[server] Fix prune index handling #1928

[michaelvlach]

No description provided.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
agdb	Ready	Preview	Sep 14, 2026 2:01pm UTC

[Copilot]

🔵 Needs a closer look

The change alters distributed consensus resync/pruning recovery behavior — a safety-critical path — and addresses the symptom rather than the root-cause prune_index calculation named in the issue, so it warrants human sign-off.

Pull request overview

This PR fixes a cluster-resync bug in the raft consensus layer (issue #1928). When a leader's prune_index is stale after a restart (e.g., reinit() recomputes it as commit - max_log_entries, which can be lower than the actual amount of pruning that occurred in a previous session), a follower that needs already-pruned log entries would fall into a gap that the prune safety-net (prune_index > log_commit) does not catch. Previously reconcile() detected this gap, logged a warning, and returned None without taking any corrective action, leaving the follower stuck. The fix makes reconcile() set the target node's force_resync flag when a gap is detected, which is then propagated via the next heartbeat so the follower triggers a full resync — matching the existing pattern used for the append-failure threshold path.

Changes:

• In reconcile(), set force_resync = true on the target node when a log gap is detected, and update the warning message to reflect the new "Forcing resync." behavior.
• Add a TestCluster::with_max_log_entries constructor (with new delegating to it) and an expect_needs_resync helper.
• Add unit and integration tests covering the gap-detection resync path.

File summaries

File	Description
agdb_server/src/raft.rs	reconcile() now flags force_resync on gap detection; adds test helpers and unit/integration tests validating the resync-on-gap behavior.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.