search

apache / solr-operator

Official Kubernetes operator for Apache Solr
https://solr.apache.org/operator
Apache License 2.0
242 stars 112 forks source link

Remove ingress and node-services during reconcile #674

Open janhoy opened 5 months ago

janhoy commented 5 months ago

Fixes #673

DRAFT, code largely generated by GitHub copilot, just throwing up the idea

janhoy commented 4 months ago

Any takers for a review? It works when testing locally :)

janhoy commented 4 months ago

switch it to just use the headless and make sure that the node services no longer exist.

I might take a stab at such a test one day. Extending one of the existing, so after verifying that node-services are there, call an "update" on the CRD, then wait for reconsile to happen, and check that they are gone. Is there some test method that waits for reconcile or do we need to do polling with timeout?

HoustonPutman commented 4 months ago

switch it to just use the headless and make sure that the node services no longer exist.

I might take a stab at such a test one day. Extending one of the existing, so after verifying that node-services are there, call an "update" on the CRD, then wait for reconsile to happen, and check that they are gone. Is there some test method that waits for reconcile or do we need to do polling with timeout?

If you look at the existing unit tests, that's exactly how all of the Eventually...() functions work. They are eventually consistent checks. There should be some similar tests around changing state halfway through, such as creating necessary secrets for basicAuth after the SolrCloud has already been created. I can help you with the test, but obviously if you get started first, it'll help better with the learning 🙂

HoustonPutman commented 3 months ago

Btw this is going to be impacted by the work in #692 . The two approaches should not necessarily conflict with eachother though. We might not want to prune old node services too heavily, in case the user is using autoscaling. At that point it's likely better to keep the old ones around. (which will result in less rolling restarts due to the hostAliases being updated when the new node services are created.)