Compute helm chart. Which services must be run?

[roman-shakudo]

Hi

I managed to run neon on my k8s cluster based on #35

I see pageserver and three safekeepers running
I had to force using k8s service dns as BROKER_ENDPOINT and S3_ENDPOINT because default values are not correct

Since I was changing charts manually pageserver and safekeeper added dependecies which were run as dependant charts

Storage controller is falling with Error: loading public key. I suppose JWT issue
Storage Scrubber's image is not available

Questions:

1. Am I understand right that we have to wait for Compute chart? AFAIK I can't use any posgtres
2. Which services are essential to run neon? I see a lot of various services and I do not know which are core. It would great to add some doc about it
3. How can deployment be tested?
4. Can I pack Compute from docker-compose.yaml in neon as a helm cart and use it or this image is just for "testing"?
5. Am I understand right that there is no UI in the helm charts?

[roman-shakudo]

From docs I could figure out that the core services are

• pageserver
• safekeeper
• compute
• storage broker
• storage controller
• minio (for local or k8s)

Have no idea what others are

• pn-sni-router
• proxy
• storage-scrubber

And the whole architecture is kinda like that

[GangWangAI]

I wrote a compute chart that runs normally for about 6 minutes before crashing and automatically restarting. The error log shows:

PG:2024-11-05 09:48:58.052 GMT [69] PANIC:  [WP] collected propEpochStartLsn 0/1537150, but basebackup LSN 0/14EE288
PG:2024-11-05 09:48:58.053 GMT [43] LOG:  background worker "WAL proposer" (PID 69) was terminated by signal 6: Aborted
PG:2024-11-05 09:48:58.053 GMT [43] LOG:  terminating any other active server processes

Did you run it successfully afterwards? @roman-shakudo

[roman-shakudo]

GangWangAI
I did not run it
Somewhere I found information that k8s chart is not priority right now and was told to use cloud one =)
So I skipped it till it be done

Related Compute node
Did you apply patches neondb team created for a potstgres mentioned here?

Did you see that Safekeepr node receives and process WAL from Compute service?

[jcsp]

We'e clarified the README to explain that the helm charts are currently not meant to bring up a full working system: they just represents the components where we at Neon use helm at present: #119

I think this issue refers to safekeeper/pageserver patches from #35, so I'm going to close this one in favor of that one -- let's use that one as a place for folks to collaborate when they would like to try bringing up fully working systems via helm