perf: Fix cloning of executor

[dima74]

Description

I am debugging tps of a single peer (#4727), measured performance of Sumeragi::try_create_block, and noticed that executor cloning takes about 25% of try_create_block. This is because of LoadedExecutor::raw_executor which was added for snapshot creation. Could be fixed by adding Arc.

It is also interesting that other wasm-related code takes a lot of time, e.g. wasmtime::Linker::instantitate and wasmtime::Store::into_data.

Linked issue

Related: #4727

Benefits

Slightly optimize block creation performance

Checklist

• I've read CONTRIBUTING.md
• I've used the standard signed-off commit format (or will squash just before merging)
• All applicable CI checks pass (or I promised to make them pass later)
• (optional) I've written unit tests for the code changes
• I replied to all comments after code review, marking all implemented changes with thumbs up

[github-actions]

@BAStos525

[mversic]

so this comment was correct until executor wasm was added to snapshot. Was using wasmtime::Module::serialize considered as a solution instead of adding LoadedExecutor::raw_executor?

[dima74]

Was using wasmtime::Module::serialize considered as a solution instead of adding LoadedExecutor::raw_executor?

wasmtime::Module::serialize returns something different from what LoadedExecutor::raw_executor currently stores (I compared lengths of inner Vec<u8> and it differs)

[mversic]

Was using wasmtime::Module::serialize considered as a solution instead of adding LoadedExecutor::raw_executor?

wasmtime::Module::serialize returns something different from what LoadedExecutor::raw_executor currently stores (I compared lengths of inner Vec<u8> and it differs)

yes, I know that. What I'm saying is if you can directly serialize/deserialize the wasmtime::Module, you don't need raw_executor in LoadedExecutor

[dima74]

if you can directly serialize/deserialize the wasmtime::Module, you don't need raw_executor in LoadedExecutor

Changed to serialize wasmtime::Module directly. Note that this will increase size of snapshot by 1.5MB (our default executor is 0.5MB, serialized wasmtime::Module is around 2MB)

[Erigara]

I think changing to wasmtime::Module could cause problems because serialize/deserialize only works for the same version of wasmtime and in other case can produce errors.
It might be dangerous as well because random changes in serialized snapshot can cause memory safety issues.

EDIT: docs to deserialize.

[Erigara]

I think we can't guarantee that...

[mversic]

I think changing to wasmtime::Module could cause problems because serialize/deserialize only works for the same version of wasmtime and in other case can produce errors.

How can we end in this situation that we serialize with one and deserialize with another version of wasmtime?

[Erigara]

How can we end in this situation that we serialize with one and deserialize with another version of wasmtime

@mversic update of iroha version which in theory should preserve backward compatibility for block store and snapshot.

[dima74]

So we should change back to LoadedExecutor::raw_executor approach

[dima74]

Opened #5009