https://dagster.io/ logo
Title
m

Mykola Palamarchuk

01/18/2022, 10:53 PM
Hi! We need to execute jobs in different clusters due to security restrictions. Can I use k8s_job_executor for that somehow?
j

johann

01/18/2022, 10:56 PM
Hi Nick, can you expand on this? Would the different jobs ideally all be visible within one Dagit instance?
m

Mykola Palamarchuk

01/18/2022, 11:03 PM
We have a couple of accounts with different resources. Dagit runs in a k8s cluster in one of the accounts. We want to collect data from databases in all the accounts, but the direct cross-account database access is prohibited. So I think about possibility to run jobs in different clusters each under its own account.
I see that worker should have access to the dagster database even if it runs on the remote cluster. That is also a case for celery as I understand. Is that correct?
j

johann

01/19/2022, 4:19 AM
Hmm, if databases can't be accessed across accounts it's a little tricky to consolidate the data to a single Dagit. Is running a separate Dagit instance in each account a possibility or would that defeat the purpose of the project?
Otherwise I think this likely calls for a custom run launcher or executor, and you'd provide whatever scheme for launching compute across clusters and getting events back
It's worth mentioning that Dagster Cloud operates with an agent model that could facilitate something like this- the agent in each cluster would just need outward access to the api endpoint. You could reach out to @max if you'd like to chat.