Preferred metrics source: HTTP+JSON or JMX?

Elastic Stack Elasticsearch
1

Hello,

ElasticSearch exposes cluster/nodes/shard/OS metrics via HTTP+JSON,
but also via JMX...though I think not everything that exists in the
former is present in the latter?

If I wanted to collect ES metrics, should I rely on things being in
JMX or should I always get data from HTTP+JSON API because JMX is
going to get abandoned or maybe always be secondary and thus
incomplete?

Thanks,
Otis

2

HTTP is the more complete one, which has many stats. Started with JMX as well, but slowly gave up on it a bit and prefer people will use the HTTP interface because, well, that it sucks...

On Friday, January 27, 2012 at 8:40 AM, Otis Gospodnetic wrote:

Hello,

Elasticsearch exposes cluster/nodes/shard/OS metrics via HTTP+JSON,
but also via JMX...though I think not everything that exists in the
former is present in the latter?

If I wanted to collect ES metrics, should I rely on things being in
JMX or should I always get data from HTTP+JSON API because JMX is
going to get abandoned or maybe always be secondary and thus
incomplete?

Thanks,
Otis

http://sematext.com/

3

Shay,

Parity here would be nice. No defense of JMX's nonsuckiness but most
organizations have an ops tool that consumes JMX and when adopting a new
technology it removes some frowns from the typically stressed and skeptical
operations folk. 2c.

Mike

On Fri, Jan 27, 2012 at 3:51 PM, Shay Banon kimchy@gmail.com wrote:

HTTP is the more complete one, which has many stats. Started with JMX as
well, but slowly gave up on it a bit and prefer people will use the HTTP
interface because, well, that it sucks...

On Friday, January 27, 2012 at 8:40 AM, Otis Gospodnetic wrote:

Hello,

Elasticsearch exposes cluster/nodes/shard/OS metrics via HTTP+JSON,
but also via JMX...though I think not everything that exists in the
former is present in the latter?

If I wanted to collect ES metrics, should I rely on things being in
JMX or should I always get data from HTTP+JSON API because JMX is
going to get abandoned or maybe always be secondary and thus
incomplete?

Thanks,
Otis

http://sematext.com/

4

Sure, if JMX is what people are after, it pretty simple to add the relevant stats on JMX as well (there is actually quite a framework built in ES to easily expose JMX data), just open an issue and it can be easily added. But, the REST API is the prefer way to do it (JMX has a very large overhead).

On Friday, January 27, 2012 at 10:55 PM, Michael Sick wrote:

Shay,

Parity here would be nice. No defense of JMX's nonsuckiness but most organizations have an ops tool that consumes JMX and when adopting a new technology it removes some frowns from the typically stressed and skeptical operations folk. 2c.

Mike

On Fri, Jan 27, 2012 at 3:51 PM, Shay Banon <kimchy@gmail.com (mailto:kimchy@gmail.com)> wrote:

HTTP is the more complete one, which has many stats. Started with JMX as well, but slowly gave up on it a bit and prefer people will use the HTTP interface because, well, that it sucks...

On Friday, January 27, 2012 at 8:40 AM, Otis Gospodnetic wrote:

Hello,

Elasticsearch exposes cluster/nodes/shard/OS metrics via HTTP+JSON,
but also via JMX...though I think not everything that exists in the
former is present in the latter?

If I wanted to collect ES metrics, should I rely on things being in
JMX or should I always get data from HTTP+JSON API because JMX is
going to get abandoned or maybe always be secondary and thus
incomplete?

Thanks,
Otis

http://sematext.com/