JinHyuk Kim created HBASE-30132:
-----------------------------------
Summary: Expose per column family storeFileSize JMX metric
Key: HBASE-30132
URL: https://issues.apache.org/jira/browse/HBASE-30132
Project: HBase
Issue Type: Task
Components: metrics
Reporter: JinHyuk Kim
Assignee: JinHyuk Kim
h2. Summary
{{storeFileSize}} is exposed at table, region, and regionserver level, but not
per column family. We sums all stores per table, so on multi-CF tables the
per-family disk footprint is invisible.
h2. Motivation
In day-to-day operation we sometimes need per-family size to answer questions
the table-level metric can't:
- Which CF in this multi-CF table is actually growing?
- Is one CF dominating disk usage while others stay small?
Today the only way to get this is by walking HDFS or scraping the {{table.jsp}}
UI, neither of which is suitable for monitoring/alerting pipelines.
h2. Proposal
Add a per-CF gauge reusing the naming pattern HBASE-24205 introduced for read
counts:
{code:java}
Namespace_${ns}_table_${table}_columnfamily_${family}_metric_storeFileSize
{code}
The existing per-table {{storeFileSize}} gauge is preserved.
h2. Testing
Manual verification against a standalone HBase:
Create a test table with two column families and put some data:
{code:java}
create 'test', 'f1', 'f2'
put 'test', 'r1', 'f1:a', 'v1'
put 'test', 'r1', 'f2:a', 'value_with_long_value'
flush 'test'
{code}
Query the regionserver JMX endpoint and confirm both per-CF gauges appear:
{code:java}
curl -s http://localhost:16030/jmx | grep -E
'columnfamily_(f1|f2)_metric_storeFileSize'
{code}
Result:
{code:java}
"Namespace_default_table_test_columnfamily_f1_metric_storeFileSize": 4956,
"Namespace_default_table_test_columnfamily_f2_metric_storeFileSize": 4975,
{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
