fix(valkey): seed on the current master, not a reachable replica

[mreho]

Summary

Problem

The Sentinel bootstrap seeds on the node currently reporting role:master, but the wait loop only ran that discovery when the static seed (node-0) was unreachable. After a failover, node-0 can be a live replica, so a reachability-only check can seed Sentinel from a read-only node and advertise the wrong master.

Fix

Discover the node reporting role:master among the data-node peers first, and fall back to the static seed only if none answers within the bounded budget.

Type Of Change

• New chart
• Existing chart change
• Documentation only
• CI / repository workflow

PR Governance

• I linked an existing issue in this PR body (Resolves #NNN or Related to #NNN)
• If this is a new chart PR, labels enhancement and type:feature are applied

Checklist

• I created this branch from updated main
• My PR targets main
• My commit message and PR title follow Conventional Commits
• I did not edit version in Chart.yaml manually
• I updated chart docs for behavior or default changes

Site Sync (GR-007)

• Site PR: docs(valkey): sync failover hardening options site#365

Local Validation

• make validate-chart CHART=valkey TIMEOUT=1200 passed end-to-end: FULLY VALIDATED (21 layers), including k3d behavioral scenarios for default, cluster, dual-stack, existing-secret, external-secrets, metrics, networkpolicy, replication, sentinel, standalone, tls-cluster, and tls-replication.
• make site-sync-check CHART=valkey passed on site#365.
• make release-check REPO=charts passed with the expected GR-077 post-merge release confirmation warning.
• make attribution-check REPO=charts passed.

Remote Validation

• GitHub Actions are green; standards-check is skipped by fork policy and covered locally.
• CodeRabbit completed with no actionable comments.
• Review threads: 0 unresolved.

Issue: #633

[coderabbitai]

📝 Walkthrough

Walkthrough

The Sentinel entrypoint's initial bootstrap logic in the Valkey Helm chart's configmap is modified to discover the current master by repeatedly calling find_current_master instead of pinging the seed node, falling back to SEED_MASTER after bounded attempts. Corresponding Helm test assertions are updated to match the new script content.

Changes

Sentinel Bootstrap Discovery Update

Layer / File(s)	Summary
Discovery loop rewrite charts/valkey/templates/configmap.yaml	Replaces ping-based seed reachability loop with repeated find_current_master calls; falls back to SEED_MASTER and updates logging after 6 unsuccessful iterations.
Test assertions updated charts/valkey/tests/cluster_domain_test.yaml, charts/valkey/tests/ha_extension_test.yaml	Adds/updates matchRegex assertions to validate the new find_current_master-based master assignment logic in the entrypoint script.

Estimated code review effort: 2 (Simple) | ~10 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly describes the main change: seeding from the current master instead of a reachable replica.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}