netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Stephen Hemminger <stephen@networkplumber.org>
To: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Cc: David Ahern <dsahern@gmail.com>, Chris Mi <chrism@mellanox.com>,
	netdev@vger.kernel.org, gerlitz.or@gmail.com
Subject: Re: [patch iproute2 v3 3/4] tc: Add -bs option to batch mode
Date: Wed, 27 Dec 2017 13:40:24 -0800	[thread overview]
Message-ID: <20171227134024.0706d33f@xeon-e3> (raw)
In-Reply-To: <20171227203929.GB22042@localhost.localdomain>

On Wed, 27 Dec 2017 18:39:29 -0200
Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> wrote:

> > > +			send = false;
> > > +		else
> > > +			send = true;
> > > +
> > > +		ret = do_cmd(largc, largv, batch_size, msg_iov_index++, send);  
> > 
> > What happens if tc commands are interlaced in the file -- qdisc add,
> > class add, filter add, then a delete, show, exec, etc.? Right now each
> > command is handled one at a time so an add followed by a delete will
> > work. Your proposed batching loop won't work for this case as some
> > commands are executed when that line is reached and others are batched
> > for later send. Not all of the tc commands need to be batched in a
> > single message so perhaps those commands cause the queue to be flushed
> > (ie, message sent), then that command is executed and you start the
> > batching over.
> > 
> > Further, I really think the batching can be done without the global
> > variables and without the command handlers knowing it is batching or
> > part of an iov. e.g., in the case of batching try having the commands
> > malloc the request buffer and return the pointer back to this loop in
> > which case this loop calls rtnl_talk_msg and frees the buffers.  
> 
> Sounds like the batching is being done at the wrong level. If it was
> done by rtnl_talk(), it should be easier.
> We can keep rtnl_talk() for previous users and make rtnl_talk_msg() do
> the batching, mostly independent of which kind of msg it it.
> 
> As you need to inform it that it was the last entry, that may be
> detected with feof(stdin). Just add a 'bool flush' parameter to it.
>    rtnl_talk_msg(...., flush=feof(stdin));
> 
> Next step then would be to add a memory manager layer to it, so
> libnetlink wouldn't need to copy the messages but recycle pointers:
>   rtnl_get_msgbuf(): returns a buffer that one can use to fill in the
>     msg and use with rtnl_talk_msg()
>   and the free is done by libnetlink itself when the message is
>   finally sent, so no need to keep track of what one needs to free or
>   can reuse.


What about using sendmmsg instead?
That woudl allow sending multiple messages in one syscall.

  reply	other threads:[~2017-12-27 21:40 UTC|newest]

Thread overview: 16+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2017-12-25  8:46 [patch iproute2 v3 0/4] tc: Add -bs option to batch mode Chris Mi
2017-12-25  8:46 ` [patch iproute2 v3 1/4] lib/libnetlink: Add a function rtnl_talk_msg Chris Mi
2017-12-27 14:33   ` David Ahern
2018-01-02 13:59     ` Chris Mi
2017-12-25  8:46 ` [patch iproute2 v3 2/4] utils: Add a function setcmdlinetotal Chris Mi
2017-12-27 14:34   ` David Ahern
2018-01-02 14:03     ` Chris Mi
2017-12-25  8:46 ` [patch iproute2 v3 3/4] tc: Add -bs option to batch mode Chris Mi
2017-12-27 15:39   ` David Ahern
2017-12-27 20:39     ` Marcelo Ricardo Leitner
2017-12-27 21:40       ` Stephen Hemminger [this message]
2017-12-27 22:06         ` Marcelo Ricardo Leitner
2018-01-02 14:17     ` Chris Mi
2017-12-27 19:56   ` Marcelo Ricardo Leitner
2018-01-02 14:19     ` Chris Mi
2017-12-25  8:46 ` [patch iproute2 v3 4/4] man: Add -bs option to tc manpage Chris Mi

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20171227134024.0706d33f@xeon-e3 \
    --to=stephen@networkplumber.org \
    --cc=chrism@mellanox.com \
    --cc=dsahern@gmail.com \
    --cc=gerlitz.or@gmail.com \
    --cc=marcelo.leitner@gmail.com \
    --cc=netdev@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).