Midnight Indexer 4.3.2
https://github.com/midnightntwrk/midnight-indexer/releases/tag/v4.3.2
Midnight Indexer 4.3.2
Midnight Indexer 4.3.2 Release Notes
Release date: 2026-05-15
Release type: Bundle component — Midnight 1.1 (patch release on the development line)
Git tag: v4.3.2
Tree hash: 9b2de1658b375e22323e6420eef372c0bf924aec
Environment: All public networks (mainnet, preprod, preview, devnet, qanet)
Note: This release pairs with node 1.0.0 and ships as part of the Midnight 1.1 bundle. Operators on node 0.22.x should track the 4.0.x maintenance line, not this release.
The 4.3.x line pairs with node 1.0.0 / ledger 8.1 and carries accumulated schema and fee-computation changes since the 4.0.x maintenance line (node 0.22.x). Index state built on 4.0.x cannot be reused — reset the indexer and re-index from genesis on upgrade. Budget ~3–4 days at mainnet/preprod scale. Upgrades within the 4.3.x line (4.3.0 → 4.3.1 → 4.3.2) are drop-in and require no reset.
Summary
Small patch release on the development line. Adds a /live HTTP liveness endpoint to indexer-api, allowing kubernetes to detect a stuck process the existing file-based liveness check could not. Fixes a dustGenerations subscription bug where the termination signal could miss in-flight wallet entries, replacing the per-wallet cursor check with chain progress and adding a re-drain to close the race window.
Dependencies
Unchanged from 4.3.1:
- Paired ledger: 8.1
- Paired node: 1.0.0 (supports 0.22.x as well)
- Bundle tracking issue: Midnight Release 1.1
- Sibling release: indexer 4.0.x maintenance line continues for operators on node 0.22.x (latest: 4.0.2)
Docker Images
Five images publish at this tag (per .github/workflows/build-indexer-images.yaml at v4.3.2):
midnightntwrk/chain-indexer:4.3.2midnightntwrk/indexer-api:4.3.2midnightntwrk/wallet-indexer:4.3.2midnightntwrk/indexer-standalone:4.3.2midnightntwrk/spo-indexer:4.3.2
Audience
These release notes are intended for:
- Operators running the Midnight Indexer against node 1.0+
- Wallet integrators consuming the
dustGenerationssubscription, especially clients passing anendIndexthat does not coincide with a wallet-owned generation index - SREs configuring kubernetes liveness probes on indexer-api
- QA and release managers tracking the Midnight 1.1 bundle
(Operators on node 0.22.x should track the 4.0.x maintenance line, not this release.)
What Changed
| Change | Type | PR |
|---|---|---|
/live HTTP endpoint for kubernetes liveness probe |
New feature | #1145 |
dustGenerations subscription terminates on chain progress, not per-wallet cursor (with race-window re-drain) |
Bug fix | #1137 |
New Features
/live HTTP endpoint for kubernetes liveness probe (#1145)
Adds an HTTP /live endpoint to indexer-api alongside the existing /ready endpoint. SREs can configure this as the kubernetes liveness probe, replacing the previous file-existence check (cat /var/run/indexer-api/running) which could not detect a stuck HTTP server. With /live, kubelet auto-recovers indexer-api pods that have entered an unresponsive state by triggering a restart after the probe failure threshold.
Operator note: this is opt-in. To use it, update the deployment manifest's livenessProbe to an httpGet against /live on the api port. The file-existence probe continues to work for backward compatibility.
Breaking Changes or Required Actions
None. Drop-in upgrade for operators already on the indexer 4.x.y development line.
Known Issues
Runtime-park hang under heavy concurrent merkle-tree-update load (midnight-ledger#521)
Under heavy concurrent merkle-tree-update load, Sp::force_as_arc (called via BackendLoader::get from make_dust_commitments_collapsed_update, make_dust_generations_collapsed_update, and related collapsed-update paths) can park the entire tokio runtime indefinitely. The pod produces zero log output (no errors, no panics, no background heartbeats) until kubelet's /live probe fails and the container is SIGTERM'd.
Per @tkerber on the ledger ticket:
Note: We are moving to a mainnet release with this as a known defect. This is likely to have the following effects:
- Indexer instances will be forcibly restarted when a deadlock/stall is detected
- Such stalls are increasingly likely to occur under heavy load
- An indexer restart may cause disconnection in connected wallet instances, which may lead to stalls in syncing until reconnected. This has not been fully tested.
We are moving ahead with a release without fixing this because:
- A prior fix attempt midnight-ledger#499 did not resolve this issue
- Evidence pinpointing the cause is very weak, and we cannot definitively attribute it.
- Restart/reconnect was deemed an acceptable mitigation while we work to find the underlying cause.
The /live HTTP endpoint shipped in this release is the detection half of that mitigation; the kubernetes restart path closes the loop. Root cause investigation ongoing at midnight-ledger#521.
Operators tracking other open issues in the indexer should consult the priority:critical and priority:high labels in the indexer GitHub issues.
Fixed Defects
| Defect | Description |
|---|---|
| #1137 | dustGenerations subscription hung in live-tail when the user-supplied endIndex did not coincide with a wallet-owned generation index. Termination signal moved from the per-wallet cursor (cursor > end_index strict) to chain progress (chain_first_free > end_index), aligning with the canonical pattern used by dustNullifierTransactions and shieldedNullifierTransactions. A re-drain step before yielding Progress closes the small race window between the storage drain ending and the chain-progress read. Wallets passing the chain's current dustGenerationEndIndex as endIndex now terminate cleanly. |
Links and References
(Per-PR links are embedded in the What Changed and Fixed Defects tables above.)
