Description
I see this as an incremental improvement toward a larger more powerful metrics framework in phoenix-server, phoenix-client, and phoenix-connectors. Today we describe our available metrics as https://phoenix.apache.org/metrics.html
Long term a tag based multi-dimensional data model (similar to prometheus/open tsdb) may be superior but for now this targets simple generalization of the existing metrics in phoenix. Today much of the metrics are built on hbase/hadoop offering.
Some initial discussion/brainstorming:
Metric Types:
Phoenix uses/provides sufficient verbosity of metrics, but does not define the metric types cleanly. Typical counter, gauge, histogram
Monitoring Models:
Events (Event Metrics), metrics with no summarization or aggregation.
Example of this will be request metrics. On this request you used 5 scanners.
Metrics (Summary/Aggregated Metrics),
Example of this might be total queries
Reporting:
Today JMX is one of the main approaches used in phoenix for metrics. For events most of them are retrievable from PhoenixRuntime. Long term these should be pluggable.
Storage:
End user based on reporting is envisioned.
May consider storing in HBase/Phoenix as an option? The query log could store events with that query log entry.
Attachments
Issue Links
- depends upon
-
HBASE-20904 Prometheus metrics http endpoint for monitoring integration
-
- Resolved
-