Cache RMN Contract in OnRamp

[asoliman92]

Motivation

Every call to arm_contract.NewARMContract deserialized ARM contract abi under the hood. Considering that OnRamp.IsSourceCursed is called very often by the background worker it creates a huge CPU overhead. Please initialize ARMContract only once and cache it instead of doing it with every method call

Solution

• Add new cachedRmnContract field to OnRamp struct
• Cache it once per OnRamp initialization
• fix: Use correct cache function for StaticConfig
• Add Benchmark before and after caching

Note: I did benchmark the caching function against using a map with the address as key and the function caching was a bit faster. Also using the caching is more consistent with the current implementation like StaticConfig.

Benchmark results:

BenchmarkIsSourceCursedWithCache-14        90266             12526 ns/op
BenchmarkIsSourceCursedWithCache-14        88867             12411 ns/op
BenchmarkIsSourceCursedWithCache-14        90930             12379 ns/op
BenchmarkIsSourceCursedWithCache-14        91108             12259 ns/op
BenchmarkIsSourceCursedWithCache-14        90915             12306 ns/op
BenchmarkIsSourceCursedOriginal-14          3002            397356 ns/op
BenchmarkIsSourceCursedOriginal-14          2996            402220 ns/op
BenchmarkIsSourceCursedOriginal-14          3009            399806 ns/op
BenchmarkIsSourceCursedOriginal-14          2575            397963 ns/op
BenchmarkIsSourceCursedOriginal-14          2596            396812 ns/op

[github-actions]

Go solidity wrappers are out-of-date, regenerate them via the make wrappers-all command

[dimkouv]

only difference is that it's not loaded on init?

[dimkouv]

cc @mateusz-sekara

[asoliman92]

only difference is that it's not loaded on init?

Yes. When doing the benchmarks I found out that it was being called every time. Needed to wrap it within CallOnceOnNoError to actually cache it.

[makramkd]

Are these called upon creating a new CCIP job? The static config one makes an rpc call, if it errors, what happens? We want to avoid a job not starting up due to intermittent rpc errors, c.f #894

[asoliman92]

Good question! I asked it myself when I started working on it initially as well. As you can see it's wrapped within CallOnceOnNoError, so in case of errors it will not actually cache. And as this is a function definition and not actually calling while starting, it is going to be called lazily when IsSourceCursed is called. That's why I think it shouldn't affect the job starting.

[makramkd]

Ah I see, good note, so these aren't called here on these lines, but they will be called when IsSourceCursed is called, and only cached if there's no error. You're right, that doesn't block job startup.