Memory Leak of BalancedResolverGroup

[johnsiu]

Bug Report

• Driver: 0.8.8
• Database: n/a
• Java: n/a
• OS: n/a

Current Behavior

Stack trace

// your stack trace here

Table schema

Input Code

-- your SQL here;

Steps to reproduce

Input Code

// your code here;

Expected behavior/code

We use this library in our project and found a memory leak.
Heap dumps were taken and it was discovered that there were many many instances of BalancedResolverGroup in EpollEventLoop.terminationFuture.listeners.listeners. It doesn't help that a new BalancedResolverGroup is created whenever a connection is acquired. Whenever BalancedResolverGroup.getResolver() is called, a new listener is added to the EventLoop's termination listeners.
see https://github.com/netty/netty/blob/netty-4.1.65.Final/resolver/src/main/java/io/netty/resolver/AddressResolverGroup.java#L89

Possible Solution

AddressResolverGroup should almost always be a singleton.
This line should use the static instance (BalancedResolverGroup.INSTANCE) rather than instantiating.
https://github.com/pgjdbc/r2dbc-postgresql/blob/v0.8.8.RELEASE/src/main/java/io/r2dbc/postgresql/client/ReactorNettyClient.java#L392

Versions

Additional context

[kitaev]

Hello @johnsiu

I have encountered this leak too. Did you manage to find a workaround besides patching r2dbc-postgres?

[kitaev]

@mp911de I think another option to fix this leak would be to call BalancedResolverGroup.close() in ReactorNettyClient.handleClose() method.

[mp911de]

Feel free to submit a pull request. Leaving BalancedResolverGroup open is clearly a bug on our side.

[kaladhar-mummadi]

@mp911de
When Is 0.8.9 scheduled to be released ? We are facing the same issue.