19913d85-f662-4518-a9a1-7adc1793dfd4.png){kind=small}
Jumping into a k8s-based deployment -- going to dr...
# announcementst
Travis Cline
05/28/2020, 8:21 PMJumping into a k8s-based deployment -- going to drop some initial thoughts into a thread here.
Travis Cline
05/28/2020, 8:22 PM1. Initially expected to be able to control the run launcher configuration via the execution key in the pipeline config object.
Travis Cline
05/28/2020, 8:23 PM2. not having more dynamic configuration (AFAICT so far) prevents some desirable patterns such as per-pipeline service account specification.
Travis Cline
05/28/2020, 8:43 PM3. Is it true that the dagit UI can't launch pipeline runs that use run_launcher configuration?
a
alex
05/28/2020, 8:46 PMdagit UI can’t launch pipeline runsthere should be a drop down to select between
startlaunchalex
05/28/2020, 8:46 PM@nate for other notes
t
Travis Cline
05/28/2020, 8:47 PMAh yes, thanks @alex
a
alex
05/28/2020, 8:47 PMwe are working to remove that distinction in
0.8.0alex
05/28/2020, 8:47 PMit is confusing and goofy
n
nate
05/29/2020, 5:04 PMhey Travis - on 1, definitely agree the split config between instance YAML and pipeline execution config is pretty rough right now, so agree here.
for 2, are there other fields you’d like to be able to configure per-pipeline-run? or just service account?
nate
05/29/2020, 5:04 PMcc @cat
t
Travis Cline
05/29/2020, 7:48 PMdefinitely interested in per-run resource requests/limits (this would allow auto-scaling up when a pipeline run would use a GPU, for example)
Travis Cline
05/29/2020, 7:52 PMSupplying the image on a per-pipeline is a likely requirement for many folks (would enable distinct dependency management between pipelines)
n
nate
06/05/2020, 3:01 PMhey Travis, just following up here, I’ve updated our Celery/K8s configs: https://dagster.phacility.com/D3192 - so you should be able to configure these parameters on a per-pipeline basis now
t
Travis Cline
06/07/2020, 11:31 PMbadass!
n
nate
06/08/2020, 2:23 PMlet us know if you have other feedback, we’re actively working on improving this integration!