Wil Genovese wrote:
> We've been working on this setup for well over two months.  First we 
> read every post and blog we could find to get answers.  Then I started 
> posting to the Adobe forums and got a little help.  While it is true I 
> just posted to House of Fusion, I and my sys admin have been asking for 
> help in the Adobe forums for several weeks.  Each week getting a little 
> closer to a working setup but still finding yet another problem.  I 
> figured as a last resort I would post here.  So, I guess it is my fault 
> for assuming the Adobe Coldfusion forums was the correct place to go for 
> help as it is the Official Support Forum. 

To put it bluntly: you should go to the place the people with answers 
hang out. There is a clear correlation between the level of the users 
and the medium they communicate through, and forums are at the bottom of 
the food chain.


> Oh, we've had time to work on this.  We've been searching for then 
> asking for help.  The real point is is should not be this hard to set 
> get a working cluster setup.

It is as hard as it is and it would be counterproductive to reduce that 
complexity and force everybody into a one-size-fits-all solution.

I think the real problem with cluster setup in CF lies in the user 
interface. There is a fine line between the user interface simplifying 
and automating tasks and between hiding necessary complexity. I think 
the user interface in CF 7 crossed that line. The fact of the matter is 
that for no cluster I have ever set up the user interface was 
sufficient, I have always had to drop down to the XML files to change 
settings manually.


> In the end myself and our sysadmin would prefer to have a properly 
> working load balanced, fail over, session sharing cluster.  Not being 
> able to achieve that in two months work we're forced to go to a less 
> preferred but functional setup.  I've talked it over with our sysadmin 
> and we are still willing to setup test servers in house using our 
> staging databases to see if this type of setup will ever work.

For the lockup when an instance joins the cluster:

Between which loglines is the delay exactly (most easily seen if you run 
JRun from the command line or with "tail -f")?

Can you set up a simple template that appends a timestamp to a logfile 
and then sends a meta refresh to the browser with one second delay? 
Deploy that template on both instances and run it from localhost through 
the build-in webserver. Log what happens during a forced cluster break 
(unplug the network until you see in the log that the cluster break is 
detected, can take up to 2 MSL / 4 minutes, then plug back in).
Then try again through a shutdown / restart cycle (you will only get one 
logfile out of that).


For the failure to detect an instance is down and fail over:

What algorithm did you use for the connector?

If you enable verbose logging in the webserver connector (-l), what do 
you see at the time of fail over?

How long does it take exactly to fail over?

If it takes more then 4 minutes for the webserver detector to detect 
that an instance is down, how long does it take after you have changed 
your TCP stack settings to use a lower value for the TIME-WAIT timeout 
(RFC 793)?


Please make sure and manually force a time synchronization before you 
start each test.

Jochem

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~|
Create robust enterprise, web RIAs.
Upgrade & integrate Adobe Coldfusion MX7 with Flex 2
http://www.adobe.com/products/coldfusion/flex2/?sdid=RVJP

Archive: 
http://www.houseoffusion.com/groups/CF-Server/message.cfm/messageid:6327
Subscription: http://www.houseoffusion.com/groups/CF-Server/subscribe.cfm
Unsubscribe: 
http://www.houseoffusion.com/cf_lists/unsubscribe.cfm?user=11502.10531.10

Reply via email to