Node Hangs on Cold Start with External Signer + Batch Poster

[Bidon15]

Fresh node startup hangs indefinitely when batch-poster.enable=true with external signer configuration. The hang occurs after TLS certificate loading but before HTTP server starts.

Environment

• Nitro Version: main 1f915ab
• Parent Chain: Sepolia
• Configuration: External signer for batch poster with mTLS client certificates

Configuration

--node.batch-poster.enable=true
--node.batch-poster.data-poster.external-signer.url=<url>
--node.batch-poster.data-poster.external-signer.address=<address>
--node.batch-poster.data-poster.external-signer.method=eth_signTransaction
--node.batch-poster.data-poster.external-signer.client-cert=/certs/client.crt
--node.batch-poster.data-poster.external-signer.client-private-key=/certs/client.key
--node.batch-poster.data-poster.external-signer.root-ca=/certs/ca.crt
--node.staker.enable=false

Steps to Reproduce

1. Deploy a new Orbit chain
2. Start node with above configuration and empty data directory
3. Node hangs indefinitely

Observed Behavior

Last logs before hang:

INFO [...] DA providers configured                  totalWriters=1
INFO [...] Client certificate for external signer is enabled

Node never reaches:

INFO [...] HTTP server started endpoint=[::]:8547

Key Observations

1. Cold start hangs: Fresh data directory + batch-poster + external-signer = hang
2. Warm start works: If node is first started with batch-poster.enable=false, then restarted with batch-poster.enable=true, it works
3. Certificate logs appear twice: Even with staker disabled, the "Client certificate for external signer is enabled" message appears twice before the hang

Workaround

# 1. Start with batch-poster disabled
--node.batch-poster.enable=false

# 2. Wait for HTTP server to start

# 3. Restart with batch-poster enabled
--node.batch-poster.enable=true

Node starts successfully after this sequence.

Notes

• External signer is reachable (verified independently)
• Alt-DA is used
• Certificates are valid (log message confirms successful loading)
• Issue only occurs on fresh/cold start
• Once database is initialized, subsequent restarts with batch-poster enabled work fine

---

[peter941221]

I traced the current master path for this and found one concrete startup-blocking mechanism that still exists in the current code.

What I confirmed locally:

• cmd/nitro/nitro.go creates the consensus node before stack.Start()
• inside that path, arbnode.getBatchPoster(...) calls NewBatchPoster(...)
• NewBatchPoster(...) immediately calls dataposter.NewDataPoster(...)
• NewDataPoster(...) immediately dials the external signer when data-poster.external-signer.url is set
• that dial goes through rpc.DialOptions(...) with the configured mTLS client-certificate setup

So a stalled external-signer connection can block cold startup before the HTTP server is up, which matches the "hang after certificate logs but before HTTP server started" symptom.

I have not yet fully explained the cold-start vs warm-start asymmetry from code alone, and the duplicate Client certificate for external signer is enabled log still suggests there may be another initialization path touching the same config. So I would treat that part as still open rather than over-claiming it here.

The narrowest fix shape looks like:

• keep the configured sender address available during construction
• but defer the actual external-signer dial out of NewDataPoster(...)
• and only connect in batch-poster Start(...) or on first real signing use

That would remove the confirmed pre-HTTP blocking path without widening into a larger batch-poster redesign.

One validation caveat from my local clone: I could not run the full arbnode/dataposter or cmd/nitro package tests yet because this checkout is missing the generated solgen/go/* bindings that the current Go build expects. But the constructor chain above is from the current source tree, not a stale snapshot.