1 Mar
2005
1 Mar
'05
2:15 a.m.
While a nice feature, this will only help in those cases where the source stream is managed. In traditional send-and-forget-udp syslog, this would not help the risk of data loss when the destination is a FIFO. On Mon, 28 Feb 2005 17:08:24 +0100, Balazs Scheidler <bazsi@balabit.hu> wrote:
How exactly does that work? You can only "stop writing" for so long, before a buffer fills up and you have to drop messages...
Simple: syslog-ng stops reading. if you have flow-controlled sources (e.g. unix-stream, or TCP with the other end doing flow control as well) then at the end the application will block on sending the message.
They're dropped, lost, unrecoverable. There really is nothing else syslog-ng could do in those situations.