The read timeout isn't firing for slow replying endpoints

[2polev]

I was investigating the way how you can configure different timeouts for WebClient. I was pretty sure that setting response timeout like this:

        var finalProperties = getFinalProperties(name, properties);
                var connectionProvider = ConnectionProvider.builder(name)
                        .maxConnections(finalProperties.getMaxConnections())
                        .maxLifeTime(finalProperties.getMaxConnectionLifeTime())
                        .pendingAcquireMaxCount(-1) // no limit
                        .pendingAcquireTimeout(finalProperties.getAcquireTimeout())
                        .metrics(finalProperties.isMetrics())
                        .build();

    var client = HttpClient.create(connectionProvider)
            .compress(true)
            .wiretap(true)
            .option(ChannelOption.CONNECT_TIMEOUT_MILLIS, Long.valueOf(finalProperties.getConnectionTimeout().toMillis()).intValue())
            .option(ChannelOption.SO_KEEPALIVE, true)
            .responseTimeout(finalProperties.getResponseTimeout());

    if (finalProperties.isInsecure()) {
        client = client.secure(t -> t.sslContext(createInsecureSslContext()));
    }

    var connector = new ReactorClientHttpConnector(client);

    var webClient = WebClient.builder()
            .defaultHeader(HttpHeaders.CONTENT_TYPE, MediaType.APPLICATION_JSON_VALUE)
            .clientConnector(connector)
            .filter(new LogbookExchangeFilterFunction(logbook));

    if (finalProperties.isMetrics()) {
        webClient.observationRegistry(observationRegistry);
    }

I would have total time which takes from getting first bytes from server until the last one so total time of getting response back. I prepared some integration test for that using toxiproxy and wiremock:

final ToxiproxyClient toxiproxyClient = new ToxiproxyClient(TOXI_PROXY.getHost(), TOXI_PROXY.getControlPort()); final Proxy proxy = toxiproxyClient.createProxy("wiremock", "0.0.0.0:8666", "wiremock:" + "8080"); proxy.toxics() .slicer("slicer", ToxicDirection.DOWNSTREAM, 1, Duration.ofMillis(100).toMillis() * 1000);

In my test I call endpoint on Wiremock via Toxiproxy. Toxproxy is configured to reply with new bytes every 100ms. My configured timeout is 17sec. Here are wiretap logs:

2025-07-18 11:22:49.345 [main] TRACE [] o.s.w.r.f.client.ExchangeFunctions - [37e64e37] HTTP GET http://localhost:64175/albums/1/photos, headers={masked} 2025-07-18 11:22:49.650 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [d16433fd] REGISTERED 2025-07-18 11:22:49.665 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [d16433fd] CONNECT: localhost/127.0.0.1:64175 2025-07-18 11:22:49.667 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [d16433fd, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] ACTIVE 2025-07-18 11:22:49.672 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientConnect - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] Handler is being applied: {uri=http://localhost:64175/albums/1/photos, method=GET} 2025-07-18 11:22:49.675 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientOperations - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] No sendHeaders() called before complete, sending zero-length header 2025-07-18 11:22:49.680 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] WRITE: 156B +-------------------------------------------------+ | 0 1 2 3 4 5 6 7 8 9 a b c d e f | +--------+-------------------------------------------------+----------------+ |00000000| 47 45 54 20 2f 61 6c 62 75 6d 73 2f 31 2f 70 68 |GET /albums/1/ph| |00000010| 6f 74 6f 73 20 48 54 54 50 2f 31 2e 31 0d 0a 61 |otos HTTP/1.1..a| |00000020| 63 63 65 70 74 2d 65 6e 63 6f 64 69 6e 67 3a 20 |ccept-encoding: | |00000030| 67 7a 69 70 0d 0a 75 73 65 72 2d 61 67 65 6e 74 |gzip..user-agent| |00000040| 3a 20 52 65 61 63 74 6f 72 4e 65 74 74 79 2f 31 |: ReactorNetty/1| |00000050| 2e 32 2e 35 0d 0a 68 6f 73 74 3a 20 6c 6f 63 61 |.2.5..host: loca| |00000060| 6c 68 6f 73 74 3a 36 34 31 37 35 0d 0a 61 63 63 |lhost:64175..acc| |00000070| 65 70 74 3a 20 2a 2f 2a 0d 0a 43 6f 6e 74 65 6e |ept: */*..Conten| |00000080| 74 2d 54 79 70 65 3a 20 61 70 70 6c 69 63 61 74 |t-Type: applicat| |00000090| 69 6f 6e 2f 6a 73 6f 6e 0d 0a 0d 0a |ion/json.... | +--------+-------------------------------------------------+----------------+ 2025-07-18 11:22:49.681 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] FLUSH 2025-07-18 11:23:05.749 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientOperations - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] Received response (auto-read:false) : RESPONSE(decodeResult: success, version: HTTP/1.1) 2025-07-18 11:23:25.349 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientOperations - [d16433fd-1, L:/127.0.0.1:64179 - R:localhost/127.0.0.1:64175] Received last HTTP packet

What I've observed is:

• The read timeout isn't firing, even though it should trigger after 17 seconds.
• The read timeout is firing when timeout is set to less than 16 seconds.
• The read timeout is currently being measured from when the request is flushed until the log line 'Received response (auto-read:false)' appears. However, it should be measured until the 'Received last HTTP packet' log entry.
• There is no option to configure real readtimeout which refers to the maximum duration allowed between each network-level read operation.
• ReadTimeoutHandler does not allow to set values less than seconds

I am using reactor-netty-http:1.2.5

[violetagg]

@2polev HttpClient#responseTimeout is specifies the maximum duration allowed between each network-level read operation while reading a given response. It does not measure neither from start sending the request to the end of the response nor from start receiving the response to the end of the response. Is that what you expected or something else? It is more readTimeout but unfortunately we cannot change the name of the API because of backwards compatibility.

Here is the javadoc https://projectreactor.io/docs/netty/1.2.6/api/reactor/netty/http/client/HttpClient.html#responseTimeout-java.time.Duration-

[2polev]

As I understand "duration between read operation" is time between subsequent reads from socket. If server sends new bytes every 100ms so ReadTimeoutException should occur if a read operation does not start within X seconds after the request is sent or if subsequent data chunks are not received within the configured interval.

So HttpClient#responseTimeout is probably not what you are refering to. It looks like something that measures time until http client reseices status line and headers and not a whole response.

In my case wiretap logs are as follows (new read every 100ms):

2025-07-18 17:59:06.556 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ: 1B +-------------------------------------------------+ | 0 1 2 3 4 5 6 7 8 9 a b c d e f | +--------+-------------------------------------------------+----------------+ |00000000| 0a |. | +--------+-------------------------------------------------+----------------+ 2025-07-18 17:59:06.556 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ COMPLETE 2025-07-18 17:59:06.657 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ: 1B +-------------------------------------------------+ | 0 1 2 3 4 5 6 7 8 9 a b c d e f | +--------+-------------------------------------------------+----------------+ |00000000| 43 |C | +--------+-------------------------------------------------+----------------+ 2025-07-18 17:59:06.657 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ COMPLETE 2025-07-18 17:59:06.759 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ: 1B +-------------------------------------------------+ | 0 1 2 3 4 5 6 7 8 9 a b c d e f | +--------+-------------------------------------------------+----------------+ |00000000| 6f |o | +--------+-------------------------------------------------+----------------+ 2025-07-18 17:59:06.759 [reactor-http-nio-2] DEBUG [] r.netty.http.client.HttpClient - [5d54ad2b-1, L:/127.0.0.1:53846 - R:localhost/127.0.0.1:53801] READ COMPLETE .... ....
and it keeps reading bytes until this is displayed:

2025-07-18 18:08:45.250 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientOperations - [283241ca-1, L:/127.0.0.1:54422 - R:localhost/127.0.0.1:54409] Received response (auto-read:false) : RESPONSE(decodeResult: success, version: HTTP/1.1) HTTP/1.1 200 OK Content-Type: <filtered> Matched-Stub-Id: <filtered> Transfer-Encoding: <filtered> 2025-07-18 18:08:45.255 [reactor-http-nio-2] TRACE [] o.s.w.r.f.client.ExchangeFunctions - [7bc8ffbc] [283241ca-1, L:/127.0.0.1:54422 - R:localhost/127.0.0.1:54409] Response 200 OK, headers={masked}
and then it keeps reading body (no time is measured since this moment and no timeout triggered) until this is displayed:

2025-07-18 18:09:04.880 [reactor-http-nio-2] DEBUG [] r.n.http.client.HttpClientOperations - [283241ca-1, L:/127.0.0.1:54422 - R:localhost/127.0.0.1:54409] Received last HTTP packet

It seems there is a difference in how Reactor Netty HttpClient and Apache HttpClient handle read timeouts. I have noticed that with Apache HttpClient, a 90ms read timeout reliably triggers a ReadTimeoutException, while 150ms allows the operation to complete without error. This suggests that Apache HttpClient's "read timeout" behaves as a true timeout for the read operation itself. Please find logs from my test.

logs.txt