https://dagster.io/ logo
a

Agon Shabi

04/04/2021, 5:58 PM
Has any work been done to date to support executions via knative (and therefore google cloud functions)?
m

mrdavidlaing

04/04/2021, 10:44 PM
Depends what you mean by "extensions". It is possible to run dagster on Cloud Run in "single container" mode
I deploy a simple "web runner" flask app (codeDocker setup) that runs 
execute_pipeline()
 when triggered by a HTTP request from Cloud Scheduler (docs). I run 
dagit
 locally (connecting to the same GCP Postgres instance) to observe the run output.
No fancy "spread load across multiple containers"; but it can work out very inexpensive since you're only paying for the time your pipelines are actually running - my simple pipeline that runs once / day costs ~ €0.01 for 385.47 vCPU-second / day
a

Agon Shabi

04/05/2021, 5:45 PM
Nice, that's very similar to how I've been using it - I have a persistent SQL Server instance, on-demand ephemeral compute (on prem; triggerable via HTTP request), and a way to run dagit pointed at the database. I might try and license a managed k8s service, but since my workloads don't currently require any special hardware resources, I thought it might be simpler to forego the k8s process overhead and rely entirely on cloud functions for execution of solids, and maybe app engine for dagit.
m

mrdavidlaing

04/05/2021, 5:47 PM
One problem I haven't been able to solve with hosting Dagit on GCP Cloud run is (HTTP) authentication. If you come up with a solution that isn't too expensive I've love to learn what you did