[systemd-devel] systemd-journald doesn't follow max datagram length when forwarding messages to syslog

[Umut Tezduyar Lindskog]

On Wed, Aug 13, 2014 at 6:59 PM, Lennart Poettering
<lennart at poettering.net> wrote:
> On Sun, 10.08.14 21:07, Umut Tezduyar Lindskog (umut at tezduyar.com) wrote:
>
>> The problem is due to sending socket's SO_SNDBUF limit. My guess at
>> this point is the amount of messages you can queue to syslog socket is
>> determined by 3 factors. a) Sending socket's SO_SNDBUF b) Receiving
>> socket's SO_RCVBUF c) Global variable
>> /proc/sys/net/unix/max_dgram_qlen.
>>
>> My workaround is adding SendBuffer=8M to
>> systemd-journald-dev-log.socket but I am not sure if the way we
>> account things is intuitive. My propose is passing
>> systemd-journald-dev-log.socket to systemd-journald so that
>> systemd-journald can forward messages to rsyslog through it. Or not
>> deal with this since newer rsyslog fetches logs.
>>
>> References
>> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/tree/net/unix/af_unix.c#n1445
>> http://vger.kernel.org/~davem/skb_sk.html
>
> I have now changed git to also set SendBuffer=8M for the .socket unit.
>
> I must agree though, it's a bit surprising that we use the same socket
> here for recieving syslog and forwarding it.
>
> I was tempted to just add a second socket for the forwarding, that is
> internal to journald, and never seen outside. While this appears
> "cleaner" on one hand, it also would require us to either hardcode the
> buffer size or come up with a new config setting for it. Neither of
> these options are particularly attractive, in particular given that we
> can currently just set SendBuffer=8M and be done with it. Hence that's
> what I did.
>
> Does that make sense?

Thanks. I think it is good enough.

>
> Lennart
>
> --
> Lennart Poettering, Red Hat