I know that a read and a connection timeouts can be configured in reactor-netty HttpClient, like:
public WebClient xsdWebClient() {
HttpClient httpClient = createHttpClient(config.getConnectionTimeout(), config.getReadTimeout());
return WebClient.builder()
.clientConnector(new ReactorClientHttpConnector(httpClient.followRedirect(true)))
.baseUrl(config.getHost())
.build();
}
private static HttpClient createHttpClient(int connectionTimeout, int readTimeout) {
return HttpClient.create()
.option(
ChannelOption.CONNECT_TIMEOUT_MILLIS,
(int) TimeUnit.SECONDS.toMillis(connectionTimeout))
.doOnConnected(c -> c.addHandlerLast(new ReadTimeoutHandler(readTimeout)));
}
But what are the default read and connection timeouts for reactor-netty HttpClient?
I would recommend you to use the response timeout configuration provided by Reactor Netty instead of ReadTimeoutHandler. You can configure the response timeout either globally on HttpClient level for all requests or per request.
The default values for various timeouts provided by Reactor Netty you can find in the reference documentation.
Referring to answer given by one of the devs of reactor-netty, the read default time is 10 seconds.
Related
I use a httpClient of apache in java to call REST APIs which is configured as below in a static block of class, what I want is to change the connection timeout per request. is it possible? how?
static {
PoolingHttpClientConnectionManager pool = new PoolingHttpClientConnectionManager();
pool.setMaxTotal(ChatSettings.HTTP_CLIENT_THREAD_POOL_SIZE);
httpClient = HttpClients
.custom()
.setConnectionManager(pool)
.setDefaultRequestConfig(
RequestConfig
.custom()
.setConnectTimeout(DEFAULT_HTTP_TIMEOUT)
.setSocketTimeout(DEFAULT_HTTP_TIMEOUT)
.build()
)
.build();
}
One thing you can implement for sure is to follow this example: https://github.com/apache/httpcomponents-client/blob/5.1.x/httpclient5/src/test/java/org/apache/hc/client5/http/examples/ClientAbortMethod.java
It demonstrates how a client can fire a request, but eventually change it's mind and tell the server it no longer wants to receive the response.
Combine this with a 'watchdog' thread that fires after some time/when the time runs out.
I have to call different urls with different connection timeout with webclient. But i found that we can set connect timeout globally only and not per resquest .. what to do please to set this timeout on a request without creating a new weblient each time.
You need to instantiate one webclient per url. For each webclient, you can set the connection timeout :
HttpClient httpClient = HttpClient.create()
.option(ChannelOption.CONNECT_TIMEOUT_MILLIS, 10000);
WebClient client = WebClient.builder()
.baseUrl("http://yourendpoint:8080")
.clientConnector(new ReactorClientHttpConnector(httpClient))
.build();
Then reuse this webclient for all the calls to the configured url.
I have 2 microservices. First one is simply echo service which returns after 3 seconds delay. Second one calls first over spring webclient. When I simulate 501 concurrent users to request second microservice, first 500 users get response in 3 seconds and the last one gets response in 6 seconds. Means spring webclient supports only 500 concurrent connections by default. Corresponding value is configurable in apache client. But I can't find how to configure it in spring webclient.
Similar situation with WebSocket, it supports only 256 concurrent connections. How to configure it?
Tests are done with jMeter. If I test first microservice directly - there is no limit.
public TestController(WebClient.Builder webClientBuilder) {
this.webClient = webClientBuilder.baseUrl("http://localhost:8081").build();
}
#GetMapping("/test1")
public Mono<String> test1() {
return webClient
.get().uri("/test/HTTP")
.retrieve().bodyToMono(String.class);
}
Thanks Mark, solution from you link works.
Either this:
HttpClient httpClient = HttpClient.create(ConnectionProvider
.elastic("myConnectionProvider"));
this.webClient = webClientBuilder.clientConnector(new ReactorClientHttpConnector(httpClient))
.baseUrl("http://localhost:8081").build();
Or that:
HttpClient httpClient = HttpClient.create(ConnectionProvider
.fixed("myConnectionProvider", 1000));
this.webClient = webClientBuilder.clientConnector(new ReactorClientHttpConnector(httpClient))
.baseUrl("http://localhost:8081").build();
Spring docs says it is required to configure http client for WebClient manually to set timeouts: https://docs.spring.io/spring/docs/current/spring-framework-reference/web-reactive.html#webflux-client-builder-reactor-timeout.
But since WebClient returns reactive Mono, it's possible (api-wise) to apply .timeout method.
Does it have the same effect?
Moreover, when one uses .timeout method, Reactor's TimeoutException is expected. Will the same error appear in the stream if configuration is done manually i.e. will doOnError(TimeoutException.class, ...) work?
My findings
Setting a timeout in a http client specific way will lead to http client specific exception i.e. WebClient doesn't wrap exceptions:
#Test
void test() {
var host = "localhost";
var endpoint = "/test";
var port = 8089;
var timeout = Duration.ofSeconds(3);
WireMockServer wireMockServer = new WireMockServer(wireMockConfig().port(8089));
wireMockServer.start();
WireMock.configureFor(host, wireMockServer.port());
WireMock.stubFor(get(urlEqualTo(endpoint))
.willReturn(aResponse().withFixedDelay((int) timeout.toMillis())));
HttpClient httpClient = HttpClient.create()
.tcpConfiguration(client ->
client.doOnConnected(conn -> conn
.addHandlerLast(new ReadTimeoutHandler((int) (timeout.toSeconds() / 2)))
.addHandlerLast(new WriteTimeoutHandler((int) (timeout.toSeconds() / 2)))));
WebClient webClient = WebClient.builder()
.baseUrl(format("http://%s:%d", host, port))
.clientConnector(new ReactorClientHttpConnector(httpClient)).build();
webClient.get().uri(endpoint).retrieve().bodyToMono(Recommendation.class).block();
}
This will lead to io.netty.handler.timeout.ReadTimeoutException.
.timeout(timeout.dividedBy(2)).block() leads to regular TimeoutException (java.util.concurrent) but it's still an open question whether a web client takes care about connections afterwards (probably not).
My solution is to use http client specific configuration to ensure native and correct way to utilize connections while adding new handler that wraps http client related exception into more generic ones (or java.util.concurrent.TimeoutException) so that WebClient clients won't depend on provider exceptions.
This problem has blocked our whole team half a day!
We use apache httpclient 4.3.x to post and get data from an storage server which provides http api. In order to improve performance, we used PoolingHttpClientConnectionManager:
public HttpClient createHttpClient() {
Registry registry = RegistryBuilder.create()....build();
PoolingHttpClientConnectionManager connectionManager = new PoolingHttpClientConnectionManager(registry);
connectionManager.setMaxTotal(50);
connectionManager.setDefaultMaxPerRoute(50);
CloseableHttpClient httpClient = HttpClients.custom()
.setConnectionManager(connectionManager)
.build();
return httpClient;
}
Then we hold an instance of the httpClient in our program, reuse it with every http request:
Global httpClient:
HttpClient httpClient = createHttpClient();
Post some data:
HttpPost httpPut = new HttpPost("...");
HttpResponse response = httpClient.execute(httpPut);
// Notice we get the response content here!
String content = EntityUtils.toString(response.getEntity());
System.out.println(content);
httpPut.releaseConnection();
response.close();
Then get:
HttpGet httpGet = new HttpGet("...");
// Blocked at this line !!!!
HttpResponse response = httpClient.execute(httpGet);
String content = EntityUtils.toString(response.getEntity());
System.out.println(content);
httpPut.releaseConnection();
response.close();
Please notice the line: // Blocked at this line !!!!
The program has blocked at that line and never go to next line. In debugging mode, I can see it has been blocked at:
SocketInputStream.socketRead0()
I've searched for a lot of questions and documents, but no lucky.
My colleage just fix it by setting NoConnectionReuseStrategy.INSTANCE:
HttpClients.custom()
.setConnectionManager(connectionManager)
// Following line fixed the problem, but why?
.setConnectionReuseStrategy(NoConnectionReuseStrategy.INSTANCE)
.build();
Now it doens't blocked, but why?
What does "reuse connection" mean? And is there performance issue by using NoConnectionReuseStrategy?
Thank you, guys~
I tried to reproduce the blocking http-get (also as an exercise for myself) but even without closing responses I could not get it to block. The ONLY time I managed to make the http-get block is by doing a response.getEntity().getContent() without reading from the returned InputStream and without closing the returned InputStream.
For my tests I used Tomcat 7.0.47 with two very simple servlets (one responding "OK" to a get, the other echoing a post) as a server. The client started 50 threads with each thread performing 30 alternating http-get and http-post request (total of 1500 requests). The client did not use the RegistryBuilder, instead the default one is used (created by the PoolingHttpClientConnectionManager itself).
About the NoConnectionReuseStrategy: by default (HttpClient created with HttpClients.createDefault(), I used org.apache.httpcomponents:httpclient:4.3.1) a connection pool is used with a maximum of 2 connections to 1 server. E.g. even if 5 threads are doing all kinds of requests at the same time to 1 server, the connection pool opens only 2 connections, re-uses them for all requests and ensures that 1 connection is used by 1 thread at any given time. This can have a very positive impact on client performance and significantly reduces load on the server. The only thing you must make sure is to call response.close() in a finally-block (this ensures the connection is returned to the connection pool). By using the NoConnectionReuseStrategy you basically disable the connection pool: for each request a new connection will be created. I recommend you enable debug-logging for category org.apache.http.impl.conn.PoolingHttpClientConnectionManager, it is very informative.
A note about httpPut.releaseConnection(): this does not actually release a connection, it only ensures that you can re-use the "httpPut" object in a next request (see the apidocs, follow the shown link). Also note that in your code for the "httpGet", you call releaseConnection() on "httpPut" instead of "httpGet".
Ran into this problem just a while back. In case someone else comes across this problem, this post might be useful.
I am using a Java Servlet to service my requests. When I wrote to the response stream using the PrintWriter instance my client blocked. Tried writing to the OutputStream directlyresponse.getOutputStream.write("myresponse") and it worked.