debian-med-packaging/attachments/20251106/da0b238b/attachment.sig>


More information about the Debian-med-packaging mailing list