Avoid computing columns from EL blobs if block has already been imported

[jimmygchen]

Issue Addressed

From PeerDAS local testing I noticed some errors with failing to send computed columns to receiver:

lighthouse/beacon_node/beacon_chain/src/fetch_blobs.rs

Line 252 in dd7591f

error!(log, "Failed to send computed data columns"; "error" => ?e);

I believe there is a race condition, the sequence:

1. Fetch EL blobs triggered, received 4 blobs from the EL, triggered column computation
2. All 8 custody columns received via gossip
3. Columns written to db, block imported. (Column receiver dropped)
4. Computed columns but fail to send as receiver was dropped.

We should not even try to compute the columns if we've already imported the block from gossip. This consumes extra resources and bandwidth and is unnecessary.

[dapplion]

We should address the error log if the receiver is dropped after import

[jimmygchen]

We should address the error log if the receiver is dropped after import

I've downgraded the error to warn and added some comments here in the Fulu PR: https://github.com/sigp/lighthouse/pull/6795/files#diff-52c135e6abd229a6d90ebb9a954e3f28145ec85c311e95da86dd7a1c590e102fR251-R258

What do you propose the error handling to be? I think this PR should prevent this, and I think it's fine to just log it if it happens? If block has already been imported, then we just don't re-import and drop them .

[michaelsproul]

Agree debug log is better than a warning for the fail to send case

[jimmygchen]

Downgraded to debug and also just return without publishing as the block is likely available via gossip.
0df2384

[pawanjay176]

LGTM

[michaelsproul]

@mergify queue

[mergify]

queue

✅ The pull request has been merged automatically

Details

The pull request has been merged automatically at f008b84