[Spread-users] More 2 second hiccups
samv at catalyst.net.nz
Tue Nov 7 22:17:12 EST 2006
We're using spread here with the supplied Perl bindings, and are
seeing some strange delays in message delivery. We've got a simple
configuration, with a single spread daemon just as a simple mailbox;
no spread daemons are running on other machines in our cluster.
The thing we're seeing is a delay of a couple of seconds *after* the
daemon has successfully received the message, but *before* the client,
which is sitting on Spread::receive() loop, gets returned the
message. We know this, because the sending side returns successfully
from the Spread::multicast call (using RELIABLE_MESS, which I assume
waits for acknowledgment from the server before returning) 2 seconds
before the process polling the mailbox gets the message.
We're currently on 3.17.2, looking at upgrading to 3.17.3 shortly. We
are not really considering a switch to version 4, unless this
particular issue has been identified and resolved by changes -
especially in light of reports from people that this issue exists in
new versions, too.
Can anyone perhaps offer any clues as to how to go about debugging
this or working around the issue? I've seen reference to change
Hurry_timeout, are people doing that by editing the hard-coded value
in the source?
This problem is infrequent enough to be difficult to reproduce, but
frequent enough to raise the average message delivery time markedly.
Sam Vilain, Systems Architect, Catalyst IT (NZ) Ltd.
phone: +64 4 499 2267 PGP ID: 0x66B25843
More information about the Spread-users