You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2014/05/18 04:55:16 UTC
[jira] [Commented] (HBASE-4909) Detailed Block Cache Metrics
[ https://issues.apache.org/jira/browse/HBASE-4909?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14000957#comment-14000957 ]
stack commented on HBASE-4909:
------------------------------
Here is the commit Nicolas is referring to:
------------------------------------------------------------------------
r1181972 | nspiegelberg | 2011-10-11 10:45:00 -0700 (Tue, 11 Oct 2011) | 25 lines
Refactored and more detailed block read/cache and bloom metrics
Summary: As we keep adding more granular block read and block cache usage
statistics, there is a combinatorial explosion of the number of cases we have to
monitor, especially when we want both per-column family / block type statistics
and aggregate statistics on one or both of these dimensions. I am trying to
unclutter HFile readers, LruBlockCache, StoreFile, etc. by creating a
centralized class that knows how to update all kinds of per-column family/block
type statistics.
Test Plan:
Run all unit tests.
New unit test.
Deploy to one region server in dark launch and compare the new output of
hbaseStats.py to the old one (take a diff of the set of keys).
Reviewers: pritam, liyintang, jgray, kannan
Reviewed By: kannan
CC: , hbase@lists, dist-storage@lists, kannan
Differential Revision: 321147
Looking at svn diff -r1181971:1181972... the commit is all about:
+ BlockCategory blockCategory = dataBlock.getBlockType().getCategory();
...
and
+ cfMetrics.updateOnCacheMiss(blockCategory, isCompaction, delta);
... and this stuff in a class called ColumnFamilyMetrics:
+ READ_TIME("Read", true),
+ READ_COUNT("BlockReadCnt", true),
+ CACHE_HIT("BlockReadCacheHitCnt", true),
+ CACHE_MISS("BlockReadCacheMissCnt", true),
+
+ CACHE_SIZE("blockCacheSize", false),
+ CACHED("blockCacheNumCached", false),
+ EVICTED("blockCacheNumEvicted", false);
We have this. It is differently named, it is CacheStats. So, we have this detail. It came in with HBASE-4027, the slab cache issue. We need more but we have this much now so resolving as implemented.
> Detailed Block Cache Metrics
> ----------------------------
>
> Key: HBASE-4909
> URL: https://issues.apache.org/jira/browse/HBASE-4909
> Project: HBase
> Issue Type: Sub-task
> Components: Client, regionserver
> Reporter: Nicolas Spiegelberg
>
> Moving issue w/ no recent movement out of 0.95
--
This message was sent by Atlassian JIRA
(v6.2#6252)