fix: Add retry logic to kube-manifests push to avoid race conditions

[JesperTerkelsen]

Summary

Add retry logic with automatic rebase to component-deploy workflow to prevent race conditions when multiple environments deploy concurrently.

Problem

When dev and staging deployments run in parallel (or any concurrent deployments), they can encounter race conditions when pushing to kube-manifests:

1. Both jobs check out kube-manifests at roughly the same time
2. Both make changes and commit locally
3. Both try to push - one succeeds, one fails with "remote has moved ahead"

Example failure: https://github.com/monta-app/service-admin-panel/actions/runs/21146966256/job/60814766132

Solution

Replace the ad-m/github-push-action with custom retry logic that:

• Attempts to push up to 5 times
• On push failure, waits 5 seconds
• Fetches latest changes and rebases automatically
• Retries the push
• Exits with error only if rebase fails (conflicts) or max retries reached

Changes

.github/workflows/component-deploy.yml:

• Replaced ad-m/github-push-action@master step with custom bash script
• Added retry loop with configurable max attempts (5) and delay (5s)
• Automatic fetch and rebase on push failure
• Clear logging for each attempt and outcome

Benefits

• Eliminates race conditions in concurrent deployments
• No manual intervention needed when pushes collide
• Maintains deployment reliability across all environments
• Clear logging shows retry attempts and outcomes

Testing

This should be tested with concurrent deployments (dev + staging) to verify the retry logic works as expected.

🤖 Generated with Claude Code