[syslog-ng]source socket keepalive patch suggestion
Thu, 12 Sep 2002 10:30:34 +0200
Thanks for your contribution.
On Wed, Sep 11, 2002 at 11:55:26AM +0100, Ted_Rule@flextech.co.uk wrote:
> I previously enquired whether it would be possible to allow for a keep-alive
> on Unix DGRAM sockets to help avoid the situation on some platforms when daemons
> such as sendmail have the /dev/log socket 'whipped from under them' during
> It also occurred to me that avoiding closing/re-opening UDP sockets on HUP
> should help to avoid
> any loss of messages across HUP on a very busy central syslog server.
> I assumed - whether this is valid or not I don't know - that the current 1.5.20
> source always rebuilds
> Unix DGRAM sockets on HUP. Likewise for UDP sockets.
> Having had a look at the YACC source, I think I may have come up with a patch to
> help with
> this - it certainly appears to compile and run but it's a bit difficult to test
> all the possible variations on the theme,
> and whether it actually fulfils my original requirements.
No it doesn't. The keepalive flag is not implemented for unix-dgram sockets.
> The changes involved in the patch are as follows:
> Split parsing of Unix DGRAM and STREAM options apart a little more - previously
> it seems that
> the Yacc allowed for a max-connections options on a Unix DGRAM socket, even
> though I don't believe
> this is actually meaningful(?). The original source also seems to be at variance
> with the documentation here.
> Force default keepalive setting on TCP socket. The documentation suggests this
> is the default, but my reading
> of the Yacc suggests otherwise.
> Enable keepalive option on both flavours of DGRAM socket
I think it is better to use a separate option, like 'keep-socket-alive' or
'keep-listener-alive' as it might be a good idea to enable the same
behaviour for stream sockets as well (e.g. not to reopen the socket)
> Disable max-connections option on Unix DGRAM socket.
> Force default keepalive setting on both flavours of DGRAM socket.
> Could you enlighten me further as to whether this patch actually helps to
> acheive my original intent.
I'm afraid it doesn't.
PGP info: KeyID 9AF8D0A9 Fingerprint CD27 CFB0 802C 0944 9CFD 804E C82C 8EB1