[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [RFC PATCH V4 0/4] Introduce COLO-compare

From: Zhang Chen
Subject: Re: [Qemu-devel] [RFC PATCH V4 0/4] Introduce COLO-compare
Date: Fri, 17 Jun 2016 10:25:40 +0800
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.8.0

On 05/31/2016 07:06 PM, Jason Wang wrote:

On 2016年05月31日 16:28, Zhang Chen wrote:

On 05/31/2016 02:16 PM, Jason Wang wrote:

On 2016年05月31日 11:54, Zhang Chen wrote:

On 05/30/2016 11:19 AM, Jason Wang wrote:

On 2016年05月25日 20:50, Zhang Chen wrote:
COLO-compare is a part of COLO project. It is used
to compare the network package to help COLO decide
whether to do checkpoint.

the full version in this github:

     - add some comments
     - fix some trace-events
     - fix tcp compare error
     - add rcu_read_lock().
     - fix trace name
     - fix jason's other comments
     - rebase some Dave's branch function
     - colo_compare_connection() change g_queue_push_head() to
     - g_queue_push_tail() match to sorted order.
     - remove QemuMutex conn_list_lock

Looks like conn_list lock is still there. I still prefer to do all thing in the comparing thread. Have you tried Fam's suggestion to use g_main_context_push_thread_default()? If it does not work, does it work simply by replacing all:

g_source_attach(x, NULL);


g_souce_attach(x, g_main_context_get_thread_default());

after call g_main_context_push_thread_default()?


I have tried fam's suggestion it does not work.

so I tried what you suggestion like that:

static void *colo_compare_thread(void *opaque)
    CompareState *s = opaque;
    GSource *source;
    GMainContext *worker_context;

    source = g_source_new(&source_funcs, sizeof(DemoSource));
    worker_context = g_main_context_new ();

    g_source_attach(source, g_main_context_get_thread_default());
    g_source_set_callback(source, NULL, NULL, NULL);
    g_main_context_push_thread_default (worker_context);
+    g_source_unref(source);
qemu_chr_add_handlers(s->chr_pri_in, compare_chr_can_read,
                          compare_pri_chr_in, NULL, s);
    qemu_chr_add_handlers(s->chr_sec_in, compare_chr_can_read,
                          compare_sec_chr_in, NULL, s);
    g_main_context_pop_thread_default (worker_context);

g_source_attach() should be done after g_main_context_push_thread_default()

Like that? (+    g_source_unref(source);)
But compare_pri_chr_in always run in main loop.

Zhang Chen

Not very sure just from the above codes, since I don't see any connection between source and compare_pri_chr_in. The point is to attach the source of compare_pri_chr_in to thread default context instead of global default context I think.

Hi~ jason.

I tried a lot of ways to make it run in compare thread, but it not work.

Because that:
void g_main_context_push_thread_default (GMainContext *context);
Acquires context and sets it as the thread-default context for the current thread.

So, this function g_main_context_push_thread_default() just can set thread-default context, not the global default context. and I can't find a function to set global default context in glib.
The qemu's QIOChannel uses g_source_attach(source, NULL); to attach GSource.
g_source_attach (GSource *source,
                 GMainContext *context);

a GMainContext (if NULL, the default context will be used).

But he not say "the default context" is global default context or thread-default context.
So I read glib codes find that:
g_source_attach (GSource *source,
                 GMainContext *context)
if (!context)
    context = g_main_context_default ();

g_main_context_default ()
Returns the global default main context.

So...I think we can't make qemu_chr_add_handlers() run in colo thread in this way.
Do you have any comments about that?

but it does not work too. Do you mean like this?

Not exactly the same, but should be worth to try also. My suggestion is to replace all g_source_attach(src, NULL) in qemu source (especially the ones in qemu-char.c) with g_source_attach(src, g_main_context_get_thread_default()). (Maybe not that easy, but better to have a try).

You may refer https://developer.gnome.org/glib/unstable/glib-The-Main-Event-Loop.html for more docs on this. I believe there should have some solutions on this.

Zhang Chen

     - remove pkt->s
     - move data structure to colo-base.h
     - add colo-base.c reuse codes for filter-rewriter
     - add some filter-rewriter needs struct
     - depends on previous SocketReadState patch
     - except move qemu_chr_add_handlers()
       to colo thread
     - remove class_finalize
     - remove secondary arp codes
     - depends on previous SocketReadState patch

   - rebase colo-compare to colo-frame v2.7
   - fix most of Dave's comments
     (except RCU)
   - add TCP,UDP,ICMP and other packet comparison
   - add trace-event
   - add some comments
   - other bug fix
   - add RFC index
   - add usage in patch 1/4

   - add jhash.h

   - initial patch

Zhang Chen (4):
   colo-compare: introduce colo compare initialization
   colo-compare: track connection and enqueue packet
   colo-compare: introduce packet comparison thread
   colo-compare: add TCP,UDP,ICMP packet comparison

  include/qemu/jhash.h |  61 +++++
  net/Makefile.objs    |   2 +
  net/colo-base.c      | 183 +++++++++++++
  net/colo-base.h      |  92 +++++++
net/colo-compare.c | 745 +++++++++++++++++++++++++++++++++++++++++++++++++++
  qemu-options.hx      |  34 +++
  trace-events         |  11 +
  vl.c                 |   3 +-
  8 files changed, 1130 insertions(+), 1 deletion(-)
  create mode 100644 include/qemu/jhash.h
  create mode 100644 net/colo-base.c
  create mode 100644 net/colo-base.h
  create mode 100644 net/colo-compare.c





reply via email to

[Prev in Thread] Current Thread [Next in Thread]