APM agent for Elasticsearch

[rschirin (Robert)]

Hi all,
I would like to know, is there a way to think about APM agent applied directly on Elasticsearch? I mean, to monitor the single transaction starting from the first step (Elasticsearch is contacted to receive a document), should I use the xpack.monitoring or is there something different?

[stephenb (Stephen Brown)]

Hi Roger thanks for trying out Elastic APM

I'm a little unclear on your question perhaps you can help clarify

Do you want a monitor a whole distributed trace from say the web browser to your application to access the document from Elasticsearch and see that entire trace from web browser to example Java service to Elasticsearch and back?

Or

Are you trying to monitor the internals of Elasticsearch once that search request reaches Elasticsearch?

[rschirin (Robert)]

hi @stephenb, my initial idea was the second one. I was thinking about the internals of Elasticsearch getting, for instance, the time of every action (search, fetch and so on). as I said, I know sure if APM should be considered the correct tool.

[stephenb (Stephen Brown)]

I am a little curious what you are trying to accomplish do you have a real use case / real performance issue? or is this just for experiment / for your own education.

Typically you wouldn’t run the APM Agent on Elasticsearch itself to monitor Elasticsearch performance. (I am not sure how viable that would be, I would need to look further, there would be some challenges I suspect due to distributed nature with TCP protocols)

In general to monitor Elasticsearch search and index performance teams could use one or more of the following along with other techniques

Elasticsearch Monitoring: Query Per Second, Average Query Latency etc… Host Metrics, Logs etc.

https://www.elastic.co/guide/en/elastic-stack-overview/current/xpack-monitoring.html

Query Profile, Thread Pools, Hot Nodes,

https://www.elastic.co/guide/en/elasticsearch/reference/7.2/search-profile.html

https://www.elastic.co/guide/en/elasticsearch/reference/7.2/cat-thread-pool.html

https://www.elastic.co/guide/en/elasticsearch/reference/7.2/cluster-nodes-hot-threads.html

https://www.elastic.co/guide/en/elasticsearch/reference/7.2/cat-pending-tasks.html

Elasticsearch Slow Logs

https://www.elastic.co/guide/en/elasticsearch/reference/7.2/index-modules-slowlog.html

[Eyal_Koren]

@rschirin adding to @stephenb references, all Elastic APM agents (besides the .NET agent which is still in beta) provide instrumentation to the corresponding Elasticsearch clients. Is there any information/visibility you are missing by using those or the Elasticsearch monitoring options provided above?

We are thinking on a way to add dedicated instrumentation to Elasticsearch, but we are more thinking about how it can assist us to debug issues, rather than provide more visibility to end users.