hbase-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Stack <st...@duboce.net>
Subject Re: Region-level metrics being phased out?
Date Thu, 19 Feb 2015 01:59:38 GMT
On Wed, Feb 18, 2015 at 1:57 PM, Otis Gospodnetic <
otis.gospodnetic@gmail.com> wrote:

> Well, here are a few related things, since you're asking :)
> 1. Are there table-level metrics?  I didn't see any in JMX. Or is one
> supposed to collect all region-level metrics and compute table-level
> metrics from that?

If not in JMX, it don't exist.

Yeah, no table-level.  Aggregate.

> 2. http://hbase.apache.org/book.html#_hbase_metrics lists mainly
> RegionServer-level metrics and NO region-level metrics.  Are region-level
> metrics generally less useful in practice?  Or maybe they are hard for a
> human to analyze when a cluster has hundreds or more regions?

The latter -- they can be overwhelming when more than a few regions on a RS
-- and just oversight; we should add some region-level metrics docs.

> 3. Are there some region-level metrics that are considered really important
> to keep an eye on?

 Here is a region from /metrics on a RS:

: 1,
: 3,
: 76081792,
: 323832848,
: 34,
: 3709543822,
: 114,

Store count can probably be ignored. It doesn't change.

Store file count would be good to keep an eye on. Could alert if a
region is in excess of some threshold.

Ditto memstore size.

Store file size could be good for seeing how regions are growing over time.

The compactions related ones are probably of secondary interest (would
be good though I suppose to be able to see regions that are compacting
a lot).

So, I suppose all but one or two important (smile).

Go easy Otis,


  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message