Handle the base exception from the requests library #396
Conversation
We were handling one of the subclasses, and some errors slipped through, causing missed reductions.
| @@ -6,20 +6,20 @@ | |||
| # will be started when the CPU usage rises above the configured threshold. | |||
| horizontalPodAutoscaler: | |||
| enabled: true | |||
| minReplicas: 15 | |||
| maxReplicas: 15 | |||
| minReplicas: 25 | |||
There was a problem hiding this comment.
We should probably talk about how to introduce autoscaling someday when you have the bandwidth...
There was a problem hiding this comment.
Do we need to set a max replicas here? Or just assume if we only set a min, they are equal?
There was a problem hiding this comment.
We should probably talk about how to introduce autoscaling someday when you have the bandwidth...
Jashan did prototype this at one time. https://github.com/LCOGT/banzai/pull/359/files
Autoscaling based on queue size sounds great Joey, but I am very hesitant to commit my time to this as a development priority. I think it is relatively high effort to low gain, given that we can simply run more pods than we need most of the time.
I would love if someone other than me could spearhead that effort.
There was a problem hiding this comment.
Do we need to set a max replicas here? Or just assume if we only set a min, they are equal?
Yes we need to. https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.31/#horizontalpodautoscalerspec-v2-autoscaling
This has always been the case.
There was a problem hiding this comment.
Yeah. Ok. github split the file and I missed the next line.
helm-chart/banzai/values-prod.yaml
Outdated
|
|
||
| image: | ||
| repository: ghcr.io/lcogt/banzai | ||
| tag: "1.18.1" | ||
| tag: "1.18.2" |
There was a problem hiding this comment.
Should this match the version number above in the changes? That's currently 1.18.3...
This is to speed up and make more robust the serialization of frames endpoint responses, particularly for BPMs which have many thousands of related frames.
This reverts commit 1942cbc.
We were handling one of the subclasses, and some errors slipped through, causing missed reductions.
Also update our deploy to use in-cluster URLs to speed up the DNS resolution and make us more resilient to DNS issues.
Sorry for all of the whitespace changes, but it looks like we had some extra end-of-line whitespace in the changelog.