Repair Redis Sentinel Promotion Quorum Drift

Sentinel clusters often fail silently during network partitions, leading to split-brain scenarios where nodes disagree on the current master. This skill diagnoses unstable primary promotions and quorum drift by auditing state snapshots and epoch transitions. It prevents accidental data loss by mapping write-exposure windows before you attempt a recovery.

What it does

• Sentinel Vote Analysis identifies discrepancies in node voting records and configuration epochs.
• Epoch Transition Tracing maps the sequence of master promotions to find where the cluster diverged.
• Write-Risk Assessment calculates the window where split-brain writes may have occurred.
• Recovery Plan Generation builds a quorum-safe rollback and synchronization strategy based on actual node offsets.

How it works

1. Log Ingestion provides the agent with Sentinel logs, node configuration snapshots, and transaction timelines.
2. Evidence Mapping runs a four-stage check to compare expected vs. observed cluster states.
3. Risk Quantification identifies the specific epoch where the quorum drifted and flags potential data loss windows.
4. Action Proposal generates a non-destructive recovery plan with specific rollback criteria.

Frameworks & tools

This skill is designed for Redis Sentinel deployments. It requires access to sentinel.conf files, INFO command outputs, and system logs from the affected nodes.

Why this beats prompting it yourself

Generic prompts often miss the nuances of epoch synchronization and fail to account for the specific sequence of network recovery. This skill enforces a strict verification process, labeling every finding as VERIFIED or INFERRED to prevent dangerous recovery assumptions.

Use cases

• Resolve a split-brain scenario where two nodes both claim to be the primary.
• Audit a failed promotion after a major network partition or cloud provider outage.
• Map data loss risks before manually resetting Sentinel configurations.

Known limitations

Requires manual provision of log files and state snapshots as it does not have live execution capabilities. It proposes recovery steps but does not execute them directly.