RE: Hard Failures, KeepAlive, and Failover --Follow-up

Question

Hi,
It's a really challenging question. However, what do you want to do after
the network crash? Failover or just stop the service? Should we assume
that when the network is down, and so do your name service?
One idea is to use externalconnection to "listen" to your external non-forte
alarm, so do "whatever" after you receive the alarm instead of letting the
"logical connection" to time out or hang.
Regards,
Peter Sham.
-----Original Message-----
From: Michael Lee [SMTP:[email protected]]
Sent: Wednesday, June 16, 1999 12:44 AM
To: [email protected]
Subject: Hard Failures, KeepAlive, and Failover -- Follow-up
I've gotten a handful of responses to my original post, and the suggested
solutions are all variations on the same theme -- periodically ping remote
nodes/partitions and then react when the node/partition goes down. In
other circumstance this would work, but unless I'm missing something this
solution doesn't solve the problem I'm running into.
Some background...
When a connection is set up between partitions on two different nodes,
Forte is effectively establishing two connections: a "physical
connection"
over TCP/IP between two ports and a "logical connection" between the two
partitions (running on top of the physical connection). Once a connection
is established between two partitions Forte assumes the logical connection
is valid until one of two things happen:
1) The logical connection is broken (by shutting down a partition from
Econsole/Escript, by killing a node manager, by terminating the ftexec,
etc.)
2) Forte detects that the physical connection is broken (via its KeepAlive
functionality).
If a physical connection is broken (via a cut cable or power-off
condition), and Forte has not yet detected the situation (via a KeepAlive
failure), the logical connection is still valid and Forte will still allow
method calls on the remote partition. In effect, Forte thinks the remote
partition is still up and running. In this situation, any method calls
made after the physical connection has been broken will simply hang. No
exceptions are generated and failover does not occur.
However, once a KeepAlive failure is detected all is made right.
Unfortunately, the lowest-bound latency of KeepAlive is greater than one
second, and we need to detect and react to hard failures in the 250-500ms
range. Using technology outside of Forte we are able to detect the hard
failures within the required times, but we haven't been able to get Forte
to react to this "outside" knowledge. Here's why:
Since Forte has not yet detected a KeepAlive failure, the logical
connection to the remote partition is still "valid". Although there are a
number of mechanisms that would allow a logical connection to be broken,
they all assume a valid physical connection -- which, of course, we don't
have!
It appears I'm in a "Catch-22" situation: In order to break a logical
connection between partitions, I need a valid physical connection. But
the
reason I'm trying to break the logical connection in the first place is
that I know (but Forte doesn't yet know) that the physical connection has
been broken.
If anyone knows a way around this Catch-22, please let me know.
Mike
To unsubscribe, email '[email protected]' with
'unsubscribe forte-users' as the body of the message.
Searchable thread archive <URL:http://pinehurst.sageit.com/listarchive/>-
To unsubscribe, email '[email protected]' with
'unsubscribe forte-users' as the body of the message.
Searchable thread archive <URL:http://pinehurst.sageit.com/listarchive/>

Answer 1

Make sure you chose the right format, and as far as partitioning in concerned, you have to select at least one partition, which will be the entire drive.