[
https://issues.apache.org/jira/browse/HBASE-14836?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HBASE-14836:
--------------------------
Component/s: documentation
> Document how to deal with WARN "BucketCache: Failed allocation for...too big
> size=1113254; adjust BucketCache sizes ..to accomodate..."
> ---------------------------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-14836
> URL: https://issues.apache.org/jira/browse/HBASE-14836
> Project: HBase
> Issue Type: Bug
> Components: BlockCache, BucketCache, documentation
> Reporter: stack
>
> A busy, internal *smart* user was having their logs filled with messages like
> this:
> {code}
> BucketCache: Failed allocation for c326ba1d8a134b4487539239fce60a99_0;
> org.apache.hadoop.hbase.io.hfile.bucket.BucketAllocatorException: Allocation
> too big size=1113254; adjust BucketCache sizes hbase.bucketcache.bucket.sizes
> to accomodate if size seems reasonable and you want it cached.
> {code}
> .. and wasn't sure how to address them.
> Had to explain what the above is about (we don't have a bucket of a size into
> which we could put this object).
> Need to doc that c326ba1d8a134b4487539239fce60a99_0 is a filename and an
> offset. Need to doc how to look at this object that we are trying to cache
> using hfile tool.
> Then had to describe how you'd add a bucket that was big enough. Below I add
> 786432 and 1048576 beyond the default bucket sizes.
> <property>
> <name>hbase.bucketcache.bucket.sizes</name>
> <value>4096, 8192, 16384, 32768, 40960, 49152, 57344, 65536, 98304, 131072,
> 196608, 262144, 393216, 524288, 786432, 1048576</value>
> </property>
> The operator asked some questions about how can I tell rates at which Cell
> sizes are being served.
> Also, we were looking at the bucket numbers in the bucket L2 tab and it
> became obvious that bucket sizes, accesses, and frees should be summary
> stats, not per bucket instance. TODO.
> Anyway, this issue is about doc'ing how to deal w/ the above. We'll see it
> more when bucket cache is on always as we'd like to do in 2.0 hbase.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
