bug: Tailcall with deduplication doesn't respond after making multiple concurrent requests

[beelchester]

Prerequisites

• [x] I have checked the existing issues to ensure this is not a duplicate.
• [x] I am running the latest version of Tailcall.

Describe the bug

On dedupe:true, If I make multiple concurrent n+1 query requests It doesn't respond.

Steps to reproduce

1. Open codespace on https://github.com/tailcallhq/graphql-benchmarks
2. use dedupe: true in graphql/tailcall/benchmark.graphql
3. run the ./run.sh script or tailcall start benchmark.graphql (for better logs)
4. In wrk/wrk.lua change wrk.body to '{"operationName":null,"variables":{},"query":"{posts{id,userId,title,user{id,name,email}}}"}'
5. run ./wrk/bench.sh.

Expected behavior

Tailcall should respond and the benchmarks should work all the time.

Actual behavior

On the first run of ./wrk/bench.sh it might work correctly but try running it 2-3 more times. It doesn't respond and the benchmark shows all 0

Screenshots

https://github.com/tailcallhq/tailcall/assets/73148455/7905b2de-45f0-419e-ad58-ef873ae772c3

Environment information:

• Operating System: Github codespace
• Tailcall Version: 0.92.6
• Rust Version: rustc 1.79.0

Additional context

I'm working on https://github.com/tailcallhq/graphql-benchmarks/pull/258

[bnchi]

For me this showed up at the 3rd run of ./wrk/bench.sh but then I restarted the server and tried to reproduce the issue and couldn't do it

Using 4 core 16GB RAM codespace

[meskill]

I've prepared a fix https://github.com/tailcallhq/tailcall/pull/2291 for this.

You can try it to see if it fixes the issue for you.

[beelchester]

Yes, it has fixed the issue for me. Thanks!