Hi all,
First of all, sorry for the long story...
I have a strange situation with my CSS boxes and our enviroment. Let me describe the infraestructure.
Our CSS boxes was deployed to serve the application on 2008. The hole solution was based on Microsoft IIS webservers. It's an e-commerce website with many stores.
Since then all, the requests always were answered by the same servers, the content URLs config is a wildcard option, as the example bellow:
OLD SITE CONTENT CONFIG
content old-platform
vip address xxx.xxx.xxx.xxx
balance weightedrr
flow-timeout-multiplier 1
advanced-balance arrowpoint-cookie
protocol tcp
port 80
url "/*"
no persistent
add service ms-webserver1
add service ms-webserver2
add service ms-webserver3
active
This was the approach adopted from the IT staff on 2008 during the CSS implementation.
So, today we have another web application based on Apache webserver. This new application's goal is to replace the old one. We are doing the migration now in the following manner:
We are selecting some of the stores to be migrated following the company strategy.
So, after migrating the first selected store, we've had to rollback some hours latter due to database problems, and at that point, the CSS behaves completly crazy.
My approach to migrate the stores was:
- Create a new content, with the new apache-servers inside declaring the store URL that I wanted to migrate on the content config as follows:
NEW SITE CONTENT CONFIG
content new-platform
vip address xxx.xxx.xxx.xxx (same VIP address from the old site content config)
balance weightedrr
advanced-balance arrowpoint-cookie
protocol tcp
port 80
url "//new.website.com/*" (Same URL from the old store that was not declared explicitly on OLD content)
add service apache-webserver1
add service apache-webserver2
add service apache-webserver3
add service apache-webserver4
active
When I was planning the migration, my thought was, if I declare the URL, the CSS will match the requested URL with this new content and the new website will be delivered. This worked fine.
The problem I've ran across was at rollback time. I thought, if I type suspend, the CSS will back to match the requested URL with the old content and the old website will be delivered.
So what's happend was, the CSS starts to blink the status led in Ambar, indicating a fail, and becames unavailable to answer request and even to telnet. The only available access was console.
I have two CSS with box-to-box redundancy enabled. To eliminate the chances that was a problem with the box, I've inverted the master and slave order, and the same happend when I type, suspend on the new content on backup unity (as a master).
The question is, have I adopted the wrong approach to handle this migration?
Is there a best practice to achieve this task?
A reboot brings all to normal state.
Best regards,
Fabiano Martins
