Closed GoogleCodeExporter closed 9 years ago
This could easily be done with a hook in case the master thread exits. This
hook would solve several other issues on the list as well. The user callback is
called with an event MG_EXIT_MASTER and may be used to stop/cancel the
operations of worker threads or other processes (CGI/DB/..).
e.g., at DEBUG_TRACE(("stopping workers")) insert something like
memset(&request_info, 0, sizeof(request_info));
request_info.user_data = ctx->user_data;
if (ctx->user_callback) {
ctx->user_callback(MG_EXIT_MASTER, (struct mg_connection *) 0, &request_info);
}
Original comment by bel2...@gmail.com
on 7 May 2012 at 7:26
I would go with number one but have the parameter take a mg_connection
structure. The mg_is_running function is expected to be called from within a
long running response which should have a connection associated with it.
@bel I would expect to receive the MG_EXIT_MASTER callback only once all the
worker threads have actually stopped. The patch I have in Issue 345 will soon
be updated with master thread entry and exits. Would a MG_STOPPING_MASTER be
needed? Thus MG_ENTER_MASTER is called when stop_flag is set to 0,
MG_STOPPING_MASTER is set when stop_flag is set to 1, and MG_EXIT_MASTER is
called when stop_flag is set to 2.
Original comment by jmucchie...@gmail.com
on 10 May 2012 at 4:19
I use an augmented clone of mongoose and have these two functions exported for
precisely this purpose (the custom code is serving very long running requests,
and the custom code both needs to see when mongoose is shutting down
(mg_get_stop_flag()) and signal the mongoose to shut down (mg_signal_stop()),
e.g. when an fatal error occurs in the custom code.
Code snippet from mongoose_ex.c:
int mg_get_stop_flag(struct mg_context *ctx)
{
return ctx && ctx->stop_flag;
}
void mg_signal_stop(struct mg_context *ctx)
{
ctx->stop_flag = 1;
}
Original comment by ger.hobbelt
on 12 May 2012 at 5:42
I think a callback at DEBUG_TRACE(("stopping workers")) is required. At this
location it tells asynchronous operations that the server is supposed to shut
down, so these operations/threads could be canceled. If a worker thread is busy
and no one tells it to cancel the operation one will not reach the point of
DEBUG_TRACE(("exiting")).
I don’t know if there is need for a second callback at
DEBUG_TRACE(("exiting")) when all threads actually finished - maybe yes, since
it is an asynchronous shutdown one might need to know when the shutdown starts
and when it is completed.
Original comment by bel2...@gmail.com
on 21 May 2012 at 3:02
Submitted
https://github.com/valenok/mongoose/commit/15a0d819ede70398db936977ddb0ecf6cf32a
195, thanks!
Original comment by valenok
on 16 Aug 2012 at 4:26
Original issue reported on code.google.com by
kai.nick...@gmail.com
on 6 May 2012 at 11:31