Pierre Fersing

From my search the logger that support deletion of history are: * log4j 1.2 but with size based rotation: http://logging.apache.org/log4j/1.2/apidocs/org/apache/log4j/RollingFileAppender.html => maxBackupIndex * log4j 2.8, using DefaultRolloverStrategy that has delete...

I don't see how it help. It should already be used by [log4j.xml](https://github.com/jmxtrans/jmxtrans/blob/3dc555c34e2e5b48fc3adb541bb2ab6ef1088fc9/jmxtrans/src/main/resources/log4j.xml#L46) and don't seem to have any option to limit history size. If you are thinking to https://logging.apache.org/log4j/extras/apidocs/org/apache/log4j/rolling/RollingFileAppender.html...

(closed my mistake, sry)

I'm not sure jmxtrans itself can do something for this bug. I've opened at PR on appassembler (which generate the init.d script for RPM): https://github.com/mojohaus/appassembler/pull/60

I'm also hitting this issue (uWSGI restarted often, resulting in lots of files which cause /metrics endpoints to become too slow). I'm interested and willing to help to see this...

In uWSGI case, mark_process_dead will probably run at process exit (so with own PID). At least it's own I did it (using Python atexit handler). But in Gunicorn, mark_process_dead will...

BTW, this race-condition is not new. Gauge livesum and liveall already had this issue (new worker open the file, then mark_process_dead run by master remove them). Or did I miss...

In psutil, the create_time has more precision that in our case (it's a float number of seconds). But I agree that this event look like a very low probability (and...

Is it possible that a reconnection happen and you are using QoS = 0 message ? If yes, it's probably fixed by #796

To my understanding this issue was fixed with release 1.6.0 in commit https://github.com/eclipse/paho.mqtt.python/commit/9a4941efd758ca4231d4f16a8d3951686e24cb92: The ack of QoS = 1 message now occur after on_publish callback. If the issue isn't fixed,...