HTTP errors are not being commited since version 1.5.3

[tmpolaczyk]

Starting from the 1.5.3 release, when a node encounters an HTTP error or a timeout in one of the sources of a data request, it will not commit any value at all.

Run a git bisect and e33986d is the first bad commit.

Besides, one single failing source would make the node to infer that the whole data request failed and, again, the node would refrain from committing any value or error at all. This is something that was presumably fixed before 1.5.3.

[tmpolaczyk]

The affected nodes show this log line:

witnet-rust/node/src/actors/chain_manager/mining.rs

Line 566 in 2fa9a2f

log::error!("Couldn't resolve rad request {}: {}", dr_pointer, e);

which means that the ResolveRA message returns an Err. So we need to ensure that the ResolveRA handler only returns an Err when we want the node to refrain from committing any value at all.

[aesedepece]

@tmpolaczyk nice catch. Very interested about what could be the actual consequences of this, specially its impact on resolution of data requests, and reputation / collateral slashing.

[tmpolaczyk]

@aesedepece when a data source returns an http error, nodes running on 1.5.3 or 1.5.4 will not participate in the resolution of that data request. So we should see an "Insufficient commits" error in mainnet. But instead we see the "http error", even when it shouldn't be the result (in case of 1 error and 4 non-error sources). So we believe that there are some nodes running modified versions of the node, and they are the ones participating in these requests.

[guidiaz]

@aesedepece @tmpolaczyk What if we let nodes refrain from committing errors only if majority of errors correspond to either http/401 (unauthorized) or http/403 (forbidden) ??