fix(acl01): raise the identify size limit so the relay is still seen as a relay

[NiKrause]

Browsers cannot get a circuit reservation from the production relay, so peers appear in the list and none of them can be reached.

Cause

libp2p rejects an identify response above maxMessageSize (8192 by default) and rejects it whole. The relay announces one libp2p protocol per database it holds open — /orbitdb/heads/<address>, from OrbitDB's own sync — and has crossed that line:

Measured on the production relay	Protocols	of those /orbitdb/heads/*	identify bytes
first measurement	122	109	10538
an hour later	129	116	11056

So every browser drops the message, and /libp2p/circuit/relay/0.2.0/hop disappears from the protocol list along with bitswap and meshsub. circuitRelayTransport({ discoverRelays: 1 }) looks for relays in exactly that list, finds none, and logs not discovered enough relays 0/1. No reservation, no /p2p-circuit address.

Nothing is wrong with the relay itself: it answers hop immediately if you open the stream by name, and a client that names the relay in addresses.listen still gets a reservation. Only discovery is broken — and discovery is what this app uses.

Proof

Same client twice — listen: ['/p2p-circuit'], circuitRelayTransport({ discoverRelays: 1 }), bootstrapped to the relay, exactly as configured here. The identify limit is the only difference:

A · default 8192
   hop in the protocol list: no
   /p2p-circuit address: none after 30 s

B · maxMessageSize 65536
   hop in the protocol list: yes
   /p2p-circuit address: yes, after 2.0 s

Why 64 KiB

Today's payload is ~11 KiB and grows by roughly 72 bytes per database the relay opens. 64 KiB leaves room for about 700 of them. The cost is a larger identify message accepted from any peer, which at this size is modest.

Not the fix, only the guard

The relay side is tracked in NiKrause/orbitdb-relay#50 — which now reports the payload size on /health and in metrics so the cliff is visible — and NiKrause/orbitdb-relay#51, which asks whether one protocol id per database can work for a relay at all. Neither is resolved; this keeps browsers connecting meanwhile and is worth having regardless, since no client should lose a relay silently because a peer got chatty.

svelte-check: 0 errors, 0 warnings.

🤖 Generated with Claude Code