beam.smp - high frequency of running on single node instance
Looking at processor utilization, I've notice that beam.smp runs every second to few seconds ...
1) What is the reason for this?
2) Is this due to stat collection?
3) Is this due to XDCR?
4) Is this writing to disk as well?
5) In either case is there way (especially during idle and/or during no update) to stop this?
Thanks
Ash
1) Understood
2) Understood
3) Understood
4) No views have been defined...
5) Tried both the versions (community/enterprise) of the final 2.0 release and have not seen any reduction in the beam.smp ...
Some details:
Machine (Centos 5.8 4GB Dual quadcore)
Couch (assigned 1GB) - no replication, no indexes, no views etc
Even with no data ... beam.smp hits up to 50% of CPU usage every second ..
It is my assumption that when there is nothing to do, with no data etc, it should not be hitting such high CPU usage.. Is this correct?
I understand that as the statistics and rules dictate, compaction will occur and beam.smp will be managing/executing such.
Your thoughts are much appreciated. :)
Cheers
Ash
1) beam.smp is the erlang process. We currently have all of our view, cluster management, and xdcr code written in erlang and all of these are part of beam.smp.
2) The cluster manager does stats collection from each node every second. I wouldn't expect this to cause a lot of cpu compared to the other process run in beam.smp. Stats collection is also likely a network bound process as opposed to cpu bound.
3) Some of this will be due to XDCR, but again this isn't the only thing running in the beam.smp process.
4) The view engine creates indexes and we also run the compaction process from erlang. This will contribute to disk activity in the beam.smp process.
5) I would need more information to understand exactly what you might be able to do to mitigate this issue. I can tell you though that we are continuing to work to make the cpu consumption as small as possible in beam.smp. The final 2.0 release will likely contain fixes to lower the consumption and future releases will definitely have fixes to lower the cpu consumption even further.