|[eclipse.org-committers] Quick update - 3.1 postmortem, upgrades and planned outage on the way|
Greetings to all, In this issue: 1. We survived 3.1... *phew* 2. Access the Bugzilla database from PHP! 3. New backend servers - outage on the way 4. CVS Proxy for firewalled users 1. We survived 3.1Looking back at it, a few things caused the eclipse.org infrastructure to be generally unavailable for the 18 hours following the release of 3.1:
- Files were available for public download as soon as they were on the servers, and mirrors didn't have time to sync up. After a while, there were so many connections to eclipse.org that mirrors couldn't even get through.
- It took too long for me to identify that we were being attacked.- We didn't have enough bandwidth to meet the demand. This could have been avoided by allowing more time for mirrors to sync up.
- We didn't have a bittorrent distribution channel set up to take advantage of P2P
The big day-saver was a WhiteBox server we're using as a disaster-recovery box. It's designed to be an offsite "plug-in-and-serve" box for critical eclipse.org services, and within minutes of connecting it up to a separate 100Mb line, eclipse.org was saved.
2. Access the Bugzilla database from PHP! You can now access the Bugzilla database using PHP scripts. For integrity purposes, the database is read-only, and you cannot read user passwords or e-mail addresses. This may be useful to pull your own bugzilla stats. Other database access, such as user-databases and access to the downloads stats database should be available soon after the new hardware is installed (See 3.) For more information on database access, please use your committer account and log into the Committer Tools at https://dev.eclipse.org/committers/ 3. New backend servers - outage required The three IBM p5 550 servers will be ready for production use soon. Two redundant servers will be handling the backend requests, such as Database, File Serving and user authentication. These servers areequipped with 16 GB of RAM and better I/O devices than the current backend server (15 disks vs. 3), so bottlenecks related to replication are likely to disappear. Also, total storage capacity will go from 250GB to 1TB.
In order to replace the current backend server with the new servers, we will have to shut down many services, including Bugzilla and CVS, for a period of less than one hour. This outage is scheduled to occur during Sunday, July 17/2005 at 6:00am Eastern time. 4. CVS Proxy for pserver and SSH connectionsAs per bug 87918 (https://bugs.eclipse.org/bugs/show_bug.cgi?id=87918), we've configured our load balancer to allow connections on standard ports to be redirected to the CVS ports for those unable to connect to CVS because of firewall restrictions. Here are the details:
proxy.eclipse.org port 80 (http) --> dev.eclipse.org port 2401 (PSERVER) proxy.eclipse.org port 443 (https) --> dev.eclipse.org port 22 (SSH) The attached screenshot shows a sample configuration to Extssh. We will be updating our CVS How-To pages shortly with this information. As usual, your comments, suggestions and questions are welcome. Thanks, Denis -- Eclipse WebMaster - webmaster@xxxxxxxxxxx Questions? Consult the FAQ at http://www.eclipse.org/webmaster/faq.html View my status at http://www.eclipse.org/webmaster/main.html
Back to the top