describegraph contains edges with zero capacity

[wereHamster]

The describegraph command produces a JSON with lots of channels that have zero capacity, but are enabled on both sides, and according to both amboss and 1ml are still active.

For example this edge:

{
  "channel_id": "616942571506761729",
  "chan_point": "0000000000000000000000000000000000000000000000000000000000000000:0",
  "last_update": 1749533813,
  "node1_pub": "0217890e3aad8d35bc054f43acc00084b25229ecff0ab68debd82883ad65ee8266",
  "node2_pub": "02450500fa04c03e76ce1f01d5338006111bf5347fabfc14bc2344d4193dfd8abb",
  "capacity": "0",
  "node1_policy": {
    "time_lock_delta": 144,
    "min_htlc": "1000",
    "fee_base_msat": "1000",
    "fee_rate_milli_msat": "1",
    "disabled": false,
    "max_htlc_msat": "382295000",
    "last_update": 1749533813,
    "custom_records": {},
    "inbound_fee_base_msat": 0,
    "inbound_fee_rate_milli_msat": 0
  },
  "node2_policy": {
    "time_lock_delta": 144,
    "min_htlc": "1000",
    "fee_base_msat": "1000",
    "fee_rate_milli_msat": "1",
    "disabled": false,
    "max_htlc_msat": "378473000",
    "last_update": 1749512371,
    "custom_records": {},
    "inbound_fee_base_msat": 0,
    "inbound_fee_rate_milli_msat": 0
  },
  "custom_records": {}
}

• On 1ml
• On amboss

Weird that chan_point is null(?), there seems to be a strong correlation in the describegraph JSON between chan_point being null and zero capacity.

I was running the node with a neutrino backend in the beginning, before switching to a local bitcoind backend. Maybe this has something to do with that?

Any idea how to trigger a re-scan so that lnd will find the correct chan_point and populate the capacity?

[wereHamster]

I also see a lot of [WRN] GRDB: Channel=??? not found in graph cache, even though the channels are active (according to 1ml or amboss)... I really wonder if my LND has corrupted graph cache..

[ellemouton]

I was running the node with a neutrino backend in the beginning, before switching to a local bitcoind backend. Maybe this has something to do with that?

Yes, this is probably the cause as if you are running neutrino then it doesnt lookup & validate the funding transactions of an announced channel & therefore does not fetch the actual outpoint and channel capacity.

Any idea how to trigger a re-scan so that lnd will find the correct chan_point and populate the capacity?

Currently there is no safe way to do this that I'm aware of. In future, we can potentially make it possible to add a "gossip resync" option which basically would start with an empty graph store and re-sync gossip with peers (BUT dont do this now by deleting the graph store because this would also delete your own nodes channel&node announcements).
Other option is that we can add specific functionality to address this: ie, iterate through all our channels and re-validate them (and ie, fetch their capacity&outpoints)

[guggero]

I think you can restore the auth proofs by updating all your channel policies. See lncli updatechanpolicy.

[wereHamster]

I don't see anything in my updatechanpolicy command to inject auth proofs (v0.19.1).

I eventually fixed this by starting up another LND node to construct the full graph, including auth proofs (via ChannelAnnouncement gossip messages), and then injecting them into my channel.db via a custom chantools command (graphDB.AddEdgeProof(shortChanID, &proof)).

I think it's unfortunate that LND rejects ChannelAnnouncement messages for its own channels that are coming from peers. If it allowed that, it would be possible to bootstrap the whole network graph without manual intervention. Or is there something in the protocol that would forbid other nodes sending me my own channel announcements?

There seem to be nodes in the lightning network that send ChannelAnnouncement messages of my own channels to my node. In my logs I see these messages:

Error processing remote msg *lnwire.ChannelAnnouncement1: ignoring remote ChannelAnnouncement1 for own channel

[guggero]

I don't see anything in my updatechanpolicy command to inject auth proofs (v0.19.1).

You don't use the command to "inject" auth proofs. You generate new ones by sending out a new channel update that is singed by your node.

[wereHamster]

So you're saying that by updating the channel policy, lnd will generate new auth proofs and write into the local channel.db?

[guggero]

Yes.

[ellemouton]

see a lot of [WRN] GRDB: Channel=??? not found in graph cache, ev

Something to note here is that this log line was recently added. So if there is a bug here, it is probably not new - it is just that we were not logging this before.