A Helm release is stuck in failed state
A chart upgrade that hit a template error or a timed out hook, leaving the HelmRelease unable to progress without manual intervention or a targeted annotation.
Releases roll forward cleanly
Running Flux CD?
Flux is quietly powerful and does not tell you when it has given up. A HelmRelease that paused mid upgrade, a Kustomization that drifted away from Git because somebody kubectl applied a hotfix, or a GitRepository that cannot pull a private repo all sit in a status field and age. We walk the reconciliation tree and get the cluster back in line with the repo.
What it is
Flux CD is a GitOps toolkit for Kubernetes that reconciles clusters against sources held in Git or OCI registries. The first version came out of the Weaveworks team in 2016, and Flux v2 with its toolkit style controllers reached general availability in 2022, now hosted at the CNCF as a graduated project.
The design splits responsibilities across several controllers: source-controller fetches from Git, OCI or Helm repositories, kustomize-controller applies Kustomizations, helm-controller reconciles HelmReleases, and notification-controller handles alerts and webhooks. That makes Flux composable but also means a failed reconciliation lives inside one of those controllers, and reading the status chain is a skill of its own.
The honest version
No technology is all upside. This is where Flux CD earns its place and where it tends to cause trouble.
Sound familiar?
The Flux CD problems that bring people to us, and what you have once each one is fixed.
A chart upgrade that hit a template error or a timed out hook, leaving the HelmRelease unable to progress without manual intervention or a targeted annotation.
Releases roll forward cleanly
A deploy key that was removed, a token that expired, or a secret that was renamed without updating the GitRepository resource.
Sources sync, status is green
A kubectl edit or an emergency hotfix that was applied directly to the cluster without being committed back to Git, leaving Flux to reconcile it away at the next pass.
Cluster and Git back in step
An update to a field Kubernetes does not allow to change on a live object, usually a selector or a StorageClass, so apply fails every cycle.
Changes land without conflict
A large monorepo or a Helm repo index that exceeds the controller's memory, causing reconciliation to loop on OOM kills.
Controller sized for the sources
A signing key that rotated without updating the public key config, or an image pushed without the signature step in CI.
Signed artifacts verify correctly
A resource whose finalizer is waiting on an external controller, blocking Kustomization pruning and stalling the next reconciliation.
Pruning completes on schedule
A regex filter on the ImagePolicy that no longer matches the tags being pushed, so automation looks healthy but never updates.
Tag policy matches real releases
What we do
Bring us one failing Kustomization, or hand us the GitOps platform for an end to end review.
How it works
Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.
We fix the cause, not the symptom. If something is down, getting it back up comes first.
Backups, monitoring, updates and access, so the same problem can't quietly come back.
You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.
Often found together
We work on the whole stack, so the problem can be wherever it actually is.
Flux's helm-controller relies on the Helm release state machine, and once a release lands in failed or pending-upgrade, Flux needs either a manual rollback through helm CLI or the right annotations on the HelmRelease to let it recover. The usual cause is a chart upgrade that hit a template error, a hook that timed out, or a webhook that rejected a resource. Fixing the template is step one, clearing the release state is step two.
Flux reconciles toward Git but does not block a human with cluster admin from running kubectl apply. The practical controls are to lock down admin access to break glass accounts, turn on Flux notifications so drift gets flagged the moment it happens, and treat any manual change as a commit that still has to land in the repo within the shift. Policy as code with Kyverno or OPA raises the floor further.
Both are mature and conformant GitOps tools. Flux fits teams that prefer Kubernetes native resources, OCI artifacts, image automation and minimal UI surface. Argo CD fits teams that want a strong dashboard, application sets and a GUI for developers. We pick based on what the team already uses, how many clusters the platform manages, and whether a UI is a hard requirement.
The source controller fetches and caches tarballs of Git repositories and Helm chart indexes. Large monorepos, repos with a lot of binary assets, or very long chart indexes push it past its default memory. Raising the controller resource limits is a short term fix; trimming the source paths, moving large assets out of the repo and limiting chart repo scope is the durable one.
Flux v1 and v2 are architecturally different, so the migration is not an upgrade in place. The pattern is to install Flux v2 alongside v1 on a copy cluster, port the sync spec into Kustomizations and HelmReleases, verify behavior, and only then cut over by removing v1. We do it one tenant or one cluster at a time.
Estimates come free. Describe the cluster count and where Flux is stalling, and you get a price before any resources are touched.
Free estimate
Share the controller logs or a sketch of the repo layout, and a GitOps engineer replies with a plan and a quote.
Your message is with a real person, and we'll get back to you soon.
Rather talk now? Call us: