Memory Leak in reactor-netty when using Spring WebClient with blocked flux. Issue: Databuffer is not released. about reactor-netty HOT 15 CLOSED

@KrishnakanthYachareni From the configuration I don't see any changes related to the memory management, which means that you are running with Pooled ByteBufs. Is it possible that you see this reserved memory?
Check this https://www.programmersought.com/article/9322400832/ ,
https://netty.io/wiki/reference-counted-objects.html and may be this https://projectreactor.io/docs/netty/release/reference/index.html#faq.memory-leaks

from reactor-netty.

@violetagg Yes, my application has been using default Pooled ByteBus, as I did not have any manual configuration related to mememory management. You can see below it is a reserved memory.

If you see the memory usage of the pod which is gradually increasing in the pod and memory is not collecting by GC.

Why does ByteBuf memory is not clearing, is the following snippet not releasing the buffer properly?

Flux<FcmResponseWrapper> response = fcmRestClient.sendPush(fluxRequests, appName, eventIdentifiers);

List<Tuple2<Long, FcmResponseWrapper>> listOfSendResponses = response.elapsed()
                  .collectList()
                  .block(); // Thread blocking.

from reactor-netty.

Is it possible to enable memory metrics provided by Reactor Netty
https://projectreactor.io/docs/netty/release/reference/index.html#_metrics_6
reactor.netty.bytebuf.allocator.active.heap.memory
reactor.netty.bytebuf.allocator.active.direct.memory

from reactor-netty.

@violetagg Just to clarify that in my code no other objects are allocating more memory and memeory is leaking from reactor related objects. That I verfied during JVM profiling.

bytebuf are not emitted but ConnectionProvider metrics are emitting. Following code used to enable the reactor metrics. am I missing anything?

Schedulers.enableMetrics(); // in Spring boot main method.

private ConnectionProvider getConnectionProvider() {
        return ConnectionProvider.builder("fcmConnection")
                .metrics(true)
                .build();
    }

    @Bean
    public NettyServerCustomizer nettyServerCustomizer(){
        return httpServer -> httpServer.metrics(true, Function.identity());
    }

My suspect is .subscribeOn(Schedulers.fromExecutor(executor) method with reactor scheduler may causing the leak, Can you please validate the code?

messages.flatMap(msgPair -> webClient.post()
                            .uri(appURI)
                            .header(HttpHeaders.AUTHORIZATION, accessToken)
                            .body(Mono.just(msgPair.getFirst()), FirebaseRequest.class)
                            .exchangeToFlux(clientResponse -> this.handleResponse(clientResponse, msgPair.getSecond())))
                    .subscribeOn(Schedulers.fromExecutor(executor));

from reactor-netty.

Add this to the client httpClient.metrics(true, Function.identity());

from reactor-netty.

@violetagg I collected the byteBuff memory metric and both seems always constant and same value.

from reactor-netty.

@KrishnakanthYachareni You need the active not the overal memory that is reserved

reactor.netty.bytebuf.allocator.active.heap.memory
reactor.netty.bytebuf.allocator.active.direct.memory

vs

reactor.netty.bytebuf.allocator.used.heap.memory
reactor.netty.bytebuf.allocator.used.direct.memory

from reactor-netty.

@violetagg My bad, I miss understood the mertics name. Both memories are most of the time showing as 0 but some time direct memory have certain value and active heap memory is always 0. what does it mean?

Note: Just to mention Non-Heap memory is gradually increasing and not going down.

from reactor-netty.

Took this screenshot after few hours of running and later again both memories went down to 0

from reactor-netty.

@KrishnakanthYachareni So to summarise the active memory is not increasing constantly but it returns to 0?

from reactor-netty.

@violetagg Yes. that is right!

from reactor-netty.

@KrishnakanthYachareni So looks like no memory leaks?

from reactor-netty.

@violetagg I did these changes, this could be a reason memory leak got fixed.

From

webClient.subscribeOn(Schedulers.fromExecutor(executor));

To

webClient.publishOn(Schedulers.fromExecutor(executor));

from reactor-netty.

@KrishnakanthYachareni I cannot comment on the change from subscribeOn/publishOn without having a reproducible example. That said should I close this, if the above is working for you?

from reactor-netty.

@violetagg yes please, you can close it.

from reactor-netty.