Page MenuHomePhabricator

Liftwing studio chart
Closed, ResolvedPublic

Description

Build and deploy a working instance of liftwing studio on dse eqiad cluster.
This involves the deployment of a pgsql instance (done) and the creation of a dedicated liftwing studio chart.
This task is to track the progress on the chart work.

Liftwing studio is a web frontend to the LLM models hosted by the ML team on ml-serve k8s clusters.

Details

Related Changes in Gerrit:
SubjectAuthorRepoBranchLines +/-
Dpogorzelskioperations/deployment-chartsmaster+905 -0
Dpogorzelskioperations/deployment-chartsmaster+42 -0
Dpogorzelskioperations/deployment-chartsmaster+33 -2
Dpogorzelskioperations/deployment-chartsmaster+59 -13
Dpogorzelskioperations/deployment-chartsmaster+152 -107
Dpogorzelskioperations/deployment-chartsmaster+7 -2
Dpogorzelskioperations/deployment-chartsmaster+33 -2
Dpogorzelskioperations/deployment-chartsmaster+947 -0
Dpogorzelskioperations/puppetproduction+7 -0
Dpogorzelskioperations/deployment-chartsmaster+37 -8
Dpogorzelskioperations/deployment-chartsmaster+72 -1
Dpogorzelskioperations/deployment-chartsmaster+37 -8
Dpogorzelskioperations/deployment-chartsmaster+210 -0
Dpogorzelskioperations/dnsmaster+2 -0
Dpogorzelskioperations/deployment-chartsmaster+2 K -0
Dpogorzelskioperations/puppetproduction+2 -0
Dpogorzelskioperations/puppetproduction+13 -0
Show related patches Customize query in gerrit

Related Objects

StatusSubtypeAssignedTask
OpenDPogorzelski-WMF
ResolvedDPogorzelski-WMF

Event Timeline

DPogorzelski-WMF changed the task status from Open to In Progress.Sep 11 2026, 1:11 PM

Change #1339756 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: chart for Open WebUI and LiteLLM

https://gerrit.wikimedia.org/r/1339756

There is a candidate chart that i will test out of band before submitting. The pgsql isntance is up and running and secrets in the private repo are set up and ready.

The chart is working i will be investigating if/how it connects to the llm models on ML cluster.
I will also add 2 small changes to expose the service on a domain and make sure it's routed correctly.
After that i will investigate central authentication.
The service will not be exposed to the outside world until this is done.

Change #1341128 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/dns@master] dns: add liftwing-studio records

https://gerrit.wikimedia.org/r/1341128

Change #1341130 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/puppet@production] liftwing-studio: add to service catalog

https://gerrit.wikimedia.org/r/1341130

Change #1341163 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/puppet@production] services_proxy: send SNI on the inference listener

https://gerrit.wikimedia.org/r/1341163

Change #1341174 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: chart deployment

https://gerrit.wikimedia.org/r/1341174

Change #1341130 merged by Dpogorzelski:

[operations/puppet@production] liftwing-studio: add to service catalog

https://gerrit.wikimedia.org/r/1341130

Change #1341163 merged by Dpogorzelski:

[operations/puppet@production] services_proxy: send SNI on the inference listener

https://gerrit.wikimedia.org/r/1341163

Change #1339756 merged by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: chart for Open WebUI and LiteLLM

https://gerrit.wikimedia.org/r/1339756

Change #1341128 merged by Dpogorzelski:

[operations/dns@master] dns: add liftwing-studio records

https://gerrit.wikimedia.org/r/1341128

Change #1341174 merged by jenkins-bot:

[operations/deployment-charts@master] liftwing-studio: chart deployment

https://gerrit.wikimedia.org/r/1341174

Change #1341903 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: add external-services egress to the chart

https://gerrit.wikimedia.org/r/1341903

Change #1341904 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: OIDC auth against idp.wikimedia.org

https://gerrit.wikimedia.org/r/1341904

Change #1341909 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: OIDC auth against idp.wikimedia.org

https://gerrit.wikimedia.org/r/1341909

Change #1341909 abandoned by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: OIDC auth against idp.wikimedia.org

Reason:

Duplicate: created accidentally when an amend lost the Change-Id. Superseded by patchset 2 of change 1341904.

https://gerrit.wikimedia.org/r/1341909

Change #1342135 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/puppet@production] idp: add the liftwing_studio service

https://gerrit.wikimedia.org/r/1342135

Change #1341903 merged by jenkins-bot:

[operations/deployment-charts@master] liftwing-studio: add external-services egress to the chart

https://gerrit.wikimedia.org/r/1341903

Change #1341904 merged by jenkins-bot:

[operations/deployment-charts@master] liftwing-studio: OIDC auth against idp.wikimedia.org

https://gerrit.wikimedia.org/r/1341904

Change #1342135 merged by Dpogorzelski:

[operations/puppet@production] idp: add the liftwing_studio service

https://gerrit.wikimedia.org/r/1342135

SSO login works and it's the only form of authentication allowed now.
Only wmf and nda groups are allowed. All new users are regular users without admin permissions.

I will now switch the setup to contact inference models via sidecar proxy.

Change #1342606 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] mesh: Accept websocket upgrades in configuration 1.15.4

https://gerrit.wikimedia.org/r/1342606

Change #1342607 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: accept websocket upgrades at the tls terminator

https://gerrit.wikimedia.org/r/1342607

Change #1342611 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] Copy mesh.configuration 1.15.3 -> 1.15.4

https://gerrit.wikimedia.org/r/1342611

Change #1342612 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] mesh.configuration: Accept websocket upgrades

https://gerrit.wikimedia.org/r/1342612

Change #1342613 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: accept websocket upgrades at the tls terminator

https://gerrit.wikimedia.org/r/1342613

Change #1342606 abandoned by Dpogorzelski:

[operations/deployment-charts@master] mesh: Accept websocket upgrades in configuration 1.15.4

Reason:

wrong structure

https://gerrit.wikimedia.org/r/1342606

Change #1342607 abandoned by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: accept websocket upgrades at the tls terminator

Reason:

wrong structure

https://gerrit.wikimedia.org/r/1342607

There is a proposed patch to fix websocket connections, waiting for approval as there is another similar solution pending.
I will also start looking into swapping openweb ui with libre chat as this is what the ML team is leaning towards as the final choice.

Change #1345056 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] ferretdb: make the Service name configurable

https://gerrit.wikimedia.org/r/1345056

Change #1345057 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: back the database with ferretdb

https://gerrit.wikimedia.org/r/1345057

Change #1345058 had a related patch set uploaded (by Dpogorzelski; author: Dpogorzelski):

[operations/deployment-charts@master] liftwing-studio: replace Open WebUI with LibreChat

https://gerrit.wikimedia.org/r/1345058

Change #1345056 merged by Dpogorzelski:

[operations/deployment-charts@master] ferretdb: make the Service name configurable

https://gerrit.wikimedia.org/r/1345056

Change #1345057 merged by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: back the database with ferretdb

https://gerrit.wikimedia.org/r/1345057

Change #1345058 merged by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: replace Open WebUI with LibreChat

https://gerrit.wikimedia.org/r/1345058

I swapped out openwebui with libre chat, currently can be tested here: https://liftwing-studio.discovery.wmnet:30443
I will close this task now and open a new one to review and publish this service under a public URL.

Change #1342613 abandoned by Dpogorzelski:

[operations/deployment-charts@master] liftwing-studio: accept websocket upgrades at the tls terminator

Reason:

we switched to librechat so this is not needed anymore

https://gerrit.wikimedia.org/r/1342613

Change #1342612 abandoned by Dpogorzelski:

[operations/deployment-charts@master] mesh.configuration: Accept websocket upgrades

Reason:

we switched to librechat so this is not needed anymore

https://gerrit.wikimedia.org/r/1342612

Change #1342611 abandoned by Dpogorzelski:

[operations/deployment-charts@master] mesh.configuration: Copy 1.15.3 -> 1.15.4

Reason:

we switched to librechat so this is not needed anymore

https://gerrit.wikimedia.org/r/1342611