https://prefect.io logo
Title
m

Manuel Ledesma

01/05/2022, 5:19 PM
Good morning, I'm trying to deploy prefect helm chart on EKS, deployment is fine but when trying to setup the UI service as load balance, I'm always getting 400 as response, according to nginx documentation the proxy_protocol must be setup to allow ELB traffic, but I don't see an way to change the nginx default.conf file.
a

Anna Geller

01/05/2022, 6:18 PM
Why would you load balance the UI service? Not sure whether this is required. I think the main issue to maintain availability would be ensuring that Apollo server scales rather than the UI. I would be curious to hear from others who deployed Server
m

Manuel Ledesma

01/06/2022, 9:42 PM
I fixed the issue, now how we set the apollo URL permanent through helm chart deployment?
a

Anna Geller

01/06/2022, 10:34 PM
I believe you can set it through helm chart values, specifically this one
nice work fixing the load balancing btw!
m

Manuel Ledesma

01/07/2022, 12:23 PM
thx
another quick question, why HPA is not setup ?
a

Anna Geller

01/07/2022, 1:07 PM
I think the helm chart is not-opinionated in terms of how you want to scale and that’s why it’s up to you to configure HPA if you need it. Maybe someone from the community did it and can share? Let’s see
m

Manuel Ledesma

01/07/2022, 2:22 PM
thx
Do you have a guide on production settings for each component (memory/cpu) ?
a

Anna Geller

01/07/2022, 5:02 PM
We don’t have a separate guide for this. There is only the recommendation to leverage a separate Postgres database and to use the same prefectVersionTag for Server as the version you are using for the agent. In general, the README is quite informative, but if you feel like something is missing, we are open to your contribution.
m

Manuel Ledesma

01/07/2022, 6:15 PM
what about memory/cpu requirements for each component assuming an average load
Any page that can help getting the recommended settings for each one
a

Anna Geller

01/07/2022, 6:29 PM
Good question! I don’t know the answer to that. Even the “average load” is kind of subjective and can be understood differently by different users. Maybe someone from the community can guide you through the values they used. Until then, perhaps you can start low and increase the values if you see bottlenecks?