[
https://issues.apache.org/jira/browse/HBASE-12199?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14165314#comment-14165314
]
Elliott Clark edited comment on HBASE-12199 at 10/9/14 4:16 PM:
----------------------------------------------------------------
Different versions of Hadoop have different syncing behaviors. Some will sync
the local filesystem (1.0 and FB's old 0.20) others won't sync (2.0+). The
patch speeds up the older hadoops a lot by making everything even. It also
speeds up the newer hadoops a little bit by not waiting for the hlog syncer at
all.
I think this behavior was chosen because it can't change semantics of any test
while it can provide a speed up. Since our tests are a single process that
never continues after crash syncing is wasted time; linux and osx will always
give a read by this process the dirty page.
was (Author: eclark):
Different versions of Hadoop have different syncing behaviors. Some will sync
the local filesystem (1.0 and FB's old 0.20) others won't sync (2.0+). The
patch speeds up the older hadoops a lot by making everything even. It also
speeds up the newer hadoops a little bit by not waiting for the hlog syncer at
all.
I think this behavior was chosen because it can't change semantics of any test
while it can provide a speed up. Since our tests are a single process that
never continues after crash syncing is wasted time; linux and osx will always
give a read by this process the dirty pag.
> Make TestAtomicOperation and TestEncodedSeekers faster
> ------------------------------------------------------
>
> Key: HBASE-12199
> URL: https://issues.apache.org/jira/browse/HBASE-12199
> Project: HBase
> Issue Type: Bug
> Components: test
> Affects Versions: 1.0.0
> Reporter: Manukranth Kolloju
> Assignee: Manukranth Kolloju
> Priority: Trivial
> Fix For: 2.0.0, 0.98.7, 0.99.1
>
> Attachments: 0001-Fix-TestAtomicOperation.patch,
> 0001-INTERNAL-Fix-TestAtomicOperation.patch
>
>
> TestAtomicOperation has a bunch of tests which spawn a bunch of tests that do
> puts, appends, deletes, inc and then does gets to make sure that things
> happened. It uses TEST_UTIL.getLocalHRegion() to obtain an HRegion object.
> So, in essence the test doesn't need to sync to disk to make sure the edits
> have persisted. There is no regionserver/master to take care of wal replay or
> any failure for that matter. So, converting all the Durability settings for
> the edits to ASYNC_WAL seems like the right thing to do here to make the test
> run faster.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
