[Libwebsockets] Using websockets as a pure buffer manipulation engine.

Andy Green andy at warmcat.com
Fri Nov 18 03:10:13 CET 2016


On Thu, 2016-11-17 at 09:50 -0500, Alan Conway wrote:
> On Thu, 2016-11-17 at 22:04 +0800, Andy Green wrote:
> >  
> > > I'm working on qpid.apache.org/proton, an AMQP messaging library.
> > > It
> > > currently has a single-threaded, event-based "reactor" API using
> > > sockets & poll. It also has a low-level "protocol driver" that
> > > deals
> > > strictly with byte buffers in memory - no IO assumptions.
> > 
> > I understand the kind of abstraction you mean, but I think it's not
> > enough for a practical system without also considering flow control
> > (equivalent to POLLIN/POLLOUT)
> 
> Ohhhh yes indeedy. If I had a dime for every distributed system 

Great.

> screwup
> I've fixed (or caused) due to ignoring flow control, I'd have a lot
> of
> dimes :)
> 
> Here's how it works for proton: 2 variable sized but bounded buffers
> for read and write (they are actually variable sized windows on a
> fixed
> buffer). Normal operation is something like:
> 
> - non-blocking or async read of max read-buffer-size bytes
> - parse & handle events which puts stuff in the write-buffer
> - non-blocking or async partial write of write-buffer till it's empty
> 
> EPOLLIN == read-buffer-size > 0
> EPOLLOUT == write-buffer-size > 0

Right... what's implied by that are elective events lws can receive on
the connection triggered by changes in the window state.

This brings a nice additional feature actually it's possible to know
how much can be written with confidence.  There's an exported api to
expose that already existing for the http/2 tx window,
lws_get_peer_write_allowance().  So the user code can tune what it
tries to write to match what may be written.

> If the write buffer fills up, the read-buffer is disabled till there
> is
> some write space.

Lws offers immediate quench of rx activity if you ask for rx to be flow
controlled... this is done both by removing POLLIN (or disabling rx
event if libuv etc), and by internally making any buffered RX wait.

Basically over and above "just buffers" you must

 - track connection lifecycle through lws (so it can do timeouts,
manage his own wsi object for the connection etc).  This means you must
create a logical "fd" to represent the connection, which lws allows to
be typedef'd to an opaque struct.  But he will pass it back on read /
write / close so you can act on the correct connection context etc.

 - allow lws to close connections

 - create a serialized event system around Proton buffers that can call
into lws on rx, or space in the tx buffer appearing.  The event enables
are controlled by lws using POLLIN / POLLOUT on lws side, but you can
translate that into whatever on the proton side.

Basically you must recreate your own "fd" and the basic event loop
actions on proton side.

> There's definitely room for improvement on the buffering scheme but
> it
> does handle the flow control problem pretty well.
> 
> > > The FD notion also isn't very helpful for fast native windows
> > > IOCP
> > > - we
> > > ended up having to fake a poll()-like loop which is a terribly
> > > inefficient way to use IOCP.
> > 
> > Dunno what iocp is, but --->
> 
> Windows IO Completion Ports: a multi-threaded, proactive IO
> framework.
> Most apps deal with it by making it look like a single-threaded
> poller
> (e.g. libuv) which gives acceptable single-threaded performance but
> does not scale as well on multi-cores as proper multi-threaded use.
> 
> > > So I'd love to see a byte-buffer based websocket "driver" that
> > > could be
> > > used independently of the libwebsocket loop and has no
> > > assumptions
> > 
> > The 'flow control' type concerns are not theoretical... 
> 
> +1000, preaching to the choir. Beware unbounded data structures,
> there
> are no computers with unbounded memory ;)
> 
> > Sure, it should not be that hard since the recv() and send() bits
> > are
> > already isolated.  But you will need a flow control method even to
> > duplicate the mirror demo reliably.
> 
> Yep. What do you think of the scheme described above? There are other
> ways to skin that cat. We know proton's buffer scheme has some
> limitations (control of allocation, may force copies from externally-
> owned buffers) but it has the advantage of being simple. If you want
> to
> come up with something more sophisticated I'd enjoy that discussion.

There are a few things above the buffering that need getting sorted
before it will work.  So probably just wiring up what you got with the
buffering until it works would be the first step.

LWS has a concept of platform-specific code, but that doesn't fit here
since you might want this on different platforms.  The closest place to
put it is the concept of the selectable event library, eg, default
poll() or use libuv, stuff like enable / disable POLLIN/OUT go through
that, so you can cleanly map that to whatever is firing the events.

So far I think the steps would look like this

1) define a new cmake define like LWS_PROTON to allow you to force
lws_sockfd_type to be a pointer to the opaque per-connection struct in
"proton".  Add a context creation time server option flag to select to
use proton.c, same way there is one for libuv, libev etc.

2) defeat listen sockets etc that are handled out of scope now

3) call lws_adopt_socket_vhost(vhost, sockfd) when there is a new
connection at proton (it is already exported).

4) clone libuv.c into proton.c or so and select it from LWS_PROTON, fix
up the api names, visit any preprocessor use of LWS_USE_LIBUV looking
for what needs doing.

5) Have proton.c translate POLLIN / POLLOUT enable / disable to calls
into your proton code (libuv.c does this kind of translation already)

6) Provide an exported entry point in proton.c for asynchronous events
to arrive at (these must be serialized by the caller...).  Again
libuv.c already has this kind of code for the necessary functions since
that is also how libuv works.

7) Look in output.c about lws_ssl_capable_read_no_ssl() for recv
overload and lws_ssl_capable_write_no_ssl() for send() overload, divert
these to proton

8) Adapt close() usage in libwebsockets.c to do it via proton.c and
call into proton there to get it done

If there is some heavy earthmoving inside lws that will make this
better or easier I can consider to help with that.

-Andy

> 



More information about the Libwebsockets mailing list