Hello, I am using Apache 2.2 mpm prefork and am having a problem with requests spawning child processes. I am using Firefox on Ubuntu 11.04.
Here's the situation: Firefox allows 6 persistent connections. KeepAlive in on with a Timeout of 5 s 6 requests are made the 7th request is blocked until the 6 requests are done the 7th request waits for 1 s more than it should... I believe Apache is spawning a child process My issue is that I think that once the 6 requests are done, there should be 6 child processes idle (because of KeepAlive). I would think that the 7th request should be served by one of these 6 now idle child processes. Instead it seems like the 6 initial processes idle and the 7th must wait for a new child process to spawn. Also, shouldn't the KeepAlive max count in the headers decrease for every request served by the Apache child process? It seems that many requests are always at the initial max count which I think means that the persistent connection is not being reused. It basically gets used once and then is unavailable. For a pic of the problem: http://s24.postimg.org/qdazev9rp/website_waterfall_local.png I don't have any of what I believe the the 1s spawning delay when I set KeepAlive Off. If I create enough child processes to handle every request in parallel, I do not have any problems however any pageload done within the KeepAlive timeout limit will require more child processes... as if the originals are unresponsive. Maybe I don't understand KeepAlive but as a test if I set my MaxClients to 1, subsequent requests are delayed by the KeepAlive Timeout. I would think that the child process would be alive to handle requests before the time out but instead it seems like the child process handles a request and then will not handle another one until the keepalive timeout elapses! I'm very confused about why KeepAlive is seeming to cause problems. Thought it would make things more efficient by cutting down connection overhead on subsequent requests. I believe my content-length headers are correct but heard they could be a reason for apache processes to become unusable while they wait for data that doesn't come. Thanks for any insights you can offer! Marc
