Notes
Where we differ from the official docs
- App keys. We use gasless keys (
KEY_ADD). Most guides still show the onchain key registry. - Writes. We send through our own read node. It relays to validators.
- Ports. We keep 3381 and 3383 private and use an SSH tunnel.
- Disk. Official docs say 1.5 TB in one place and 2 TB in another. With the node's own restore, use 2 TB: one large shard alone took 744 GB on our run, and the peak lands near 1.6 TB. runfcx streams the restore itself, so 1000 GB is enough with fcx.
Troubleshooting
Every message runfcx can print, with the fix, is in the agent runbook under Errors.
What we sent upstream
- Approving gasless keys from the Farcaster app. Most custody keys live inside the app, and a verified wallet can't sign
KeyAdd. We asked for a gasless signer-request flow: comment on FIP #266. - A smaller snapshot restore. Streaming chunks into
tarso no archive is written, and resuming per shard: comment on snapchain #1046. - A docs fix.
castsByFidalso returns deleted casts asCAST_REMOVEmessages, which crashed our client in run 2: snapchain PR #1058.