Improve ECK diagnostic collection

[damianpfister]

Raising this on behalf of a customer.

Needing to generate diagnostics file for the cluster was found to be painful.

I had to build and test my own version of a wrapper script around this diagnostics script to attempt to generate a diagnostics on a kubernetes ES pod for which we didn't have an external service for (in other words, no addressable domain from which we could access elasticsearch by an app outside of kubernetes).

I was eventually about to get something working with a variety of kubectl commands although I feel this should have been much easier to accomplish.

A new member of our Observability team has had to run a diagnostic recently using v8.2.2 of the package (I was previously using 7.1.5) and unfortunately that version does not work with my helper script.

Any idea what kind of time line we could expect until we have something available for this?

Code for the wrapper script is here - note it currently only works against 7.1.5 of the ECK diagnostics script.

[pebrc]

I am confused. The whole point of the diagnostic tool in this repository is that you don't have to do any of what happens in that script you linked to.

Also there is no version 7.1.5 of the ECK diagnostics script. Actually it is no longer a script. The code in this repo is compiled to a binary that you execute. All you need is a working KUBECONFIG and enough permissions on you Kubernetes cluster. See https://github.com/elastic/eck-diagnostics/blob/main/README.md

So I am thinking this feedback is not related to the current diagnostic tooling we have, but maybe predates it (we used to have only a shell script that did not do Elastic stack diagnostics...)

[hartfordfive]

I'll provide some additional context. Up to now (at least from my knowledge), any Elastic customer running Elasticsearch via ECK still had to generate diagnostics via the usual java diagnostics tool. Unfortunately, that process is quite painful especially when you don't have externally accessible service to your Elasticsearch cluster(s) (e.g.: Kibana is the only "external" entry point). In this case, running the diagnostics ends up having to be performed via a series of kubectl commands. Although the linked script works, it's likely not the most effective or preferable method. Figuring out the proper kubectl commands to run in order to generate the diagnostics during an actual production outage of multiple clusters was unpleasant to say the least. Hopefully the eck-diagnostics tool will make this process much simpler. I would even go as far as to suggest giving the tool the ability to upload the diagnostics package to a configured S3 or GCS bucket so that users have a longer-term storage of these diagnostics as your support portal deletes all attachments after a certain amount of time.

Thanks for your help.