[
https://issues.apache.org/jira/browse/HBASE-19658?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16344949#comment-16344949
]
ramkrishna.s.vasudevan commented on HBASE-19658:
------------------------------------------------
bq.The question is exactly what are those "some cases"?
You have already answered your question.
bq.However, third cell shouldn't cause in-memory flush. So at the end two cells
in pipeline and one in active, no flakiness.
It is a classic case of multi threading.
{code}
if (this.active.keySize() > inmemoryFlushSize) { // size above flush
threshold
if (inWalReplay) { // when replaying edits from WAL there is no need in
in-memory flush
return false; // regardless the size
}
{code}
First cell is added the above check fails. Second cell is added the above check
passes. So
{code}
InMemoryFlushRunnable runnable = new InMemoryFlushRunnable();
if (LOG.isTraceEnabled()) {
LOG.trace(
"Dispatching the MemStore in-memory flush for store " +
store.getColumnFamilyName());
}
getPool().execute(runnable);
{code}
So you execute this runnable. the runnable starts running. Now by the time you
added third cell.
{code}
try {
if (LOG.isDebugEnabled()) {
LOG.debug("IN-MEMORY FLUSH: Pushing active segment into compaction
pipeline");
}
pushActiveToPipeline(this.active);
} finally {
getRegionServices().unblockUpdates();
}
{code}
Now 'this.active' has 3 cells. Now this is where flattening happens. In this
case your calculation goes wrong. That is why I was asking why to add a third
cell. Just add 2 cells it will for sure trigger your in memory flush and your
calculation will never go wrong.
> Fix and reenable
> TestCompactingToCellFlatMapMemStore#testFlatteningToJumboCellChunkMap
> --------------------------------------------------------------------------------------
>
> Key: HBASE-19658
> URL: https://issues.apache.org/jira/browse/HBASE-19658
> Project: HBase
> Issue Type: Bug
> Components: test
> Affects Versions: 2.0.0-beta-1
> Reporter: stack
> Assignee: Anastasia Braginsky
> Priority: Major
> Fix For: 2.0.0-beta-2
>
> Attachments: HBASE-19658-V01.patch, HBASE-19658-V02.patch,
> HBASE-19658-V03.patch, HBASE-19658-V04.patch, HBASE-19658-V05.patch,
> HBASE-19658.0007.patch, HBASE-19658.006.patch, HBASE-19658.05.patch,
> org.apache.hadoop.hbase.regionserver.TestCompactingToCellFlatMapMemStore-output.txt
>
>
> testFlatteningToJumboCellChunkMap was disabled so could commit HBASE-19282 on
> branch-2. This test is failing reliably. Assigned to [~anastas]. This issue
> is about fixing the failing test and reenabling it in time for beta-2. Thanks
> A.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
