Scenario-Based ArgoCD Interview Questions (2027)
26 scenario-based Argo CD interview questions with answers: OutOfSync drift, sync waves, ApplicationSets, RBAC, multi-cluster, secrets, rollback, tuning.
Quick answer: Scenario-based Argo CD interviews test whether you can debug and design real GitOps setups, not recite definitions. Expect questions on OutOfSync drift, sync waves and hooks, app-of-apps and ApplicationSets, RBAC and SSO, multi-cluster, Helm and Kustomize, secrets, rollback, custom health checks, Image Updater, sync windows and controller performance. The 26 scenarios below answer each one from the official docs.
Key takeaways
- Most OutOfSync puzzles come down to three causes: someone changed the live object, another controller owns a field (HPA replicas, webhook-injected values), or the rendered manifest differs from what you think Git contains. argocd app diff tells you which.
- Sync waves order resources inside one sync. Argo CD only moves to the next wave when the current one is healthy, so one stuck resource blocks everything after it.
- The Argo CD docs now recommend ApplicationSets (especially the cluster generator) for bootstrapping, with app-of-apps as the alternative. Anyone who can create Applications in the argocd namespace effectively has admin rights.
- argocd app rollback does not work while automated sync is enabled. In GitOps the normal rollback is a git revert.
- Argo CD's own docs strongly prefer secrets that are created on the destination cluster (Sealed Secrets, External Secrets Operator, Secrets Store CSI) over injecting them during manifest generation.
- Argo CD 3.0 changed several defaults, including annotation-based resource tracking and stricter RBAC for update and delete. Interviewers who run current versions may ask about them.
If you need the GitOps background first, read what GitOps is and how it improves infrastructure automation. For shorter definition-style questions, see our 70+ Argo CD interview questions. Answers below were checked against the Argo CD stable documentation on 6 November 2026.
How to answer an Argo CD scenario question
Interviewers usually score the order of your reasoning more than the final command. A pattern that works for most scenarios:
- State what you would look at first (application status, diff, events, controller or repo-server logs).
- Name the likely causes in order of how often they happen.
- Give the fix, and say whether it belongs in Git or in Argo CD configuration.
- Say how you would stop it happening again (policy, sync option, alert).
| Topic | Scenarios | Core Argo CD feature |
|---|---|---|
| Drift and OutOfSync | 1 to 4 | Diff, selfHeal, ignoreDifferences |
| Ordering | 5 to 7 | Sync waves, resource hooks |
| Scaling app definitions | 8 to 10 | App-of-apps, ApplicationSet generators |
| Access control | 11 to 12 | argocd-rbac-cm, AppProject, SSO |
| Multi-cluster | 13 to 14 | Cluster secrets, cluster generator |
| Helm and Kustomize | 15 to 16 | Value precedence, multiple sources |
| Secrets | 17 to 18 | Sealed Secrets, External Secrets, SOPS |
| Rollback and health | 19 to 21 | History, rollback, Lua health checks |
| Automation and change control | 22 to 23 | Image Updater, sync windows |
| Performance and upgrades | 24 to 26 | Controller sharding, repo-server tuning, 3.0 changes |
Drift and OutOfSync scenarios
1. An application shows OutOfSync right after a successful sync. What do you check?
Run argocd app diff <app> (or open the Diff tab) and look at which fields differ. Common causes are a field set by another controller or a mutating webhook, a field that the API server defaults (so Git and live never match), and a CRD whose schema strips or adds fields. If the difference is a field that should be owned by something else, add an ignoreDifferences rule. If the rendered output is wrong, check it with argocd app manifests <app> and fix Git. A hard refresh (argocd app get <app> --hard-refresh) rules out a stale manifest cache.
2. Someone ran kubectl edit on a production Deployment. How does Argo CD react, and how do you stop it?
Argo CD detects the drift and marks the app OutOfSync. Whether it reverts the change depends on the sync policy. With automated.selfHeal: true, the controller syncs again after the self-heal timeout (5 seconds by default, set by --self-heal-timeout-seconds on the application controller). Without selfHeal, the docs say live changes do not trigger automated sync, so the drift stays until someone syncs. To prevent it, give humans read-only cluster access and route all changes through Git.
spec:
syncPolicy:
automated:
prune: true
selfHeal: true
3. An HPA keeps changing replicas and the app flaps between Synced and OutOfSync. Fix it.
Remove replicas from the Deployment in Git if the HPA should own it, or tell Argo CD to ignore the field. Add RespectIgnoreDifferences=true so the ignored field is also left alone during sync, not only in the diff.
spec:
ignoreDifferences:
- group: apps
kind: Deployment
jsonPointers:
- /spec/replicas
syncPolicy:
syncOptions:
- RespectIgnoreDifferences=true
The docs also note that the HPA controller reorders spec.metrics, so keep that list in Git in the order the controller prefers.
4. Auto-sync with prune is on, and a refactor accidentally deleted a directory. What protects you?
Several things. Automated pruning refuses to sync an app down to zero resources unless allowEmpty: true is set. You can annotate critical objects (PVCs, namespaces) with argocd.argoproj.io/sync-options: Prune=false. PruneLast=true makes pruning the final step, after new resources are healthy. And branch protection with required review on the GitOps repo catches the mistake before Argo CD sees it.
Sync waves and hooks
5. A database migration must finish before the new app version starts. How do you set this up?
Run the migration as a Kubernetes Job with the PreSync hook annotation. PreSync hooks run before manifests are applied, and if one fails the sync stops, so the new Deployment never rolls out against an unmigrated schema.
apiVersion: batch/v1
kind: Job
metadata:
name: db-migrate
annotations:
argocd.argoproj.io/hook: PreSync
argocd.argoproj.io/hook-delete-policy: BeforeHookCreation
spec:
backoffLimit: 1
template:
spec:
restartPolicy: Never
containers:
- name: migrate
image: registry.example.com/shop/api:1.4.0
command: ["./migrate", "up"]
If no delete policy is given, Argo CD assumes BeforeHookCreation. Other policies are HookSucceeded and HookFailed. Add a PostSync hook for smoke tests and a SyncFail hook for cleanup or alerts.
6. Wave 1 never starts. Wave 0 contains a ConfigMap, a Deployment and a custom resource. Why?
Argo CD applies the first wave that has an out-of-sync or unhealthy resource and only moves on when that wave is healthy. If the custom resource in wave 0 has no health check, or reports Progressing forever, later waves wait. Check the resource tree for the item that is not Healthy, then either fix it or add a custom health check (scenario 21). Remember the full order: phase, then wave (lower first, negatives allowed, default 0), then kind (namespaces first, custom resources last), then name. Argo CD also waits 2 seconds between waves by default (ARGOCD_SYNC_WAVE_DELAY).
7. The team uses a Helm chart with Helm hooks. Do they still work under Argo CD?
Mostly. Argo CD renders charts with helm template and maps hooks: pre-install and pre-upgrade become PreSync, post-install and post-upgrade become PostSync, pre-delete and post-delete map to PreDelete and PostDelete, and hook-weight becomes the sync wave. Rollback and test hooks are not supported. Two traps: Argo CD cannot tell install from upgrade, so both pre-hooks can run together, and if the chart defines any Argo CD hooks, all Helm hooks are ignored. Since there is no Helm release, helm ls shows nothing.
App-of-apps and ApplicationSets
8. You need to bootstrap 30 platform add-ons on a new cluster from one command. Which pattern do you use?
Either an app-of-apps parent Application that points at a folder of child Application manifests, or an ApplicationSet. The current cluster bootstrapping guide presents ApplicationSets (often with the cluster generator) as the main recommendation and app-of-apps as an alternative. App-of-apps is still useful when each child needs hand-tuned settings. Use sync waves on the child Applications to install CRDs and operators before the things that depend on them, and add the resources-finalizer.argocd.argoproj.io finalizer to children so deleting them also deletes their resources.
9. In an app-of-apps setup, the parent shows Healthy even though a child app is Degraded. Why?
Since Argo CD 1.8, the health of an Application resource is not assessed by default, so the parent does not inherit child health. If you rely on waves between child apps, restore it with a custom health check under resource.customizations.health.argoproj.io_Application in argocd-cm that copies the child's status.health. The health docs include the Lua for this.
10. Which ApplicationSet generator fits each case?
| Requirement | Generator |
|---|---|
| One app per folder under apps/ in a repo | Git (directory) |
| One app per config.json file describing an environment | Git (files) |
| Deploy the same add-on to every registered cluster labelled env=prod | Cluster |
| Preview environment for every open pull request | Pull Request |
| Every repo in a GitHub or GitLab organisation | SCM Provider |
| Every app folder on every prod cluster | Matrix (Git x Cluster) |
| Cluster list plus per-cluster overrides | Merge |
| Fixed, small list of targets | List |
| Parameters from an external service | Plugin |
The docs also list Cluster Decision Resource and an OCI generator. A good follow-up point: ApplicationSets can create Applications in any project they template, so restrict who can write ApplicationSet manifests, and use Progressive Syncs if you need staged rollout across the generated apps.
RBAC and SSO
11. Developers in the payments team should sync only their own apps in the payments project, and never delete. Write the policy.
Put the policy in argocd-rbac-cm and bind the SSO group to a role. Objects for application resources use <project>/<app>.
apiVersion: v1
kind: ConfigMap
metadata:
name: argocd-rbac-cm
namespace: argocd
data:
policy.default: role:readonly
scopes: '[groups]'
policy.csv: |
p, role:payments-dev, applications, get, payments/*, allow
p, role:payments-dev, applications, sync, payments/*, allow
p, role:payments-dev, logs, get, payments/*, allow
p, role:payments-dev, applications, delete, payments/*, deny
g, my-org:payments-team, role:payments-dev
Test it before rollout with argocd admin settings rbac can role:payments-dev delete applications 'payments/api' --policy-file policy.csv. Pair it with an AppProject that limits the payments project to its own source repos, destination namespaces and resource kinds. Note that permissions given by policy.default cannot be removed with a deny rule.
12. After an upgrade to Argo CD 3.x, a team can update an Application but can no longer delete a Pod from the UI. What changed?
In 3.0, update and delete on an application apply only to the Application object, not its managed resources. Grant the fine-grained form, for example p, role:team, applications, delete/*/Pod/*/*, payments/*, allow, or set server.rbac.disableApplicationFineGrainedRBACInheritance to false to get v2 behaviour back. Also in 3.0, viewing logs needs an explicit logs, get permission unless the default role is readonly or admin, and Dex-based policies match on the federated_claims.user_id claim instead of sub.
Multi-cluster
13. How do you register a new EKS cluster with a central Argo CD, without using the CLI?
Create a Secret in the Argo CD namespace with the label argocd.argoproj.io/secret-type: cluster and the name, server and config fields. For EKS, config can use awsAuthConfig with an IAM role instead of a static token. The imperative alternative is argocd cluster add <kube-context>, which creates a service account in the target cluster for Argo CD to use.
apiVersion: v1
kind: Secret
metadata:
name: prod-mumbai
namespace: argocd
labels:
argocd.argoproj.io/secret-type: cluster
env: prod
type: Opaque
stringData:
name: prod-mumbai
server: https://ABC123.gr7.ap-south-1.eks.amazonaws.com
config: |
{
"awsAuthConfig": {
"clusterName": "prod-mumbai",
"roleARN": "arn:aws:iam::111122223333:role/argocd-deployer"
},
"tlsClientConfig": {
"caData": "BASE64_CA_CERT"
}
}
The env: prod label is what a cluster generator selects on in scenario 14.
14. Hub-and-spoke or one Argo CD per cluster?
A central (hub) instance gives one UI, one RBAC model and easy fleet-wide ApplicationSets, but it needs network access and credentials to every cluster and becomes a single point of failure and a high-value target. One instance per cluster isolates failures and credentials but multiplies upgrades and configuration. Many teams use one instance per environment tier. Whatever you pick, plan controller sharding (scenario 24) once a hub manages many clusters. Our post on Kubernetes multi-cluster management solutions compares other tools in this space.
Helm and Kustomize
15. Your Helm values for each environment live in a separate config repo. How do you wire that up?
Use multiple sources (available since v2.6). One source is the chart, the other is the values repo with a ref name, referenced as $ref in valueFiles. Precedence, highest first, is parameters, valuesObject, values, valueFiles (last file wins), then the chart's own values.yaml.
spec:
sources:
- repoURL: https://charts.example.com
chart: shop-api
targetRevision: 2.3.1
helm:
valueFiles:
- $values/envs/prod/values.yaml
- repoURL: https://github.com/example/shop-config.git
targetRevision: main
ref: values
Set ignoreMissingValueFiles: true if some environments do not have an override file.
16. The same base must deploy to dev, stage and prod with different replicas and image tags. Helm or Kustomize?
Either works in Argo CD. Kustomize suits plain YAML you own: one base/ and an overlay per environment, each overlay a separate Application path (or one ApplicationSet with the Git directory generator). Helm suits packaged software with many options. A sound interview answer mentions that Argo CD runs the tool itself, so the version of Helm or Kustomize bundled with the repo-server matters, and that you can preview rendered output with argocd app manifests. Our Helm chart examples show common chart layouts.
Secrets
17. The security team says no plain Secrets in Git. Which approach do you choose and why?
The Argo CD docs strongly recommend tools that create the secret on the destination cluster, so Argo CD never sees the value:
| Tool | What lives in Git | Good fit |
|---|---|---|
| Sealed Secrets | A SealedSecret encrypted with the controller's public key | Small teams, no external secret store |
| External Secrets Operator | An ExternalSecret that points to Vault, AWS Secrets Manager, Azure Key Vault or GCP Secret Manager | Teams with a central secret store and rotation |
| Secrets Store CSI Driver | A SecretProviderClass; values mount as files | Apps that can read secrets from files |
| SOPS (via KSOPS or helm-secrets) | Encrypted YAML, decrypted during manifest generation | Teams already using SOPS with KMS or age |
SOPS and argocd-vault-plugin decrypt during manifest generation, which the docs discourage: Argo CD needs access to the secrets, generated manifests (including secret values) are cached in Redis, and secret updates get tied to syncs. If you must use them, restrict network access to Redis and the repo-server. More options are in our post on secret management in GitOps workflows.
18. A rotated database password in Vault is not picked up by pods. Argo CD shows Synced. Why?
With External Secrets Operator, Argo CD only manages the ExternalSecret object, which did not change, so the app stays Synced. ESO refreshes the Kubernetes Secret on its refreshInterval. Pods that read the secret as environment variables keep the old value until they restart. Fix it by lowering the refresh interval where needed and adding a reloader or a checksum annotation that triggers a rollout when the Secret changes. This is the separation the Argo CD docs describe: secret updates happen independently of syncs.
Rollback and health checks
19. A release broke checkout. How do you roll back with Argo CD?
In a GitOps setup, revert the commit (git revert) and let Argo CD sync, so Git stays the source of truth. For an emergency, argocd app history <app> lists deployed revisions and argocd app rollback <app> <history-id> redeploys one. Without an ID it goes to the previous version. The docs state that rollback cannot be performed against an application with automated sync enabled, so you must disable auto-sync first, and remember to fix Git before re-enabling it or the next sync will reapply the broken version.
argocd app history checkout
argocd app set checkout --sync-policy none
argocd app rollback checkout 41
# after the fix is merged in Git
argocd app set checkout --sync-policy automated --self-heal
For automatic rollback based on metrics, pair Argo CD with Argo Rollouts. See our guide to deployment rollback strategies.
20. A sync succeeds but the app is Degraded. Where do you look?
Open the resource tree and find the resource that is not Healthy. For a Deployment, Degraded usually means it exceeded its progress deadline: check pod events for image pull errors, failing readiness probes, or crash loops (kubectl describe pod, kubectl logs --previous). Sync status and health status are separate in Argo CD. Synced only means the live state matches Git.
21. A cert-manager Certificate shows Healthy before the certificate is actually issued. How do you fix that?
Add a custom health check in Lua under resource.customizations.health.<group>_<kind> in argocd-cm. The script reads the resource from obj and returns a status (Healthy, Progressing, Degraded or Suspended) and an optional message.
data:
resource.customizations.health.cert-manager.io_Certificate: |
hs = {}
if obj.status ~= nil and obj.status.conditions ~= nil then
for i, condition in ipairs(obj.status.conditions) do
if condition.type == "Ready" and condition.status == "False" then
hs.status = "Degraded"
hs.message = condition.message
return hs
end
if condition.type == "Ready" and condition.status == "True" then
hs.status = "Healthy"
hs.message = condition.message
return hs
end
end
end
hs.status = "Progressing"
hs.message = "Waiting for certificate"
return hs
Argo CD already ships health checks for many popular CRDs, so check the bundled ones before writing your own.
Image Updater and sync windows
22. CI pushes a new image tag. How can Argo CD pick it up without a human editing Git?
Two common answers. First, CI commits the new tag to the GitOps repo (for example with kustomize edit set image or yq), and Argo CD syncs it. Second, Argo CD Image Updater watches the registry and updates the app. Current Image Updater releases (v1.x) are configured with an ImageUpdater custom resource, while annotations belong to the older v0.x line. Update strategies include semver, newest-build, digest, alphabetical and calver. Prefer the git write-back method for production: the default argocd method changes the Application directly, and the docs warn that those changes are lost if the Application is recreated or synced from Git. Image Updater supports Kustomize and Helm apps (and plugin apps in some setups), not plain YAML directories.
23. Production deploys must not happen during business hours, except for approved hotfixes. How?
Define sync windows on the AppProject. A deny window blocks syncs while active, and deny wins when allow and deny overlap. Set manualSync: true so an operator can still push a hotfix by hand.
apiVersion: argoproj.io/v1alpha1
kind: AppProject
metadata:
name: payments
namespace: argocd
spec:
syncWindows:
- kind: deny
schedule: "0 9 * * 1-5"
duration: 10h
applications:
- "*"
manualSync: true
timeZone: Asia/Kolkata
The same window can be added with argocd proj windows add payments --kind deny --schedule "0 9 * * 1-5" --duration 10h --applications "*" --manual-sync --time-zone Asia/Kolkata. The default time zone is UTC, which is a frequent mistake for teams in India.
Performance and upgrades
24. Argo CD manages 1,500 applications across 20 clusters and reconciliation is slow. What do you tune?
- Raise --status-processors (default 20) and --operation-processors (default 10) on the application controller. The docs use 50 and 25 as an example for about 1,000 apps.
- Shard the controller: increase StatefulSet replicas, set ARGOCD_CONTROLLER_REPLICAS to match, and choose a sharding algorithm (legacy, or the experimental round-robin and consistent-hashing) with controller.sharding.algorithm in argocd-cmd-params-cm.
- Scale the repo-server and cap concurrent manifest generation with --parallelismlimit to avoid OOM kills. Raise ARGOCD_EXEC_TIMEOUT (default 90s) if large Helm charts time out.
- Configure Git webhooks so changes are picked up without waiting for the polling interval (timeout.reconciliation, default 3 minutes).
- Watch argocd_app_reconcile and argocd_app_k8s_request_total to confirm the change helped.
25. Every commit to a monorepo refreshes all 200 apps. How do you limit that?
Add the argocd.argoproj.io/manifest-generate-paths annotation to each Application, listing the paths it depends on (semicolon separated, relative paths and globs allowed). Commits that do not touch those paths no longer invalidate the app's manifest cache. Since v2.11 this works without webhooks too. Use fully qualified refs such as refs/heads/main in targetRevision.
26. You are upgrading from 2.14 to 3.x. What do you check first?
Read the 2.14 to 3.0 upgrade notes. The items that most often bite are: resource tracking now defaults to annotations (label users may see orphaned resources with ApplyOutOfSyncOnly), fine-grained RBAC for update and delete, logs RBAC enforced, repositories defined in argocd-cm no longer supported (move them to Secrets), new default resource.exclusions, resource health no longer stored in the Application status, and a Helm upgrade where null values now override defaults. Test in a staging instance with a copy of your config first.
Practice with a trainer
These scenarios come up in our hands-on GitOps labs. You can practise them in the DevOps training in Pune (Kharadi classroom or live online across India), and the CKA training covers the Kubernetes troubleshooting that most Argo CD problems turn into. For a broader look at GitOps practice, read 18 best practices for GitOps deployment.
FAQ
What is the difference between Synced and Healthy in Argo CD?
Sync status compares the live cluster with the desired state from Git: Synced means they match. Health status reports whether resources are working: Healthy, Progressing, Degraded, Suspended, Missing or Unknown. An app can be Synced and Degraded at the same time, for example when the manifest is applied but pods crash.
What does selfHeal do in Argo CD?
With automated.selfHeal: true, Argo CD syncs again when the live state drifts from Git, after a self-heal timeout of 5 seconds by default. Without it, automated sync runs only when Git changes, and manual edits to the cluster remain until the next sync.
Should I use app-of-apps or ApplicationSets?
Use ApplicationSets when apps follow a pattern, such as one per cluster, folder or pull request. The current docs present them as the main option for cluster bootstrapping. Use app-of-apps when each child needs individual settings. Many setups combine both, and both should be writable only by admins.
How do sync waves work?
Each resource gets an integer in the argocd.argoproj.io/sync-wave annotation (default 0, negatives allowed). Argo CD applies lower waves first and waits until a wave is healthy before moving on, with a 2 second delay between waves by default.
Why can't I roll back an Argo CD application?
Rollback is blocked while automated sync is enabled. Disable auto-sync, run argocd app rollback <app> <id>, then fix Git before turning auto-sync back on. In most teams, reverting the Git commit is the preferred rollback.
How should I manage secrets with Argo CD?
Argo CD's docs recommend creating secrets on the destination cluster with Sealed Secrets, External Secrets Operator or the Secrets Store CSI Driver. Plugins that inject secrets during manifest generation work but expose values to Argo CD and its Redis cache.
Does Argo CD use helm install?
No. Argo CD renders charts with helm template and applies the output itself, so helm ls does not show the release. Most Helm hooks are mapped to Argo CD hooks, but rollback and test hooks are ignored.
References
- Automated sync policy
- Sync phases and waves
- Sync options
- Diffing customization
- Cluster bootstrapping (app of apps)
- ApplicationSet generators
- RBAC configuration
- Declarative setup (clusters and repositories)
- Helm
- Multiple sources
- Secret management
- argocd app rollback
- Resource health
- Sync windows
- High availability and performance tuning
- Upgrading from 2.14 to 3.0
- Argo CD Image Updater
- Image Updater write-back methods
What's Your Reaction?
Like
0
Dislike
0
Love
0
Funny
0
Angry
0
Sad
0
Wow
0