Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
-
None
-
None
Description
Phoenix fat client exposes a number of metrics at StatementLevel as well as ConnectionLevel. PQS uses avatica server and exposes two metrics at this point of time, which is request end to end time and request serialization/deserialization time. These metrics are not super useful for PQS operations POV, however we need them to track the bad client and mis-usage.
This Jira is to discuss how we should handle those metrics.
1. As pointed by jamestaylor, we can use SYSTEM.LOG table as a potential option here.
2. Write a shim layer to convert the metrics to hbase-metrics, which internally uses hadoop-metrics2 and define fine sink as the output.
Additional ideas welcome.