Conversation
Signed-off-by: Maria Popova <mary_ppv@mail.ru>
mary-ppv
force-pushed
the
fix/ha-replicas-by-schedulable-master
branch
from
September 24, 2026 07:58
cb1ce71 to
937204d
Compare
mary-ppv
marked this pull request as ready for review
September 24, 2026 08:53
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The on-master HA helpers took spec.replicas straight from
global.discovery.clusterMasterCount, which counts control-plane nodes by
label and ignores cordon. These workloads place one replica per node, so
a cluster with two of five masters cordoned asked for five replicas with
three placeable nodes and never reached full readiness. The release
tracker waits for readyReplicas == spec.replicas, so the module install
timed out and blocked every task queued behind it.
Add helm_lib_master_replicas_count, which caps the master count by
global.discovery.clusterSchedulableMasterCount, and use it for both
spec.replicas and the maxUnavailable threshold in the two on-master
helpers. The threshold has to follow the same number: left on the raw
master count, five masters with three schedulable would render replicas:
3 alongside maxUnavailable: 2, leaving one live replica during a rollout.
Nothing changes on a cluster that does not publish the new value. It is
read through hasKey because an absent key and a legitimate zero both read
as 0 after
int, and treating the absent key as zero would collapsethose clusters to a single replica. The floor of 1 keeps a fully cordoned
control plane from rendering replicas: 0 and dropping the workload.