Integration Test is failing on master branch

[alamb]

Describe the bug

The "Integration Test" CI test is failing on master. For example:
https://github.com/apache/arrow-rs/runs/5421632152?check_suite_focus=true

It appears to have started in c947027 (though I don't think that PR had anything to do with it)

To Reproduce
Run integration test on master

Expected behavior
I expect it to pass

Additional context
The integration test pulls from https://github.com/apache/arrow so perhaps something upstream changed.

I think we need to sort out what is wrong with this test prior to the 10.0.0 release

[xudong963]

Maybe related to hyperium/tonic#887

[xudong963]

#864 (comment)

When updated tonic to 6.0, I found the integration test also failed.

So I guess it's the problem of tonic @alamb

[alamb]

Thanks @xudong963 👍

[alamb]

I do not yet know what is going on, but wanted to post on my progress (I don't have all that much more time to spend today on this)

Last good run @ Mar 3 9:04 AM EST: 258f828

First failed run @ Mar 3 12:47 PM EST: c947027

These commits PR from arrow appears to be in that time range, but are not obviously related to the tests

• apache/arrow@ce46c1a (update flight dependencies)
• apache/arrow@9719eae

The error appears to be happening with C++ serving and rust requesting in two scenarios:

Testing file interval_mdn
Testing file duration

I am working to reproduce the error locally so I can debug further, but as I am not super familiar with the integration testing setup, it is slow going.

[jorgecarleitao]

fwiw arrow2 is also failing on those two tests and we did not change anything in tonic, flight, IPC etc that could have caused this. I am investigating it as well.

[alamb]

FYI @jorgecarleitao

I have been able to reproduce this locally.

Here are some notes I have in case that is helpful:

# check out arrow
# install archery:
cd arrow
pip install -e dev/archery[docker]
# link arrow-rs to arrow/rust
ln -s ../arrow-rs rust

# build cpp binaries
cd arrow/cpp
mkdir build
cd  build
cmake  -DARROW_BUILD_INTEGRATION=ON -DARROW_FLIGHT=ON --preset ninja-debug-minimal ..
cd ..
ninja

Then

# build rust:
cd ../arrow-rs
cargo build --all

now from arrow directory run the tests:

archery integration --with-cpp=true --with-rust=true

Run individual test:

(arrow_dev) alamb@MacBook-Pro-2:~/Software/arrow/cpp/build$ /Users/alamb/Software/arrow/cpp/build/debug/flight-test-integration-server -port 49153

Run in rust:
/Users/alamb/Software/arrow/rust/target/debug/flight-test-integration-client --host localhost --port=49153 --path /var/folders/s3/h5hgj43j0bv83shtmz_t_w400000gn/T/arrow-integration-v7xz5285/generated_dictionary_unsigned.json

Repro with this:
cd /Users/alamb/Software/arrow-rs && RUST_LOG=debug RUST_BACKTRACE=1 CARGO_TARGET_DIR=/Users/alamb/Software/df-target cargo run --bin flight-test-integration-client -- --host localhost --port=49153 --path /var/folders/s3/h5hgj43j0bv83shtmz_t_w400000gn/T/arrow-integration-v7xz5285/generated_dictionary_unsigned.json

Results in this
E0304 16:33:43.055262000 123145352572928 hpack_parser.cc:1240]         Error parsing metadata: error=invalid value key=:scheme value=grpc

[alamb]

The output from rust is:

Error: Status { code: Unknown, message: "transport error", source: Some(tonic::transport::Error(Transport, hyper::Error(Http2, Error { kind: Reset(StreamId(1), INTERNAL_ERROR, Remote) }))) }

The output from C++ is

E0304 16:47:21.317255000 123145352572928 hpack_parser.cc:1240]         Error parsing metadata: error=invalid value key=:scheme value=grpc

Which seems to suggest some sort of bug in parsing headers or something

However, since other tests run, maybe it is something related to the contents?

Given how it came on suddenly, smells like something related to a change in a dependency, but I haven't found one yet

[matthewmturner]

apologize if ive misunderstood any part of this or if i dont understand full intergration testing process - but given that both arrow and arrow2 started failing i am focusing on two scenarios:

1. Dependency that both arrow and arrow2 use which caused issue. @alamb and @jorgecarleitao are each looking into this - so i consider this "covered" in the sense that its being looked into.
2. Issue from C++ (content or dependency) that's impacted both. I know there was an email sent to dev email list mentioning the integration test fail but its not clear to me if anything beyond that has been raised on that side. Should we be raising something on that side?

[alamb]

@matthewmturner -- I agree with your assesment.

Interesting the integration tests appear to fail inhttps://github.com/apache/arrow/ as well with the same errors:

https://github.com/apache/arrow/runs/5427260660?check_suite_focus=true

I am looking into that as well