My plan is like this, suppose a master goes down just after creating the
regioninfo files in the tmp folder then we can try to use that info for
recreating the table regions on master restart.
But one problem here could be that before creating all the regioninfos in
the .tmp location if the master goes down we should be able to identify
this scenario also.
On Mon, Mar 25, 2013 at 5:38 PM, Matteo Bertozzi <[EMAIL PROTECTED]>wrote:
> snapshots knows about the /hbase/.tmp directory but unless you remove the
> without archiving or move to a location that is not the original one
> everything is fine.
> for me is ok moving the .tmp deletion later.
> do you want to restore the unfinished delete instead instead of finishing
> the deletion?
> On Mon, Mar 25, 2013 at 11:55 AM, ramkrishna vasudevan <
> [EMAIL PROTECTED]> wrote:
> > Hi Devs
> > Currently the HMaster initilization deletes the files under .tmp using
> > HFileArchiver.
> > in HMaster().new MasterFileSystem().createInitialFileSystemLayout().
> > Actually i was thinking of using this info in the .tmp directory for the
> > tables that were partially created so that on master restart i can use
> > info to recreate the table fully.
> > But i found that we are deleting this .tmp directory.
> > My question here is will it be ok to delete the .tmp directory after
> > the this.assignmentManager.joinCluster(), call in
> > HMaster.finishInitialization?
> > Because by this time i would be knowing what are the partially created
> > tables.
> > Just wanted to confirm before i can make this change? My worry was, may
> > snapshot is trying to utilize this info and doing something.
> > Regards
> > Ram