That's why I asked about user code example. I can't test what we trying to fix. Did you test it or only review? Seems that MHD read data from network after you suspended connection, which cause overflow buffer overflow and error response. You issue should be fixed by afb221c32aeeb2731e3b5362cc4ff8949d34edea.
Alternatively - may be you don't consume data in callback, which cause buffer overflow. One more reason why I asked for user code. Without your code it's only blindly guessing. Disconnect by timeout for suspended connections should be fixed, so you can stop playing with timeout values. You mentioned that you had some problems in master with epoll and timeouts. Is it still actual? -- Best Wishes, Evgeny Grin On 15.03.2017 23:26, Vitaliy T wrote: > On 15 March 2017 at 21:52, Evgeny Grin <k...@yandex.ru> wrote: >> Actually we should call MHD_connection_update_event_loop_info() in any >> case after processing of data. >> I already applied some patches to master. Check whether it helps. > > About commit ......bcfacf404 > > Seems same for me. Same place, same issue with loop inside > MHD_connection_handle_idle (). Something changes the > connection->suspended value in the function MHD_connection_handle_idle > (). The patch is still actual :) >
