Repository navigation
fix(helm): remediate igloo-pg apps by uninstall instead of rollback - #1838
Draft
slipperypenguin wants to merge 1 commit into
Draft
slipperypenguin wants to merge 1 commit into
slipperypenguin wants to merge 1 commit into
Conversation
--- kubernetes/apps/default/litellm/app Kustomization: default/litellm HelmRelease: default/litellm
+++ kubernetes/apps/default/litellm/app Kustomization: default/litellm HelmRelease: default/litellm
@@ -18,12 +18,14 @@
timeout: 20m
interval: 1h
upgrade:
cleanupOnFail: true
remediation:
retries: 3
+ strategy: uninstall
+ timeout: 20m
values:
controllers:
litellm:
annotations:
reloader.stakater.com/auto: 'true'
containers:
--- kubernetes/apps/default/open-webui/app Kustomization: default/open-webui HelmRelease: default/open-webui
+++ kubernetes/apps/default/open-webui/app Kustomization: default/open-webui HelmRelease: default/open-webui
@@ -17,12 +17,14 @@
retries: -1
interval: 1h
upgrade:
cleanupOnFail: true
remediation:
retries: 3
+ strategy: uninstall
+ timeout: 15m
values:
controllers:
open-webui:
annotations:
reloader.stakater.com/auto: 'true'
containers: |
open-webui v0.13.3 (#1831) applied alembic migrations to igloo-pg, then exceeded the 5m Helm upgrade timeout. Flux remediated with a rollback to v0.12.4, which cannot start against the migrated schema ("Can't locate revision 9f6c2a8d1b43"), leaving the release stuck in a rollback loop. For the two apps that run forward-only migrations against igloo-pg (open-webui, litellm): - upgrade.remediation.strategy: uninstall, so a failed upgrade reinstalls the desired version rather than rolling back to an incompatible image - upgrade.timeout raised (15m open-webui, 20m litellm to match its install timeout) so first-boot migrations and model downloads fit
slipperypenguin
force-pushed
the
fix/igloo-pg-remediation
branch
from
November 12, 2026 21:29
89a7a07 to
8590413
Compare
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Today's Renovate merge of open-webui v0.12.4 → v0.13.3 (#1831) left the release stuck:
igloo-pg(head now9f6c2a8d1b43).alembic CommandError: Can't locate revision identified by '9f6c2a8d1b43'), so every rollback retry failed and the HelmRelease went into a loop. Recovery needed a manualset imageto v0.13.3 with the HelmRelease suspended.A rollback can never succeed once a forward-only migration has landed, so for apps that migrate
igloo-pgthe rollback remediation is actively harmful.What
Only two apps use
igloo-pg(n8n, manyfold and readeck are on SQLite):upgrade.timeoutupgrade.remediation.strategyuninstallinstall.timeout)uninstallDISABLE_SCHEMA_UPDATE: "False"With
uninstall, a failed upgrade is remediated by uninstalling and reinstalling the desired version (rolling forward), using the existinginstall.remediation.retries: -1. Nothing persistent is removed: open-webui's data PVC is anexistingClaim, and both apps keep their state in Postgres.Trade-off
An uninstall-based remediation briefly removes the Service and HTTPRoute, so there's a short outage during remediation. Today that window was already an outage (the rolled-back pod crashlooped), so this only makes recovery automatic.