There is only 1 destination, although it is a file named with macros of date and hour.
destination workstation.log { file("/var/syslog/workstation.log.$R_YEAR$R_MONTH$R_DAY.${R_HOUR}0000" ); };
And that volume has never become full.
Also, with a file based destination I can't actually turn flow-control off since files have soft flow-control.
Because I have 3500 real connections that are all active (total of 10,000 messages per second) syslog-ng cycles through reading 100 messages from each source for 35000 messages and then writing those to disk. it is very common
for the queued messages to fluctuate from a few hundred to 200,000 messages.
I may have to add some new metrics to our statistics gathering to understand more about what is happening.
Evan Rempel.
On 2022-02-17 13:13, Laszlo Varady (lvarady) wrote:
Hi,
Do you have flags(flow-control) specified in your log paths?
If so, a dead destination in such log paths might cause the mentioned issue.
When flow-control is activated, the corresponding sources will be suspended. This suspended state does not even allow syslog-ng to truly release connections that have been closed by the clients.
This is actually more of expected behavior as we don't want to allow new connections in situations where logs could not be delivered anyway.
Please check the queued statistic counters of "syslog-ng-ctl stats" to see whether this is the case.
In case of anything else, I would suspect a bug.
--
László Várady
I am having an issue that is a little difficult to reproduce so I wanted
some input from others.
I have a syslog-ng 3.35.1 that has a TLS source defined with
max-connections(10000)
After some time the server starts logging a lot of messages
syslog-ng[12802]: Number of allowed concurrent connections reached,
rejecting connection; client='AF_INET(XXXX:61062)',
local='AF_INET(YYYY:6514)', group_name='client_network_tcp',
location='/etc/syslog-ng/syslog-ng.server.conf:61:9', max='10000'
To the best of my ability I can only find about 2500 actual connections.
Both lsof and netstat report around the 2500 connections.
I had to restart syslog-ng to stop this situation.
Has anyone seen this behavior before?
I get a lot of TLS connections without a certificate.
Error reading RFC6587 style framed data
Pperhaps the counters are not decremented for those timed out connections?
--
Evan Rempel