Re: [Libevent-users] Abnormal delay on read

[Azat Khuzhin <a3at.mail@xxxxxxxxx>]

On Thu, Sep 04, 2014 at 10:20:59AM +0200, Viallard Anthony wrote:
> On 09/04/2014 09:57 AM, Azat Khuzhin wrote:
> >Unless you have very-big buffers, and if you external functions are
> >fast, this callback must work fast (at the first glance).
> >Could you attach debugger to process when you see slow downs, and post
> >backtrace here (What I'm wonder is it enter the read callback or not)
> >
> 
> Do you want a strace output or libevent debug messages ? The two ? I wonder
> this because I think the quantity of all these messages will be a quite bit
> huge.

Not strace, but:
$ gdb -p
(gdb) thread apply all bt full

> 
> >>
> >>Moreover, after read callback is called, I don't have any remaining data in
> >>the input evbuffer (I don't see the log messages).
> >
> >You mean that at the first line of you read callback:
> >evbuffer_get_length(bufferevent_get_input(bev)) === 0 ?
> >
> 
> Actually, I talk about the evbuffer_get_length(evbuf_in); at the end of the
> function. Sometimes ago, searching why the messages processing were shift, I
> thought maybe the messages were queued in evbuffer as I don't read all data
> in it in read callback like you see. But I didn't see the log message "sock
> %p: remaining %zu byte(s) of data". So, the problem is somewhere else.

And I guess it is correct for you do drain all data from buffer, no?

> 
> >>Writing this email, I'm wondering if the evbuffer_get_length() function
> >>works in all contexts ? I mean, I disable the read callback when the
> >>"socket" is writing for example.
> >
> >AFAIK it must work in all contexts, this is just reading
> >buffer->total_len + locking, but if you disable read event then you will
> >not recieve any new data, and maybe you are expecting it?
> >
> 
> I disable read event when I call talk_socket_write() function actually. In
> this function, I disable read and enable write. I write the data in the
> output evbuffer. I enable read again (without read timeout) and disable
> write in write callback. I think it's the correct way to do when you want
> listening incoming requests, doesn't it ?

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