Access logs aren't flusing on traffic_server death

[danobi]

If there are logs in the flush buffer when we kill traffic_server, we lose these logs. There needs to be a way for ATS to flush these logs upon receiving a signal.

The obvious way would be to call Log::flush_thread_main(..) in Main::proxy_signal_handler(..), but I highly doubt that function is signal safe.

[mingzym]

well, I don't know it is still the old style, but TS have a shutdown command(way) to stop the server safely, while most init scripts just kill, you should prefer the shutdown if it still works, because you care it. flush on kill or crash sounds cool but may bring too many code complex or trouble, IMO

[danobi]

Is the shutdown command ./trafficserver stop? That makes sense if the script did a safe shutdown. The reason I bring this up is because @dragon512 is working on a testing framework and (I assume) only starting up the traffic_server binary and skipping the script. When we test for the presence of logs, if we kill too fast we get nothing in the logs.

[mingzym]

Back some version early, there is a command line traffic_line: -S, --shutdown on false Shutdown traffic_server (local node) please find out. as I am sure the scripts mostly will do a kill, and trafficserver is the sysV init script.

[mingzym]

I am not updated with the current traffic ctl commands etc, so maybe you can find it out current status on this.

[danobi]

70ef31e removed traffic_line. I worked with the traffic_ctl code for process control before and I don't recall seeing any logging related code. I have a feeling this issue has always existed but just nobody noticed.

[mingzym]

great that you take care of the ctl codes. afaict, MGMT_EVENT_SHUTDOWN handling is still in the codes, maybe you can find a way to use it?

[maskit]

#1937 may provide a chance for flushing logs.

[bryancall]

If this is still an issue please reopen.