Hello! I am seeing something weird with trace instances, that I don't see with regular non-instance traces. If I enable some sched events and set the buffer size to something tiny, like 1KB, I am expecting the buffer to rotate quickly as it overflows, but I don't see that: watch "cat trace | head -n 1" and I see the following https://paste.googleplex.com/6259651512369152?raw and if I do some heavy workloads, I still see that the top rows are not moving and are stuck showing the same thing. I am expecting the buffer to drop old events as it is being overwritten/rotation. However if I do a tail on the trace, I do see updates. So tracing is in progress... I did the exact same experiment on non-instance global tracing and it works fine. I confirmed that 'overwrite' is set in the trace_options of the instance (and non-instance). This could be some trace formatting issue or another bug? Note I am reading the tracefs trace file directly and not using trace-cmd for this. Thoughts? Thanks, Joel
|