Page MenuHomePhabricator

Migrate from Kfserving to Kserve
Closed, ResolvedPublic

Description

The first Kserve release is out: https://github.com/kserve/kserve/releases/tag/v0.7.0

I am strongly in favor of migrating to it before the release of our MVP, since we more easily upgrade at this stage.

Overall things to do:

  1. create new kserve docker images in the production-images repo (controller, storage-initializer, etc..)
  2. add a new helm chart for kserve, getting the yaml files from upstream (and apply all the custom changes that we do for kfserving)
  3. Upgrade inference-services' docker image to use kserve 0.7
  4. Deploy everything :)

Details

Related Changes in Gerrit:
SubjectAuthorRepoBranchLines +/-
Accrazemachinelearning/liftwing/inference-servicesmain+92 -55
Kevin Baziramachinelearning/liftwing/inference-servicesmain+29 -29
Accrazeoperations/deployment-chartsmaster+1 -1
Accrazeoperations/deployment-chartsmaster+2 -2
Accrazeoperations/deployment-chartsmaster+1 -1
Accrazemachinelearning/liftwing/inference-servicesmain+101 -8
Accrazemachinelearning/liftwing/inference-servicesmain+98 -8
Accrazemachinelearning/liftwing/inference-servicesmain+82 -46
Accrazemachinelearning/liftwing/inference-servicesmain+13 -13
Accrazemachinelearning/liftwing/inference-servicesmain+85 -43
Accrazemachinelearning/liftwing/inference-servicesmain+1 -1
Accrazemachinelearning/liftwing/inference-servicesmain+81 -43
Elukeyoperations/deployment-chartsmaster+23 -17 K
Elukeyoperations/docker-images/production-imagesmaster+0 -151
Elukeyoperations/deployment-chartsmaster+1 -1
Elukeyoperations/deployment-chartsmaster+45 -9
Elukeyoperations/deployment-chartsmaster+13 K -0
Elukeyoperations/docker-images/production-imagesmaster+86 -0
Show related patches Customize query in gerrit

Event Timeline

Change 730722 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/docker-images/production-images@master] Add the KServe images

https://gerrit.wikimedia.org/r/730722

Change 730761 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/deployment-charts@master] Add the kserve and kserve-inference charts

https://gerrit.wikimedia.org/r/730761

Change 730762 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/deployment-charts@master] Move the ml-serve cluster to KServe

https://gerrit.wikimedia.org/r/730762

Change 730722 merged by Elukey:

[operations/docker-images/production-images@master] Add the KServe images

https://gerrit.wikimedia.org/r/730722

Change 730761 merged by Elukey:

[operations/deployment-charts@master] Add the kserve and kserve-inference charts

https://gerrit.wikimedia.org/r/730761

Change 730762 merged by Elukey:

[operations/deployment-charts@master] Move the ml-serve cluster to KServe

https://gerrit.wikimedia.org/r/730762

Change 730807 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/deployment-charts@master] helmfile.d: update the deploy-kserve cluster role to KServe

https://gerrit.wikimedia.org/r/730807

Change 730807 merged by Elukey:

[operations/deployment-charts@master] helmfile.d: update the deploy-kserve cluster role to KServe

https://gerrit.wikimedia.org/r/730807

Change 730831 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/deployment-charts@master] Remove any trace of Kubeflow Kfserving

https://gerrit.wikimedia.org/r/730831

Change 730832 had a related patch set uploaded (by Elukey; author: Elukey):

[operations/docker-images/production-images@master] Remove Kubeflow Kfserving

https://gerrit.wikimedia.org/r/730832

Change 730832 merged by Elukey:

[operations/docker-images/production-images@master] Remove Kubeflow Kfserving

https://gerrit.wikimedia.org/r/730832

Change 730831 merged by Elukey:

[operations/deployment-charts@master] Remove any trace of Kubeflow Kfserving

https://gerrit.wikimedia.org/r/730831

From IRC (wikimedia-ml) 2021-11-22:

elukey: folks I think that we should migrate to KServer from kserve==0.7, I suspect that https://github.com/kserve/kserve/commit/c10e6271897d7fd058f5618d5e0e70b31496f64c is very important

Also @elukey flagged an issue in T296173#7520325 where our older versions of kfserving python library were causing strange results during initial load testing.

I think we should prioritize upgrading the inference services to kserve 0.7, I think we are in a good place with the new ml-sandbox to start attempting to build/test upgraded isvcs starting this week.

Change 734432 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] articlequality: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/734432

Change 743487 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] articlequality: upgrade transformer to kserve

https://gerrit.wikimedia.org/r/743487

Change 734432 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] articlequality: migrate kfserving dep to kserve

https://gerrit.wikimedia.org/r/734432

Change 743496 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] editquality: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/743496

Change 734432 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] articlequality: migrate kfserving dep to kserve

https://gerrit.wikimedia.org/r/734432

Change 743487 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] articlequality: upgrade transformer to kserve

https://gerrit.wikimedia.org/r/743487

Change 745281 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] articlequality: fix kserve imports

https://gerrit.wikimedia.org/r/745281

Change 743496 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] editquality: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/743496

Change 745281 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] articlequality: fix kserve imports

https://gerrit.wikimedia.org/r/745281

Change 745571 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] draftquality: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745571

Change 745583 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] topic: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745583

Change 745571 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] draftquality: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745571

Change 745583 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] topic: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745583

Change 745922 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] outlink: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745922

Change 745925 had a related patch set uploaded (by Accraze; author: Accraze):

[machinelearning/liftwing/inference-services@main] outlink: migrate transformer to kserve

https://gerrit.wikimedia.org/r/745925

Change 745922 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] outlink: migrate from kfserving to kserve

https://gerrit.wikimedia.org/r/745922

Change 745925 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] outlink: migrate transformer to kserve

https://gerrit.wikimedia.org/r/745925

Change 746898 had a related patch set uploaded (by Accraze; author: Accraze):

[operations/deployment-charts@master] ml-services: update revscoring-editquality image

https://gerrit.wikimedia.org/r/746898

Change 746901 had a related patch set uploaded (by Accraze; author: Accraze):

[operations/deployment-charts@master] ml-services: update revscoring-draftquality image

https://gerrit.wikimedia.org/r/746901

Change 746898 merged by Elukey:

[operations/deployment-charts@master] ml-services: update revscoring-editquality image

https://gerrit.wikimedia.org/r/746898

Change 746901 merged by Elukey:

[operations/deployment-charts@master] ml-services: update revscoring-draftquality image

https://gerrit.wikimedia.org/r/746901

Change 746949 had a related patch set uploaded (by Accraze; author: Accraze):

[operations/deployment-charts@master] ml-services: update revscoring-articlequality img

https://gerrit.wikimedia.org/r/746949

Change 746949 merged by Elukey:

[operations/deployment-charts@master] ml-services: update revscoring-articlequality img

https://gerrit.wikimedia.org/r/746949

ACraze claimed this task.

All inference-service images in the WMF registry have been upgraded to kserve v0.7.0, the ml-services helmcharts are updated and have been deployed to the ml-serve eqiad cluster.

Going to mark this as RESOLVED for now. Feel free to reopen if there are any tasks I missed.

Change 752612 had a related patch set uploaded (by Kevin Bazira; author: Kevin Bazira):

[machinelearning/liftwing/inference-services@main] Update docs to reflect KServe

https://gerrit.wikimedia.org/r/752612

Change 752612 merged by jenkins-bot:

[machinelearning/liftwing/inference-services@main] Update docs to reflect KServe

https://gerrit.wikimedia.org/r/752612