[
https://issues.apache.org/jira/browse/HBASE-14314?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14728022#comment-14728022
]
Nick Dimiduk commented on HBASE-14314:
--------------------------------------
{quote}
+ String BLOCK_CACHE_PRIMARY_HIT_COUNT = "primary.blockCacheHitCount";
{quote}
Name is inconsistent with other metrics, isn't it? Nowhere else do we have
dotted metrics names. How about 'blockCacheHitCountPrimary'? Ditto for
primary.blockCacheMissCount, primary.blockCacheEvictionCount.
{quote}
+ String L1_BLOCK_CACHE_EVICTION_COUNT = "l1.blockCacheEvictionCount";
...
+ /**
+ * Get the number of items evicted from the L1 block cache.
+ */
+ long getL1BlockCacheEvictedCount();
{quote}
L1 level statistics as well? Are we tracking all L1 metrics or just evictions?
Don't we want hits/misses/&c as well? What about L2 stats? I see nothing about
L1 in the changes to {{CacheStats.java}}, so maybe this was not intended to be
part of the patch. How about we examine this in a separate JIRA?
Over in {{BlockCacheKey.java}}, there's
{quote}
+ private final boolean isReplicaBlock;
...
+ public BlockCacheKey(String hfileName, long offset, boolean isReplica) {
{quote}
but then
{quote}
+ public boolean isPrimary() \{
+ return !isReplicaBlock;
+ \}
{quote}
Why the logical inversion? Can we keep this consistent -- all {{isPrimary}} or
all {{isReplica}}? Seems like the stores want to refer to these as "replica or
not" while the cache as "primary or not". Would be good to choose one way or
the other.
> Metrics for block cache should take region replicas into account
> ----------------------------------------------------------------
>
> Key: HBASE-14314
> URL: https://issues.apache.org/jira/browse/HBASE-14314
> Project: HBase
> Issue Type: Improvement
> Components: metrics, regionserver
> Reporter: Ted Yu
> Assignee: Ted Yu
> Attachments: 14314-v1.txt, 14314-v2.txt, 14314-v3.txt
>
>
> Currently metrics for block cache are aggregates in the sense that they don't
> distinguish primary from secondary / tertiary replicas.
> This JIRA separates the block cache metrics for primary region replica from
> the aggregate.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
