afsocket source does not set SOURCE about syslog-ng

[bazsi]

hmm... yup, that NULL setting is incorrect. it has been a performance
optimization (and probably premature at that).

Probably, way earlier the source driver didn't do anything but forward the
message, now it has some actual functionality. The entire NULL stuff could
probably be removed, I doubt it'd have any kind of measurable performance
drop. and if that would happen, I would go to the configuration compiler to
optimize the tree instead of trying to microoptimize function calls.

On Mon, Aug 4, 2014 at 2:34 PM, Gergely Nagy [email protected]
wrote:

Assuming I have this configuration:

@Version: 3.5
source s_foo { tcp(port(12345)); };
log { source(s_foo); destination { file("/dev/stdout", template("SRC: $SOURCE\n")); }; };

If I send a log message, $SOURCE will not be set, and stats for the
center counter will not be incremented, either. This is because in
afsocket_sd_init_instance()
https://github.com/balabit/syslog-ng/blob/416d3159c133a99134cf75607b74b2650cb73688/modules/afsocket/afsocket-source.c#L679,
we explicitly set the queue method to NULL, which will then fall back to
log_pipe_forward_msg(), which does not set SOURCE, nor does it increment
the center counter.

My first idea would be to remove that line, and have
log_src_driver_queue_method() do its thing. But I am unsure about the
implications.

@bazsi https://github.com/bazsi: Any insight?

(Thanks @faxm0dem https://github.com/faxm0dem for the report!)

—
Reply to this email directly or view it on GitHub
#186.

[algernon]

Thanks! That was my suspicion too. Will requeue this for 3.6 though, as the problem has not been reported until now (it was broken for a good while now), and this may break existing scripts when fixed.

Btw, I just checked, and in versions prior a24a380, it was, indeed, set to log_pipe_forward_msg(), and that commit turned it into a NULL. Sometime after, in 0e858cb, log_src_driver_queue_method() was changed. So this has been broken since 3.4 alpha, roughly.

[algernon]

I'm checking the code, and it seems that it's not only afsocket that is affected. The file() source does not go through log_src_driver_queue_method() either. streams() and program() sources appear to be unaffected.