Re: poudriere-devel and tmpfs use by inactive builders: an example

On 5/3/26 01:39, Tatsuki Makino wrote:
> Thank you very much.
> 
> However, it seems that the maximum amount of memory used does not
> change, regardless of the timing of cleanup.
> So, the number of builders should not be too high, and running them
> without swaps like I do is completely out of the question :)

The most effective and efficient way to cleanup a tmpfs that does not
need any files preserved may well be to dismount and remount it
--instead of rm -fr or the like. I'd be very surprised if memory
allocations survived a dismount/remount handling.

Did you make your own changes to poudriere(-devel) to test that? If so,
what did you do?

> 
> Regards.
> 
> On 2026/05/01 5:15, Bryan Drewery wrote:
>> On 4/30/26 2:43 AM, Tatsuki Makino wrote:
>>> When jobid03 is running and jobid06 has finished its work and is on
>>> standby, the environment of jobid06 remains intact, which means that
>>> tmpfs continues to use memory.
>>> When looking inside /usr/local/poudriere/data/.m/main-CA76-
>>> default/06/var/db/pkg/local.sqlite, all the packages installed as
>>> dependencies of the last built port remain installed, and if llvm19
>>> are installed, it seems to consume as much as 2GB.
>>> ...isn't it right?
>>
>> Yes the cleanup only happens at the start of a package build. So it is
>> possible for wasted space/memory after a build is done.
>>
>> I'll review and consider cleaning up the tmpfs at the end of
>> build_pkg() as well.

Could that work by dismounting and remounting the tmpfs, avoiding
choices about freeing vs. not the RAM+SWAP by other layers?

>>
>> Bryan
> 
> 


-- 
===
Mark Millard
marklmi at yahoo.com