Memory & connection leak in reactor-netty instrumentation

[raptium]

Describe the bug
When using WebClient in Spring WebFlux, a SimpleDequePool is created on every HTTP request and never gets disposed.

Steps to reproduce

1. Included spring-boot-starter-webflux as a dependency
2. Create a org.springframework.web.reactive.function.client.WebClient
3. Use this client to send some HTTP requests

What did you expect to see?

1. Sending requests with one WebClient should create only one connection pool and reuse connection if possible.

What did you see instead?

1. Lots of SimpleDequePool live in heap and lots of TCP connections are established. None of these will get disposed after GC.
2. Server gets OOM or reaches limit of num of FD eventually.

What version are you using?
1.9.1

Environment
Compiler: openjdk 1.8.0_312
OS: macOS 12.0.1

Additional context

• PooledConnectionProvider use PooledConnectionProvider.PoolKey as key for its internal channelPools (ConcurrentHashMap).
• Hashcode of PooledConnectionProvider.PoolKey is computed with pipelineKey and other fields.
• pipelineKey is assigned with TransportConfig.channelHash()
• channelHash() is computed with observer and other fields.
• observer is a different instance of HttpResponseReceiverInstrumenter.EndOperation on each HTTP request and it may cause PooledConnectionProvider to create a new pool every time

[sh777]

Hi, this is quite a critical issue in current release. I wonder when will we be able to release the fix?

[trask]

1.10.0 is targeted for end of next week. This issue does meet our patch release policy if you need it sooner.

[sh777]

Hi @trask, thanks a lot! We indeed need it urgently. We really appreciate if we could have a patch for it.

[trask]

@sh777 1.9.2 is released with this fix