[vlc-devel] [PATCH] VLC unable to play HLS live stream

[Ilkka Ollakka]

Hi,

I think it would be better to implement this with signal wait with
timeout instead of polling. So download thread would signal that new
stuff is ready.

On Thu, Aug 15, 2013 at 06:52:09AM -0400, Avishay Spitzer wrote:
> From 768e199c6ac6d539f1a00606a26a905a5f1b1886 Mon Sep 17 00:00:00 2001
> From: Avishay Spitzer <savishay at gmail.com>
> Date: Thu, 15 Aug 2013 06:41:35 -0400
> Subject: [PATCH 2/2] Added a retry mechanism with backoff to the read
>  function of HLS module to handle situations where
>  stream packets may delay due to slow connections or
>  short playlists. In case hls_Read returns 0, the read
>  function will wait until new data arrives.

> ---
>  modules/stream_filter/httplive.c |   48 +++++++++++++++++++++++++++++++++-----
>  1 file changed, 42 insertions(+), 6 deletions(-)

> diff --git a/modules/stream_filter/httplive.c b/modules/stream_filter/httplive.c
> index ed708bf..62ab17b 100644
> --- a/modules/stream_filter/httplive.c
> +++ b/modules/stream_filter/httplive.c
> @@ -2280,16 +2280,52 @@ static int Read(stream_t *s, void *buffer, unsigned int i_read)
>  {
>      stream_sys_t *p_sys = s->p_sys;
>      ssize_t length = 0;
> +    uint32_t retries_counter = 0, aggregate_delay = 0;

>      assert(p_sys->hls_stream);

> -    if (p_sys->b_error)
> -        return 0;
> +    while (length == 0)
> +    {
> +    	// In case an error occurred or the stream was closed return 0
> +        if (p_sys->b_error || !vlc_object_alive(s))
> +            return 0;

> -    /* NOTE: buffer might be NULL if caller wants to skip data */
> -    length = hls_Read(s, (uint8_t*) buffer, i_read);
> -    if (length < 0)
> -        return 0;
> +		/* NOTE: buffer might be NULL if caller wants to skip data */
> +		length = hls_Read(s, (uint8_t*) buffer, i_read);
> +
> +		// An error has occurred in hls_Read
> +		if (length < 0)
> +			return 0;
> +
> +		// No data is available yet so we need to wait until new data will arrive.
> +		// We use a retry mechanism with increasing backoff (up to 100ms per single sleep)
> +		if (length == 0)
> +		{
> +	    	uint32_t delay = 0;
> +
> +			retries_counter++;
> +
> +			if (retries_counter < 10)
> +				delay = 10; // ms
> +			else if (retries_counter < 25)
> +				delay = 25; // ms
> +			else if (retries_counter < 100)
> +				delay = 50; // ms
> +			else
> +				delay = 100; // ms
> +
> +			aggregate_delay += delay;
> +
> +			msleep(delay * 1000);
> +		}
> +    }
> +
> +    if (retries_counter > 0)
> +    {
> +		msg_Info(s, "got data after %u retries (delay of [%u]ms)",
> +				retries_counter,
> +				aggregate_delay);
> +    }

>      p_sys->playback.offset += length;
>      return length;
-- 
Ilkka Ollakka
Is a computer language with goto's totally Wirth-less?
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://mailman.videolan.org/pipermail/vlc-devel/attachments/20130815/0e194fff/attachment.sig>