feat!: switch back to protons

[twoeths]

Motivation

• We should switch back to protons when it's performance is comparable or better than protobufjs
• Prepare a branch to track the performance of different versions of protons

Description

Update Jul 2023: the performance is comparable to protobufjs with protons 5.1.0

protobuf
    ✔ decode Attestation message using protons 5.1.0                       3446018 ops/s    290.1900 ns/op        -    2053201 runs   60.1 s
    ✔ encode Attestation message using protons 5.1.0                       2506831 ops/s    398.9100 ns/op        -    1492091 runs   60.0 s

protobufjs

protobuf
    ✔ decode Attestation message using protobufjs                          3177024 ops/s    314.7600 ns/op   x0.960    1890821 runs   60.1 s
    ✔ encode Attestation message using protobufjs                          2500563 ops/s    399.9100 ns/op   x1.012    1485962 runs   60.0 s

Closes #453
Closes #318

[codecov-commenter]

Codecov Report

Patch coverage: 74.78% and project coverage change: -2.31% ⚠️

Comparison is base (23bf0ee) 82.30% compared to head (ba73f6a) 80.00%.
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #327      +/-   ##
==========================================
- Coverage   82.30%   80.00%   -2.31%     
==========================================
  Files          48       48              
  Lines       12039    11105     -934     
  Branches     1286     1152     -134     
==========================================
- Hits         9909     8884    -1025     
- Misses       2130     2221      +91     

Files Changed	Coverage Δ
src/metrics.ts	16.43% <3.98%> (-2.57%)	⬇️
src/index.ts	70.11% <63.73%> (+0.14%)	⬆️
src/message/rpc.ts	88.05% <88.05%> (ø)
src/message/decodeRpc.ts	93.71% <94.24%> (+0.99%)	⬆️
src/message-cache.ts	78.06% <100.00%> (ø)
src/types.ts	94.57% <100.00%> (ø)
src/utils/buildRawMessage.ts	92.21% <100.00%> (ø)
src/utils/create-gossip-rpc.ts	100.00% <100.00%> (ø)
test/decodeRpc.spec.ts	100.00% <100.00%> (ø)
test/message-cache.spec.ts	100.00% <100.00%> (ø)
... and 2 more

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[twoeths]

not sure why using proton-runtime 5.1.0 the performance is 2x slower

protobuf
    ✔ decode Attestation message using protons 5.1.0                       1645793 ops/s    607.6100 ns/op        -     983127 runs   60.1 s
    ✔ encode Attestation message using protons 5.1.0                       1630497 ops/s    613.3100 ns/op        -     972079 runs   60.0 s

should be good to go with proton-runtime 3.1.0

[twoeths]

this is not ready, got memory leak issue as noted in #318 (comment)

[twoeths]

rss is 1GB-1.5GB more than protobuf version, this is from a lodestar test mainnet node

  message ControlPrune {
    optional string topicID = 1;
    repeated PeerInfo peers = 2;
    optional uint64 backoff = 3;
  }

while with protobuf it's mostly <=6.2GB. I think the main change is backoff field being intepreted as bigint while protobufjs used number instead

 interface IControlPrune {

        /** ControlPrune topicID */
        topicID?: (string|null);

        /** ControlPrune peers */
        peers?: (RPC.IPeerInfo[]|null);

        /** ControlPrune backoff */
        backoff?: (number|null);
    }

[wemeetagain]

https://github.com/ipfs-shipyard/protons-gossipsub-benchmark

@tuyennhv might be worth testing again with the latest protons

[twoeths]

we'll need to go with latest protons 7 anyway in order to use backoff field as a number, track this work in #468 instead