Re: [Libevent-users] filter bufferevents and timeout

[Azat Khuzhin <a3at.mail@xxxxxxxxx>]

On Tue, Mar 13, 2018 at 7:42 PM, Bruno CARLUS <b.carlus@xxxxxxxxxxxxx> wrote:
> Hello,
>
> I just identified a little problem in my code:
> I use a filter bufferevent wrapped on a socket bufferevent, for example:
>
> bev = bufferevent_socket_new(base, listener, BEV_OPT_CLOSE_ON_FREE);
> struct timeval timeout_read = {5, 0};
> bufferevent_set_timeouts(bev, &timeout_read, NULL);
> bufferevent_setwatermark(bev, EV_READ, 1500, 0);
>
> bev_filter = bufferevent_filter_new(bev, readandfiltercb, NULL, BEV_OPT_CLOSE_ON_FREE, NULL, NULL);
> bufferevent_setwatermark(bev_filter, EV_READ, 3000, 0);
> bufferevent_setcb(bev_filter, freadcb, fwritecb, eventcb, NULL);
>
> I'm sending 1000 bytes packets on this socket.
> readandfiltercb is called every 2 packets: it filters data and copy it to the bev_filter input buffer.
> freadcb is then called ok.
>
> But when udp stream is over, it remains one 1000 bytes packet in the buffer.
> eventcb is called because of the timeout: then it flushes bufferevent
> readandfiltercb is called ok: it filters and copies last bytes to bev_filter input buffer.
> But then freadcb is not called at all.
>
> Is it normal?
> How to force bypass the watermark and flush bev_filter input buffer as well?

Hello,

I don't think so, my guess is that during flush it doesn't call readcb
(in be_filter_flush() there is no bufferevent_trigger_nolock_() like
in be_filter_read_nolock_()).
I will write a test today and will come back to you later, in the evening.

> Last point: Is the underlying fd of a bufferevent can be a mmap file?
> I could not find reference to the use of EVBUFFER_SENDFILE in the doc...

How you will poll mmaped fd?
Can you explain this a little bit more? Maybe I'm missing something.

-- 
  Azat.
***********************************************************************
To unsubscribe, send an e-mail to majordomo@xxxxxxxxxxxxx with
unsubscribe libevent-users    in the body.