Experimental Physics and
Industrial Control System

Geoff Savage <[email protected]> · Thu, 05 Apr 2001 13:57:30 -0500

Hi,

Make sure you close channels using ca_clear_channel when you are done.

I changed my version (R3.13.1.1) of ca_test to call ca_clear_channel at
the end and connections were made flawlessly.  I also avoided the
problem by usig a connection callback and extending the ca_pend_io
timeout to 5 seconds.

Here's the problem with some details.  With two IOCs, A and B, you run a
client on A that connects to a PV on B.  If you reboot A without closing
the connection then B still thinks that the connection still exists. 
More importantly B's TCP layer has an expected sequence number that was
generated when the initial connection was made.  Now reboot A and run a
ca client that attempts to connect to a PV on B.  When B initiates the
connection with A the sequence number is not understood by A so A tells
B and B tells A to reset.  A few seconds (~3 seconds) B attempts to
connect again and things go as expected.

			Succesful connection		
	IOC A				IOC B
1	UDP broadcast	-->	Looking for a PV		-->	UDP received
2	UDP received	<--	PV located here			<--	UDP directed
	Establish TCP connection (these are TCP messages)
3	SYN received	<--	<SEQ=100><CTL=SYN>		<--	SYN sent
4	SYN received	-->	<SEQ=300><ACK=101><CTL=SYN, ACK>-->	Established
5	Established	<--	<SEQ=101><ACK=301><CTL=ACK>	<--	Established

			Failed connection		
	IOC A				IOC B
1	UDP broadcast	-->	Looking for a PV		-->	UDP received
2	UDP received	<--	PV located here			<--	UDP directed
	Establish TCP connection (these are TCP messages)
3	SYN received	<--	<SEQ=500><CTL=SYN>		<--	SYN sent
4	SYN received	-->	<SEQ=350><ACK=150><CTL= ACK>	-->	SYN sent
5	Listen		<--	<SEQ=150><CTL=RST>		<--	SYN sent
	After a few seconds				
6	SYN received	<--	<SEQ=100><CTL=SYN>		<--	SYN sent
7	SYN received	-->	<SEQ=300><ACK=101><CTL=SYN, ACK>-->	Established
8	Established	<--	<SEQ=101><ACK=301><CTL=ACK>	<--	Established

For more information on TCP see RFC 793.

Thanks for the assistance and suggestions.

Geoff

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System