Bug#852461: systemd: Creating scopes with systemd-run unreliable

[Michael Biebl]

Am 27. Januar 2017 18:56:31 MEZ schrieb Michael Gebetsroither <michael at mgeb.org>:
>On 2017-01-24 20:20, Michael Biebl wrote:
>Hi Michael,
>
>>> The creation of scopes with systemd-run --scope is unreliable.
>>> It seems to depend on system "load" or some other factors affected
>>> by system "load".
>>> On production systems we've seen an error rate of up to 20%.
>> 
>>> % for i in `seq 50`; do (systemd-run --scope -- sleep 5 2>&1 ) &
>done 2>&1 |grep -ni failed
>> 
>> 
>> You mean you see this problem under high load? What kind of load?
>
>We are seeing this problem especially under high load, but for the
>commands above and the output
>this was on a test-system without any load on it (4 cores kvm).
>
>> Could it be that you are running into D-Bus timeouts?
>
>doesn't seem like it, see below.
>
>> Do you get any error messages in the journal?
>
>no
>
>> Can you reproduce the issue with v232 from stretch or tell us how we
>can
>> reproduce the issue?
>>
>> Even my small PI (running raspbian + systemd 215-17+deb8u6) seems to
>> handle your for loop without a hitch even if under load.
>
>thx for trying so hard to reproduce the issue!
>And sorry for wasting your time and not just sending some debug logs :/
>
># strace log of an unsuccessful systemd-run run
>
>The error sent from systemd back to systemd-run seems to be:
>"Unit run-24908.scope already exists."


Could you check if you have many scope units in /run which could slow down systemd