|
On Tue, 1 Aug 2000, Juri Haberland wrote:
> And another thing: If a real server crashes ldirectord needs a long time
> to discover that this server isn't responding. I assume it is due to the
> timeout of the read request. Maybe a ping prior to the read request
> would solve this problem...
Hi!
During some tests I did with LVS, I found that ldirectord needed a
lots of seconds to detect a failed webserver if the whole server was
disconnected from the net. I think such delays are normal if you use some
kind of TCP probe to check if the server is alive.
At those tests, I modified ldirectord to issue a Ping to each
server to check if it is alive (using Net::Ping I think). If the server
responds to the ping, ldirectord continues with its tests. If the ping
fails, the server is declared dead. Using this modification I could reduce
the time it takes to do the probes down to 3 secs.
I don't have the modified ldirectord here, but I can submit the
modifications to the author if he wants. I didn't do it at first because
(unfortunately) at the same time I did the changes, a new version came up.
Shame on me! :-)
Pablo
|
