List:Commits« Previous MessageNext Message »
From:Ingo Strüwing Date:January 27 2009 3:46pm
Subject:Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)
Bug#37780
View as plain text  
Hi Davi,

Davi Arnaut, 26.01.2009 19:30:

> On 1/21/09 3:05 PM, Ingo Strüwing wrote:
...
>> Davi Arnaut, 20.01.2009 21:22:
>>
>> ...
>>> On 1/11/09 4:05 PM, Ingo Struewing wrote:
...
>> Yes. Keeping the write direction open means that the client can still
>> receive data. The worst thing that can happen is that the client gets an
>> error on an attempt to write to the socket. If it interprets this as
>> "Lost connection to server", then we've won anyway. Just an error
>> message from the server might get lost.
> 
> My doubt is about a client polling his socket and bailing out in case
> poll returns a error if one of the directions is closed. This needs to
> be documented as a protocol change.


Ok. Documenting the half-close in KILL situations is surely a good
thing. However, i don't know where to look for the protocol
documentation. Can you help out?

OTOH we would need a "broken" implementation of the Berkeley sockets. A
reasonably conforming implementation would not take notice of the close
of the read direction on the peer. Unless it tries to actually write to
the socket. A poll on the write direction will only notice the
transition from a full write buffer to a non-full write buffer.

Anyway, if a poll misinterprets the half-close as connection failure of
some kind, it could try to work around it, or close by itself. In the
former case the full close follows soon. Finally the closed state is
reached in any case. The only problem might be wrong error messages.

...
>> I would like to keep adding sync points as easy as possible. If
>> debug_sync.h is included in mysql_priv.h, it is available in most files.
>> No extra check for "#include<debug_sync.h>" needs to be done after
>> adding a sync point. Since debug_sync.h has protection against double
>> use, I'll leave the explicit uses in the files where you added them.
> 
> Please, don't add it back to mysql_priv.h.. easy of use is not a concern
> here.


This might be a misunderstanding. I do not want to add back the
definitions. But just an #include <debug_sync.h>. As I proposed in my patch.

If this is, what you thought, but still object, please explain, what the
problem is.

...
>>>> +    /* purecov: begin inspected */
>>>>        put_error(NULL);
>>>>        free_defaults(defaults_argv);
>>>> +    batch_readline_end(status.line_buff);
>>>>        my_end(0);
>>>>        exit(1);
>>>> +    /* purecov: end */
>>> Unrelated, please commit separated so it can be pushed early.
>>
>>
>> Reporting a bug for this, and waiting for it to become tagged by triage,
>> may finally take longer than to have it included here.
> 
> OK to push separately. I didn't suggest a separate bug report.


Ok, will do. But please note that in former time it was a rule not to
make patches that are neither related to a bug nor a worklog.

...
>> Please allow me to leave the paragraph in.
> 
> OK, but if they didn't pay attention to the read and understand
> pthread_cond_wait documentation, I think they won't bother to read your
> long comments too.


Possible. And it would be ok then. And there are those, who don't
understand, and still don't read. But I hope for those who read and
consider it.

BTW I'm on vacation this week. The fixed patch will come next week only.

Regards
Ingo
-- 
Ingo Strüwing, Database Group
Sun Microsystems GmbH, Sonnenallee 1, D-85551 Kirchheim-Heimstetten
Geschäftsführer: Thomas Schröder, Wolfgang Engels, Dr. Roland Bömer
Vorsitzender des Aufsichtsrates: Martin Häring   HRB München 161028
Thread
bzr commit into mysql-6.0-backup branch (ingo.struewing:2749) Bug#37780Ingo Struewing11 Jan
  • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Davi Arnaut20 Jan
    • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Ingo Strüwing21 Jan
      • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Davi Arnaut26 Jan
        • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Ingo Strüwing27 Jan
          • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749) Bug#37780Paul DuBois27 Jan
          • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Davi Arnaut28 Jan
    • Re: bzr commit into mysql-6.0-backup branch (ingo.struewing:2749)Bug#37780Ingo Strüwing25 Feb