> > On 11/05/18 18:21, Максим Гумеров wrote: > > Hi! I am running Tomcat 8 as part of Spring Boot application. As usual, > > acceptCount defaults to 100 and maxThreads to 200. According to > > https://tomcat.apache.org/tomcat-8.5-doc/config/http.html#NIO_specific_ > configuration > > I am expecting Tomcat to accept 200 "long" requests (each taking say 20 > > seconds to complete), then accept 100 more connections but not start > > reading from them, and then reject any connection requests beyond that. > > That's why this is the expected behavior: acceptCount = "The maximum > queue > > length for incoming connection requests when all possible request > > processing threads are in use. Any requests received when the queue is > full > > will be refused." So, when first 200 requests are all taken by 200 > > processing threads, next 100 will go to the queue and subsequent will be > > ignored. At least that's how it reads to me. > > > > But what I actually see on my Mac is, tomcat successfully accepts a > little > > short of 1000 such connections (I just didn't try more connections) made > > within 5 seconds. It processes no more than 200 requests at a time, but > > does not reject connections after 300th one, as one would expect. > > > > What's the explanation for that? Can the quoted documentation be > imprecise > > or outdated? Or am I missing something in it? > > acceptCount for NIO is passed in to ServerSocket.bind(SocketAddress,int) > as the 2nd parameter (backlog). The Javadoc for that method says: > > <quote> > The backlog argument is the requested maximum number of pending > connections on the socket. Its exact semantics are implementation > specific. In particular, an implementation may impose a maximum length > or may choose to ignore the parameter altogether. > </quote> > > It looks like it is being ignored in your JVM. > > Mark >
Thank you, this looks like one possible explanation. Maybe someone else will show me some other. But, considering what you say, isn't that a documentation deficiency? They shouldn't write "acceptCount is the length of connection queue" if it's just a hint. Or, they could just add a note that actual queue length may be chosen by implementation instead. Also, they shouldn't write " up to the configured maximum (the value of the acceptCount attribute)", better formulation is "up to connection queue size (based on acceptCount attribute)". What's written there now simply contradicts the implementation, right? Maksim
