[
https://issues.apache.org/jira/browse/HBASE-24205?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17119641#comment-17119641
]
ramkrishna.s.vasudevan commented on HBASE-24205:
------------------------------------------------
A new push which has the following changes
-> Both gets and scans are covered in the same patch (we don do anything
special for these just treat as same).
-> Whether it is a scan or get - when we are at the StoreScanner loop where we
check cell by cell till a row is done - we check if all the cells in that
current loop till we return are from memstore or from file. Only then the
corresponding metric is incremented. We no longer track the total gets that has
landed on the StoreScanner. Say if out of 10 cells even if one cell is from
memstore and remaining from file we don't account it with this metric.
-> The table level metric will now have a new aggregated metric
{quote}"Namespace_default_table_TestTable_store_info0_metric_readRequestCountOnMemstore":1766,
"Namespace_default_table_TestTable_store_info0_metric_readRequestCountOnFiles":
37828,
{quote}
This will be mapped per store for each table.
-> the patch also handles the cases where same family names can be across
tables. Earlier that was not getting handled and now at the MetricStoreSource
layer we add the region name and store name for the uniqueness.
> Create metric to know the number of reads that happens from memstore
> --------------------------------------------------------------------
>
> Key: HBASE-24205
> URL: https://issues.apache.org/jira/browse/HBASE-24205
> Project: HBase
> Issue Type: Improvement
> Components: metrics
> Affects Versions: 3.0.0-alpha-1
> Reporter: ramkrishna.s.vasudevan
> Assignee: ramkrishna.s.vasudevan
> Priority: Major
> Fix For: 3.0.0-alpha-1, 2.3.0, 2.2.6
>
> Attachments: screenshot.png, screenshot_tablevscf.png
>
>
> A metric to identify number of reads that were served from memstore (atleast
> if the gets can be accounted for) then it gives a value addition to know if
> among the reads how much was targeted at the most recent data.
> Currently the existing metric framework at region level should be enough but
> we can also add a metric per store level. That will be more granular.
> We can also expose this via HbTop.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
