[bug]: Local Channel all of a sudden not in local cache

[ziggie1984]

Pre-Submission Checklist

• I have searched the existing issues and believe this is a new bug.
• I am not asking a question about how to use lnd, but reporting a bug (otherwise open a discussion).

LND Version

20 rc2

LND Configuration

native sql - postgres

Backend Version

core v0.29

Backend Configuration

OS/Distribution

Bug Details & Steps to Reproduce

For unknown reason a direct peer got dropped of the channel cache and it was impossible to readd it to the cache via a local channelUpdate on that channel.

When triggering the ChannelUpdate seeing a lot of:

GRDB: Channel=XXXX not found in graph cache

Only a restart fixed the issue, the channel was basically unusable because we use the cache for pathfinding and forwarding.

Expected Behavior

Don't drop local channels from Graph-Cache

Debug Information

No response

Environment

No response

[Filouman]

LND Version
v0.20.0-beta.rc1

LND Configuration
PostgreSQL (native SQL)

Backend Version
core v0.30

I am seeing lots of these similar logs as well:

[WRN] GRDB: Channel=892905696488390657 not found in graph cache
[WRN] GRDB: Channel=807461548420104193 not found in graph cache
[WRN] GRDB: Channel=807461548420104193 not found in graph cache
[WRN] GRDB: Channel=970410271032934400 not found in graph cache

and also
[WRN] GRDB: Node=03aa17447bdadb3d5efa388ec1340f26264bc6059090479938fad0e5b425373fec not found in graph cache

But they are not my direct peers or channels.

I am also noticing "synced_to_graph": false, after a recent restart. Usually graph sync is very quick and only takes a few minutes maximum after a restart. It's been 45 minutes and still is false.

I may try to upgrade to v0.20.0-beta.rc2 to see if there is any change.

[Filouman]

Update LND finally has synced_to_graph": true, but took almost 1 hour which seemed unusual.

[bitromortac]

Was the graph dropped at some point, e.g. using chantools (to force a resync)?

[ziggie1984]

The channel in question was not usable anymore after the Peer had almost a 1 day downtime. But this node stayed online all the time. Maybe it has something to do with the SQL implementation, maybe we drop the channel/node there for some reason. The graph stuff is new ... (no chantools was used)