Here is some more debug information:
2017-02-15 10:19:08,266 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSinkAdapter: enqueue, logicalTime=80
2017-02-15 10:19:18,267 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source RegionServer,sub=WAL
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source RegionServer,sub=IPC
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source UgiMetrics
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source RegionServer,sub=Replication
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source JvmMetrics
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source RegionServer,sub=Server
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source RegionServer,sub=Regions
2017-02-15 10:19:18,282 DEBUG [Timer for 'HBase' metrics system]
impl.MetricsSystemImpl: Snapshotted source MetricsSystem,sub=Stats
This shows that the metrics subsystem has subscribed to all "subs".
However only 2 of those make it:
2017-02-15 10:17:58,262 DEBUG [mdmsink] impl.MetricsSinkAdapter: Pushing record
RegionServer,sub=WAL.regionserver.WAL to mdmsink
2017-02-15 10:17:58,293 TRACE [mdmsink] metrics.MdmServiceClient:
{"timestamp":1487182678262,"namespace":"regionserver.WAL","dimensions":...
2017-02-15 10:17:58,356 ERROR [mdmsink] metrics.MdmServiceClient: Getting bad
status code from the service: 204
2017-02-15 10:17:58,356 DEBUG [mdmsink] impl.MetricsSinkAdapter: Pushing record
RegionServer,sub=IPC.regionserver.RegionServer to mdmsink
2017-02-15 10:17:58,356 TRACE [mdmsink] metrics.MdmServiceClient:
{"timestamp":1487182678262,"namespace":"regionserver.RegionServer","dimensions":...
2017-02-15 10:17:58,481 ERROR [mdmsink] metrics.MdmServiceClient: Getting bad
status code from the service: 204
2017-02-15 10:17:58,481 DEBUG [mdmsink] impl.MetricsSinkAdapter: Pushing record
UgiMetrics.ugi.UgiMetrics to mdmsink
2017-02-15 10:17:58,481 TRACE [mdmsink] metrics.MdmServiceClient:
{"timestamp":1487182678262,"namespace":"ugi.UgiMetrics","dimensions":[{"key":"Context","value":"ugi"},...
________________________________
From: jeff saremi <[email protected]>
Sent: Wednesday, February 15, 2017 8:33:25 AM
To: [email protected]
Subject: Fw: Certain metric groups are missing when using a custom sink in
place of FileSink
forwarding here since this seems to be a generic metric-related issue and not
necessarily HBase.
________________________________
From: jeff saremi <[email protected]>
Sent: Tuesday, February 14, 2017 9:29 PM
To: [email protected]
Subject: Certain metric groups are missing when using a custom sink in place of
FileSink
When we use a FileSink to log metrics from HBase we can see more names than
when we use a custom metric sink. Is there something undocumented that we're
missing?
For instance when using FileSink, we can see WAL, RegionServer,
Replication,Server , and Regions in the log.
However if we use a custom sink (which happens to be writing whatever it gets
to the log as well) we can only see WAL and RegionServer.
These are the config lines we're using:
hbase.sink.mdmsink.class=mycomp.metrics.MdmSink
hbase.sink.mdmsink.server=myserver:8880
*.period=10
