Our Argo CD setup has automatic sync disabled outside development and for infrastructure components. Updates wait for someone to sync them, so OutOfSync is a normal state. Does that undermine GitOps, and what are the alternatives?
Is manually syncing everything outside dev a problem for GitOps?
Answers
The practical problem is that OutOfSync now covers two very different things: an approved change waiting to deploy and an unexpected difference in the cluster. You have to investigate each one to know which it is.
A common alternative is to make the production configuration in Git represent what should be running now. Promote a tested version with a pull request, require the appropriate review, and let Argo CD sync after the merge. That puts the approval before the desired state changes rather than between the desired state and the deployment.
Reply to this answer
Manual sync can be a deliberate release choice. A team might want developers to choose when a tested change goes out, with staging kept close to production and synced automatically. That can be workable, even if it gives up continuous reconciliation in production.
The tradeoff should be explicit. Git may be ahead of production, so people need a reliable way to see what is actually deployed and what is waiting. Otherwise a sync can feel like routine repair when it is really a release.
Reply to this answer
If merging to main feels too close to deploying, point production at a separate release branch or production configuration. Promote a version there only when it is ready, then keep automatic sync enabled for that target.
That still gives you a human gate, but the gate is a reviewable Git change. Auto-heal is a separate setting to consider if you also want Argo CD to correct live changes that diverge from the desired state.
8h!ferrulecaddy_715:
That sounds like it would let us hold a release without making every pending change look like production drift.
Reply to this answer
Before changing the setting, ask what the manual sync is supposed to protect against. If the concern is untested changes, improve lower environment testing and approval. If it is an unauthorized or compromised change, an extra human gate may be part of the security model, but it needs clear permissions and a meaningful review rather than a reflexive click.
Also agree on how emergency changes are handled. If automatic sync gets paused during an incident, someone needs to check and restore the intended setting afterward.
Reply to this answer
Add your answer
Answers are reviewed before publication.
Trelvori-8b:
A diff check in the pull request can also help reviewers see what the change would apply before they approve it.