replace in clixon_event_loop select to poll #589
Conversation
|
@olofhagsand, Good afternoon, I would really like to see this functionality. |
|
Will look shortly |
@olofhagsand Thanks :) |
| AC_ARG_WITH([max-events], | ||
| AS_HELP_STRING([--with-max-events=VALUE], [Set the maximum number of events (default: 1024)]), | ||
| [MAX_EVENTS=$withval], | ||
| [MAX_EVENTS=1024] # Значение по умолчанию, если не указано |
There was a problem hiding this comment.
Please use english (more places)
| clixon_event_loop(clixon_handle h) | ||
| { | ||
| struct event_data *e = NULL; | ||
| struct pollfd fds[MAX_EVENTS]; |
There was a problem hiding this comment.
style: struct event_data *e = NULL; (indentation on the variable "e", not on the struct name "event_data"
| } | ||
| free(e); | ||
| continue; | ||
| } |
There was a problem hiding this comment.
The "continue" statement may cause starvation so that only timeouts run. This can happen for example if a timeout registers itself, and then causes a loop that only serves timeouts, causing a deadlock.
Typically if the new timeout due to overload has already timed out.
This is the way it works in select at least, I am not sure if poll is different.
Removing continue will first serve one timeout, then serve file descriptors.
It is not 100% fair either but at least interleaving timeouts with file input removes some deadlocks.
| continue; | ||
| } | ||
|
|
||
| for (int i = 0; i < nfds; i++) { |
There was a problem hiding this comment.
You also removed the primitive CLICON_SOCK_PRIO algorithm.
It is somewhat primitive but at least avoided the same kind of deadlock as timeouts could cause. That is, if there is a prioritized file, then at most one un-prioritized file is served.
@olofhagsand in continuation of the previous pull request #584
I replaced it in the main loop too, and moved the number of events to the configuration file.