HBase Replication ageOfLastShippedOp

[Asaf Mesika <as...@gmail.com>]

Hi,

I noticed that when a Region Server takes over a dead region server
replication queue, the ageOfLastShippedOp is reported of the latter. For
instance, I had an age of 500ms, and when I shutdown region server 2, then
region server took over the queue, and suddenly I had an age of 700k ms.
When you peek the metrics using JConsole and not Graphite, you see two
entries under Replication, thus we have two ageOfLastShippedOp.

Does HBase Region Server arbitrarily chooses which one to publish to its
Metrics Context?


Thank you,

Asaf

Re: HBase Replication ageOfLastShippedOp

[Asaf Mesika <as...@gmail.com>]

The problem is that when the recovered edits log is done processing
(Replicating) (queue = 0), the metrics still reports its
agetOfLastShippedOp instead of the current live active log, which seems
like a bug.



On Tue, Jun 18, 2013 at 9:55 PM, Elliott Clark <ec...@apache.org> wrote:

> On Tue, Jun 18, 2013 at 1:09 AM, Asaf Mesika <as...@gmail.com>
> wrote:
>
> > Does HBase Region Server arbitrarily chooses which one to publish to its
> >
>
> The metrics reports the last source to report in.  So for recovered edits
> since it will probably have higher throughput than the live log, you'll
> more often see that value.  In trunk and 0.95/0.96 this has been fixed and
> every source exposes it's own ageOfLastShippedLog as well as one global
> one.
>

Re: HBase Replication ageOfLastShippedOp

[Elliott Clark <ec...@apache.org>]

On Tue, Jun 18, 2013 at 1:09 AM, Asaf Mesika <as...@gmail.com> wrote:

> Does HBase Region Server arbitrarily chooses which one to publish to its
>

The metrics reports the last source to report in.  So for recovered edits
since it will probably have higher throughput than the live log, you'll
more often see that value.  In trunk and 0.95/0.96 this has been fixed and
every source exposes it's own ageOfLastShippedLog as well as one global one.