From: Benoit Daloze <eregontp@...>
Date: 2009-11-26T06:48:18+09:00
Subject: Re: DRb::DRbObject.alive?

--001636498e570ffa600479390438
Content-Type: text/plain; charset=ISO-8859-1

Yeah, you're right.

So here's how I did to not duplicate the code:

     def try_on_clients
        @clients.each { |client|
            begin
                yield client
            rescue DRb::DRbConnError
                del_drb_client(client)
            end
        }
    end

    def receive_message(msg)
        try_on_clients { |client| client.show_message(msg) }
    end

That looks already more beautiful, and doesn't spend time looking if alive.
Anyway, it's a little too bad for the client, because any connection error
will drop him away, but he can reconnect easily.

Regards,
Benoit

2009/11/25 Robert Klemme <shortcutter@googlemail.com>

> On 11/25/2009 12:04 AM, Benoit Daloze wrote:
>
>> I found a feature that seem to be missing to me in DRb.
>>
>> I'm just creating a chat with a client and a server, using only DRb to
>> interact between them.
>>
>> My problem is I didn't found a nice way to verify if a client is alive.
>> I would like to be able to update my clients' list, but I need to know if
>> they are alive.
>>
>> Currently, I'm getting around with catching the Exception
>> DRb::DRbConnError,
>> but that looks really ugly:
>> @clients.each { |c|
>>  begin
>>    c.respond_to?(:some_method_defined_in_client) #Trying to use any method
>> of c will raise the exception (except if not defined)
>>  rescue DRb::DRbConnError
>>    @clients.delete(c)
>>  end
>> }
>>
>> Using Exceptions would be meaningful when I try to send(it would raise the
>> exception if I call #send), but then I would have to manage Exceptions
>> everywhere.
>>
>
> Well, but this is the usual approach.
>
>
>  The method alive? is defined for the DRbServer and DRbTCPSocket, but not
>> for
>> DRbObject! (in fact if the Client just got disconnected, alive? will still
>> return true)
>>
>
> It should not be too difficult to add method alive? to DRbUndumped.
> However, there is a more fundamental problem that cannot be solved: even if
> you have method alive? and it is working properly what does it tell you?  It
> tells you that the object was alive when you invoked that method.  You have
> no guarantees that it will be alive when you make the next remote method
> call.  Hence you need to handle exceptions anyway. Now you have exception
> handling plus additional code that checks liveness beforehand.  Your code is
> complexer and slower.
>
>
>  Does somebody already met this?
>>
>
> This is a common issue people stumble over all the time (typically with
> things that involve interprocess communication such as for database
> connections and the like).  The issue is always the same: you do not gain
> anything with this method because you have to deal with failure anyway.
>
> Kind regards
>
>        robert
>
> --
> remember.guy do |as, often| as.you_can - without end
> http://blog.rubybestpractices.com/
>
>

--001636498e570ffa600479390438--