Jerry,
merge and compaction bypass region lock, just check the code .
in startRegionOperation:
if (op == Operation.MERGE_REGION || op == Operation.SPLIT_REGION
|| op == Operation.COMPACT_REGION) {
// split, merge or compact region doesn't need to check the
closing/closed state or lock the
// region
return;
}
-Vlad
On Wed, Feb 25, 2015 at 1:47 PM, Jerry He <[email protected]> wrote:
> Hi, Vlad
>
> For question 1. see HBASE-8760
> <https://issues.apache.org/jira/browse/HBASE-8760>.
> For question 2, the region is under startRegionOperation() --> read lock.
>
> Thanks,
>
> Jerry
>
> On Wed, Feb 25, 2015 at 10:39 AM, Vladimir Rodionov <
> [email protected]>
> wrote:
>
> > Hi, folks
> >
> > 1. In EnabledTableSnapshotHandler in
> >
> > protected void snapshotRegions(List<Pair<HRegionInfo, ServerName>>
> regions)
> >
> > why do we make snapshot for offline and split regions after all online
> > regions are done?
> >
> >
> > 2. Do we have a possible race condition in SnapshotManifest:
> >
> > a. When we get a list of a store files in
> >
> > public void addRegion(final HRegion region)
> >
> > b. then we iterate over this list and do some work
> >
> > c. then we close snapshot for region and save manifest to a file
> >
> > between a. and c. some files from list in a. can become invalid (deleted
> by
> > HFileCleaner).
> >
> > I checked the code, no exception is raised if file is no longer exists
> > during this phase.
> >
> > thanks,
> > -Vlad
> >
>
