In epoll usage, it is usually like the following:
struct epoll_event ev,events[20];
epfd=epoll_create(256);
。。。
nfds=epoll_wait(epfd,events,40,500);
Some articles are saying that the maxevents
parameter in epoll_wait
(namely the 40
in epoll_wait(epfd,events,40,500);
) should not exceed the size parameter in epoll_create
(namely the 256
).
I think the maxevents
parameter should not exceed 20
in ev, events[20]
, because events can only be registered to 20 events elements; otherwise, if there are 40 sockets which are active, then what will happen?
BTW, if I register more than 20 sockets and there are more than 20 active events(sockets), but the events array events[20]
has only 20 events, what will happen?
At any single call of
epoll_wait
you'll at most receive as many events as you have room for, but of course events don't get lost if there are more than that queued up -- you'll simply get them at a later call. Since you'll be callingepoll_wait
in a loop anyway, that shouldn't be an issue at all.The one interesting consideration I can think of is when you have multiple threads read from the same epoll-fd concurrently. In that case the size of your event array determines how many events get handled by a single thread (i.e. a smaller number might give you greater parallelism).