rowupdate timestamp

[Ski Gh3 <sk...@gmail.com>]

Hi all,

I have a question regarding the timestamp of an operation.
So if one does not supply anything, the region server will use
CurrentMilliSecond call to obtain the timestamp from the running machine,
right?
Then what happens if this region is reassigned to some other server which
has a different system time? it could result in later updates having earlier
timestamps.

I read from the cluster setup guide that the region servers can not have
time off by 1hour or more but they don't need to be strictly aligned. Then
is there any guarantee
on this issue?

Thanks!

RE: rowupdate timestamp

["Jim Kellerman (POWERSET)" <Ji...@microsoft.com>]

> -----Original Message-----
> From: jdcryans@gmail.com [mailto:jdcryans@gmail.com] On Behalf Of Jean-
> Daniel Cryans
> Sent: Tuesday, September 30, 2008 4:15 PM
> To: hbase-user@hadoop.apache.org
> Subject: Re: rowupdate timestamp
>
> Some skew mean some seconds, a minute at the very most. Time is very
> important in a distributed system so there is strictly no guarantee (for
> example, we don't have special mechanisms that handles the skew and
> corrects
> it).

Yet. See HBASE-711. However, any server cluster should run NNTPD IMO.

Re: rowupdate timestamp

[Ski Gh3 <sk...@gmail.com>]

Thanks for the answer!

On Tue, Sep 30, 2008 at 4:15 PM, Jean-Daniel Cryans <jd...@apache.org>wrote:

> Well in the API main page it is said "some skew" and not "1 hour". If this
> is the case elsewhere, please report it.

I do not remember exactly now, maybe I was confused.

>
>
> Some skew mean some seconds, a minute at the very most. Time is very
> important in a distributed system so there is strictly no guarantee (for
> example, we don't have special mechanisms that handles the skew and
> corrects
> it).
>
> NTP is a fast and easy solution.
>
> J-D
>
> On Tue, Sep 30, 2008 at 6:41 PM, Ski Gh3 <sk...@gmail.com> wrote:
>
> > Hi all,
> >
> > I have a question regarding the timestamp of an operation.
> > So if one does not supply anything, the region server will use
> > CurrentMilliSecond call to obtain the timestamp from the running machine,
> > right?
> > Then what happens if this region is reassigned to some other server which
> > has a different system time? it could result in later updates having
> > earlier
> > timestamps.
> >
> > I read from the cluster setup guide that the region servers can not have
> > time off by 1hour or more but they don't need to be strictly aligned.
> Then
> > is there any guarantee
> > on this issue?
> >
> > Thanks!
> >
>

Re: rowupdate timestamp

[Jean-Daniel Cryans <jd...@apache.org>]

Well in the API main page it is said "some skew" and not "1 hour". If this
is the case elsewhere, please report it.

Some skew mean some seconds, a minute at the very most. Time is very
important in a distributed system so there is strictly no guarantee (for
example, we don't have special mechanisms that handles the skew and corrects
it).

NTP is a fast and easy solution.

J-D

On Tue, Sep 30, 2008 at 6:41 PM, Ski Gh3 <sk...@gmail.com> wrote:

> Hi all,
>
> I have a question regarding the timestamp of an operation.
> So if one does not supply anything, the region server will use
> CurrentMilliSecond call to obtain the timestamp from the running machine,
> right?
> Then what happens if this region is reassigned to some other server which
> has a different system time? it could result in later updates having
> earlier
> timestamps.
>
> I read from the cluster setup guide that the region servers can not have
> time off by 1hour or more but they don't need to be strictly aligned. Then
> is there any guarantee
> on this issue?
>
> Thanks!
>