[systemd-devel] Journald - heap size

[Lennart Poettering]

On Wed, 20.08.14 09:08, Christian Häßelbarth (mail at christianhaesselbarth.de) wrote:

> Hi all,
> 
> I'm using systemd in version 197.
> I've observed that the heap size of the journald continuously grows.
> 
> I used "cat /proc/[PID_OF_JOUNALD]/smaps for observing the heap size.
> 
> Right after starting the heap size is about 772 kB
> 
> 10041000-10102000 rwxp 00000000 00:00 0          [heap]
> Size:                772 kB
> Rss:                 652 kB
> Pss:                 652 kB
> Shared_Clean:          0 kB
> Shared_Dirty:          0 kB
> Private_Clean:         0 kB
> Private_Dirty:       652 kB
> Referenced:          652 kB
> Anonymous:           652 kB
> AnonHugePages:         0 kB
> Swap:                  0 kB
> KernelPageSize:        4 kB
> MMUPageSize:           4 kB
> Locked:                0 kB
> 
> 
> After 7 days of operation the heap size is about 6272 kB
> 
> 10041000-10102000 rwxp 00000000 00:00 0          [heap]
> Size:                6272 kB
> Rss:                 6172 kB
> Pss:                 6172 kB
> Shared_Clean:          0 kB
> Shared_Dirty:          0 kB
> Private_Clean:         0 kB
> Private_Dirty:       6172 kB
> Referenced:          6172 kB
> Anonymous:           6172 kB
> AnonHugePages:         0 kB
> Swap:                  0 kB
> KernelPageSize:        4kB
> MMUPageSize:           4kB
> Locked:                0 kB
> 
> My journald.conf is the following
> 
> [Journal]
> Storage=none
> #Compress=yes
> #Seal=yes
> #SplitMode=login
> #RateLimitInterval=10s
> #RateLimitBurst=200
> #SystemMaxUse=
> #SystemKeepFree=
> #SystemMaxFileSize=
> #SystemMinFileSize=
> #RuntimeMaxUse=1M
> #RuntimeKeepFree=
> #RuntimeMaxFileSize=
> #RuntimeMinFileSize=
> #ForwardToSyslog=yes
> #ForwardToKMsg=no
> #ForwardToConsole=no
> #TTYPath=/dev/console
> #MaxLevelStore=debug
> #MaxLevelSyslog=debug
> #MaxLevelKMsg=notice
> #MaxLevelConsole=info
> 
> 
> Is it a known bug in version 197?

We keep a cache of open journal files with state. If we never have
written to anything we don't have anything open, so we use much less
memory.

And there's more, for example our hashmaps use an allocation cache,
which will show up as heap usage.

Just looking at the smaps file isn't really a way to identify leaks. And
allocation caches are a pretty common thing that always appear as heap
usage, even though the memory isn't leaked.

Lennart

-- 
Lennart Poettering, Red Hat