feat(serve): full ogmios v7 interop for the ouroboros N2C endpoint

[Javieracost]

Depends on txpipe/pallas#792 — the first commit bumps the pallas dependency to that revision so every commit in this PR builds standalone.

What / Why

Running ogmios v7 on top of dolos 1.5.0's ouroboros N2C endpoint fails at three independent layers today:

1. Sessions are killed at open. Ogmios opens local-tx-monitor on every connection; the endpoint only implements chainsync and statequery, so the multiplexer tears down the whole session as soon as protocol 9 is opened.
2. Statequery dies while the node syncs. Ogmios's health client acquires the volatile tip on connect and on every tip tick. Dolos validated the cursor block against the archive store — which lags behind the cursor during catch-up — and answered Failure(PointNotOnChain) to a request that cannot legitimately fail. Worse, the request loop entered the acquired state even after sending a failure, so the client's spec-correct follow-up acquire was rejected as invalid and the whole statequery protocol died.
3. Historical chain-sync hangs silently. Any client intersecting below the WAL window (ogmios, kupo, any N2C syncer) gets findIntersection instantly and then waits forever for the first block. Four distinct defects: the initial rollback was sent to the first crawled block's point instead of the negotiated intersection (consuming that block — and archive-crawled points are slot-only, so the send failed outright); that failure was silently discarded, leaving the protocol dead while the client waited; crawling from origin skipped the first block of the chain; and a client awaiting at the tip died on the next re-org, because an undo tip event returned without sending the owed reply, leaving the server receiving without agency.

This PR fixes all of the above — one commit per concern, each self-contained with protocol-level integration tests (a real pallas NodeClient over a unix socket against the session handler):

• feat(serve): add local-tx-monitor server to the ouroboros N2C endpoint — mempool-backed sessions: a snapshot is pending + inflight transactions (Confirmed excluded, as a node mempool would not hold them) tagged with the cursor slot; era-wrapped NextTx payloads (block era − 1, same convention as the tx-submission peer client); HasTx matches on hash alone so era-probing clients get correct answers; capacity mirrors the cardano-node default (2× max block body size); AwaitAcquire blocks on snapshot fingerprint change, waking on mempool events or tip changes.
• fix(serve): stop failing statequery volatile-tip acquires during sync — volatile-tip acquires acquire the current cursor directly (the node's own tip needs no archive validation), and a failed acquire or re-acquire returns the loop to the idle state.
• fix(serve): serve historical chain-sync from the archive — the opening rollback echoes the negotiated intersection without consuming a block; chain-sync session errors are logged and propagated like the other mini-protocols; origin batches start at the first block in the archive; the await loop skips undo events until the mark or apply that follows produces the owed reply (undos have no wire counterpart — the mark that follows becomes the rollback).

The three concerns are one story — making dolos a drop-in N2C backend for ogmios — but each commit stands alone; we can split the PR if you'd prefer smaller units.

Validation

Beyond the test suite, we validated end-to-end on mainnet: a full-history dolos + ogmios v7.0.0 replacing cardano-node + ogmios as the feed for a mainnet indexing pipeline.

• Derived state came out byte-identical to the node-backed reference across two legs — genesis → epoch 300, and epochs 500 → 510.
• The ogmios health endpoint holds connected / networkSynchronization 1.0 at tip; the undo-at-tip agency fix was caught live by a mainnet re-org during the soak.
• Beyond correctness, the swap simply helped us: about 3× the historical sync throughput at roughly a quarter of the memory.

Out of scope (observed while validating, pre-existing)

• Crawling past the archive end during catch-up can hit the no overlap between archive and wal panic when the crawl cursor is a slot-only point.
• Byron EBBs are not present in the slot-keyed archive, so archive-sourced chain-sync doesn't stream them (they carry no transactions, so ledger derivations are unaffected, but noting it).

Happy to follow up on either separately.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 8a53b6ed-cdc4-48dc-802e-5596c8872c33

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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