penumbra.util.tendermint_proxy.v1.TendermintProxyService.GetBlocksByHeight seems broken

[turbocrime]

Describe the bug

GetBlocksByHeight fails to respond for any attempted height

To Reproduce
Steps to reproduce the behavior:

1. Go to https://grpcui.testnet.penumbra.zone/
2. Use Service name penumbra.util.tendermint_proxy.v1.TendermintProxyService
3. Use Method name GetBlockByHeight
4. Input a valid height
5. Invoke

Expected behavior

Block data is returned

Screenshots

[zbuc]

I'm trying to write a test script for it locally but having some trouble.

I can call it with grpcurl at least, but the golang protobuf library isn't playing nicely with the way cometbft encoded timestamps:

grpcurl -d '{ "height": 326710 }' grpc.testnet.penumbra.zone:443 penumbra.util.tendermint_proxy.v1.TendermintProxyService.GetBlockByHeight
Failed to format response message 1: ns out of range [0, 1000000000)

and I can't find a command line flag to tell grpcurl to print the raw response

I then tried writing a test script in Rust, and was able to call against localhost, but not the testnet endpoint:

let mut tm_client = TendermintProxyServiceClient::connect("http://localhost:8080").await?;
let r = tonic::Request::new(GetBlockByHeightRequest { height: 326_710 });
let res = tm_client.get_block_by_height(r).await?.into_inner();

^^ that works

let mut tm_client = TendermintProxyServiceClient::connect("https://grpc.testnet.penumbra.zone:443").await?;
let r = tonic::Request::new(GetBlockByHeightRequest { height: 326_710 });
let res = tm_client.get_block_by_height(r).await?.into_inner();

^^ that doesn't:

Error: Status { code: Internal, message: "protocol error: received message with invalid compression flag: 52 (valid flags are 0 and 1) while receiving response with status: 404 Not Found", metadata: MetadataMap { headers: {"content-type": "text/plain; charset=utf-8", "x-content-type-options": "nosniff", "content-length": "19", "date": "Tue, 04 Jun 2024 17:54:11 GMT"} }, source: None }

I have no idea why it would get a 404 on grpc.testnet.penumbra.zone but not localhost, especially considering that I was able to grpcurl the endpoint.

[zbuc]

The underlying issue with using grpcui to call is the same golang parsing issue. If you use the Chrome network inspector and check the response you'll find:

  "responses": [
    {
      "message": "ns out of range [0, 1000000000)",
      "isError": true
    }
  ],

[zbuc]

Figured it out, that was Tonic feature flags.

I'm able to call the API from rust code now and get a response so I don't think the API is actually broken however there is an issue in that historical data is unavailable post-upgrade. I know @conorsch has been looking at things in that domain, maybe there's a fix?

[turbocrime]

nanos from the endpoint appear to be negative, which is against spec

https://github.com/protocolbuffers/protobuf/blob/bf7ac9f2f11e15c1a8f4b41428a75ad1c039a8fe/src/google/protobuf/timestamp.proto#L133-L144

message Timestamp {
  // Represents seconds of UTC time since Unix epoch
  // 1970-01-01T00:00:00Z. Must be from 0001-01-01T00:00:00Z to
  // 9999-12-31T23:59:59Z inclusive.
  int64 seconds = 1;


  // Non-negative fractions of a second at nanosecond resolution. Negative
  // second values with fractions must still have non-negative nanos values
  // that count forward in time. Must be from 0 to 999,999,999
  // inclusive.
  int32 nanos = 2;
}

[turbocrime]

notably, timestamp with negative nanos are unable to re-transmit to the client through our web communication channel

[zbuc]

Using an int32 to hold a value that's supposed to be positive is chaotic

[zbuc]

Closing as this isn't actually "broken" (well besides the nanos thing) and we need to build our own RPC endpoint