Re: CA client library and Connection handler

[Adam Jon DeGrush <degrush@MIT.EDU>]

Thanks Brian for the help.

On Thu, 31 Jan 2002, Brian McAllister wrote:

>
> >>> On 1/30/2002 at 19:54:19 EST, Adam Jon DeGrush wrote:
>
>   > I am trying to monitor the connection of channels using
>   > ca_change_connection_event from the client library and am having
>   > difficulty getting it to work. I would like to do a group poll
>   > (synchronous or asynchronously) of my channels at a certain frequency
>   > and would like make sure that my channels are connected before I do
>   > this.

> First, are you "monitoring" your channels (using ca_add_event()) or are you
> using ca_get() in your poll function ?

I was using ca_sg_array_get() for polling in this case. I wasn't doing any
monitoring of channels here but I was planning on installing monitors
using ca_add_event() in the future. Why? Is there problems with this?
 >
> In any case, there is a field in the CHID that holds the status of the
> connection, and a macro defined in cadef.h for accessing it (ca_state()).
> The possible values are cs_never_conn, cs_prev_conn, cs_conn, cs_closed.
>
>   > I though I could just install a connection handler like in the
>   > example:
>   >
>   >   for (pCurrent = pFirst; pCurrent != NULL; pCurrent = pCurrent->next){
>   >      status=ca_change_connection_event(pCurrent->chan, myHandler);
>   >      SEVCHK(status, "ca_change_connection_event:
>   >      couldn't change handler");
>   >   }
>
> ca_change_connection_event() is normally used to replace an existing
> connection handler, which would have been initially associated with the
> channel by specifying it in the ca_search_and_connect() call used to open
> the channel.
>
ca_search_and_connect() was used at initialization with it returning
cs_conn before replacing it with ca_change_connection_event().

>   > and then have something like:
>   >
>   > while(1){
>   > 	ca_pend_event(POLL); //
>   > 	myGroupPoll();
>   > 	nanosleep(fiveseconds,NULL);
>   > }
>   >
>   > and then have my handler either block or remove from my list those
>   > channels that have been disconnected. Instead when I run it and
>   > disconnect a channel there is a pause of about 5 seconds and then a
>   > segmentation violation( although I think handler does get called).
>
> (The "pause" is probably the time it takes for your system to write out a
> "core" file.)
>
> What do you mean by "disconnect a channel" ?  Your connection handler has
> been invoked *because* the channel has changed connection state.  Either it
> has re-connected, or it is already disconnected.

I was speaking loosely. By disconnecting I meant I my ioc was rebooting.

>
>   > The same thing happens if I remove myGroupPoll().
>
> Which clearly implies that the action which causes the seg fault is in your
> connection handler.

I still see a seg fault if I replace myHandler with NULL or I just have
myHandler printing out the name of the channel via
ca_name(connect_args.chid). The name is printed correctly before aborting.

> If all you are doing in the connection handler is flagging when a channel
> is available or not, the CA library does that for you.  You only really
> need the connection handler if you need special handling (initialization,
> message generation, cleanup, etc) when the connection changes state.
> Otherwise, just test the state before you try to access the value.
>
Thanks for this info, you're right this is probably the better way to do
this.

Adam DeGrush