Keeping Ref of dashmap over await points can lead to deadlocks

[dkales]

This might be relevant for you @bgillesp @naure @iliailia: We did some experiments/benchmarks with the GPRC networking and used the impl in this repository as a starting point.

In our scenarios we managed to deadlock it quite consistently, due to it holding the Dashmap Ref you get from .get over await points.

e.g.

iris-mpc/iris-mpc-cpu/src/network/grpc.rs

Line 56 in d11df4b

let queue = self.queues.get(sender_id).ok_or(eyre!(format!(

Changing it to tokio::sync::RwLock<HashMap> does fix it, but might not be the greatest for perf.

IDK if you have hit that in your current development

[naure]

That’s correct. I’m reviewing all these usages at the moment and probably move to async mutexes.