lnruntime: Compose native Ark channel state machines - #1191
Conversation
Consume the reviewed lnd fork that exposes external funding and lifecycle barriers without embedding a second daemon.
Instantiate lnd channel, link, invoice, payment, funding, and close state machines behind one process-owned runtime.
Expose chain notification and fee boundaries required by the native channel runtime over the existing wallet backends.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5e0f7c6a9f
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| return | ||
| } | ||
|
|
||
| case _, ok := <-registration.Reorged: |
There was a problem hiding this comment.
When a confirmation, reorg, and reconfirmation are buffered concurrently, the select may receive the newer Confirmed event before the older Reorged event. This branch discards the reorg sequence number, so it cannot reject that stale reorg and can leave the downstream lnd watcher unconfirmed even though the transaction has already reconfirmed. Track the highest shared TxConfirmation.Seq/reorg sequence and suppress older events.
Useful? React with 👍 / 👎.
| return | ||
| } | ||
|
|
||
| case _, ok := <-registration.Reorged: |
There was a problem hiding this comment.
If a spend, its reorg, and a subsequent re-spend are all buffered, this select can forward the newer re-spend first and then forward the stale reorg because the shared sequence value is ignored. The downstream lnd resolver can consequently finish in the unspent state and miss the current spender; compare SpendDetail.Seq with the reorg sequence and discard stale events.
Useful? React with 👍 / 👎.
| if epoch.Height == lastHeight && | ||
| epoch.Hash == lastHash { | ||
|
|
||
| continue | ||
| } |
There was a problem hiding this comment.
During a burst of blocks, more than one epoch can be queued after RegisterBlocks but before BestBlock returns. For example, if heights H+1 and H+2 queue while BestBlock reports H+2, the goroutine seeds H+2 and then forwards H+1 followed by H+2 because it only removes exact duplicates, producing a regressing block stream. Ignore queued epochs older than the snapshot while still allowing same-height hash changes for reorg handling.
Useful? React with 👍 / 👎.
| func (n *BackendChainNotifier) Stop() error { | ||
| n.started.Store(false) |
There was a problem hiding this comment.
When this adapter is stopped while the shared backend remains running, every outstanding confirmation, spend, and block forwarder continues indefinitely because its registration context is rooted at context.Background() and Stop retains no cancellation handles. started also does not prevent new registrations after shutdown. Give the adapter a lifecycle context or track registration cancels so Stop terminates its own workers without stopping the shared backend.
AGENTS.md reference: AGENTS.md:L77-L81
Useful? React with 👍 / 👎.
sputn1ck
changed the title
The durable Ark channel FSM must not expose an active channel until the native lnd link for its reserved SCID can actually carry HTLCs. Requiring both the open channel row and switch link removes a race where the first payment could terminally fail immediately after promotion.
eacb4d1 to
13aac6e
Compare
1 participant
Summary
This is layer 3 of 5 in the Ark channels stack. It depends on #1190; the next layer is #1192.
Testing
go test ./lnruntimego test ./oor ./arkchannel/... ./lnruntime ./waved ./dbmake buildmake lint-changed-local