Sync issue

[tbenr]

Context

When we requests new blocks\blobs we run:

public void requestMoreBlocks(final Runnable callback) {
...
    SafeFuture.allOfFailFast(blocksRequest, blobSidecarsRequest)
        .thenRunAsync(
            () -> onRequestComplete(blockRequestHandler, maybeBlobSidecarRequestHandler),
...
}            

Down the line, blocksRequest and blobSidecarsRequest call:

private <I extends RpcRequest, O extends SszData> SafeFuture<Void> sendEth2Request(

Problem

When we are rate limited or by the node we are requesting data from, the node will disconnect us by sending a GoodbyeMessage.

In that scenario we close all connections and we and up calling RpcHandler::onRemoteWriteClosed which will call Eth2OutgoingRequestHandler::readComplete.
So we will end up considering all inflight requests to the peer as completed and run onRequestComplete as nothing happened.

In that method we do an early validation of blocks and blobs. The first validation is a simple one and unlikely to fail even if we are dealing with incomplete block response. The blobSidecar validation (validateNewBlobSidecars) easely fails if we receive less blobs than expected. This could happen if the rate limit happens on the blobsByRange portion of the batch, while blocksByRange is not rate limited.

When this validation fails, we consider the peer as behaving badly and we apply a big penalty.

Possible solution

sendEth2Request could start giving more feedback instead of returning SafeFuture<Void> to be able to discriminate between an actual completed request and a request that has been interrupted along the way.