Posted by alexus on March 3, 2011 at 5:24pm
We’re trying to ascertain the best way to achieve good redundancy in an enterprise-level utilization of Drupal. Ideally, We would like to have multiple instances of Drupal on different physical hosts/VMs. For instance, if we need to perform maintenance on one host/VM, the site doesn’t go down because there’s always another instance up, or in case of DOS attack, we’d have a better chance of sustaining because we’d have multiple instances running instead of one.
Have others been able to achieved something like this? Any advice, caveats or suggestions?
Comments
Use one of the Cloud solutions
They all handle most of it transparently.
This is a great video to
This is a great video to watch to see how Acquia built their HA cloud hosting http://www.archive.org/details/BuildingScalableHighPerformanceDrupalSite...
Dead link
@langworthy do you have another link. This one is no longer avail. Thz
shorten link
@repoman: same link but shorten http://bit.ly/gmNZq4
@langworthy: I'm not looking to do Clouds, I'm interested in doing that in-house type of scenario, besides Clouds doesn't really answer HA question, you just suggesting me to use another kind of technology.
alexus.org && alexus.biz
I think the most common
I think the most common configuration for multiple servers (and the one that we use in-house for our line of clustered VPSes) is:
Varnish can either be your load balancer, or you can have an instance of Varnish running on each web-head for higher redundancy (but lower performance). Memcache can either go on the DB slave, or a dedicated server.
If you just want to ramp-up to that size of configuration you can
You can also add multiple memcache servers as necessary, and more DB slaves.
--
Dave Hansen-Lange
Director of Technical Strategy, Advomatic.com
Pronouns: he/him/his
Shared Files Dir
This module I created has significantly sped up our page generation times.
http://drupal.org/project/imageinfo_cache
Gains depends how media rich your site is. The more images the better your improvement.
@dalin: 1 load balancer /
@dalin: 1 load balancer / firewall is already not a HA since there is one point of failure, for HA redundancy is number 1 priority. Do you know how would Varnish/Drupal(apache) work in regards to logged in users, meaning if a user logges into one Drupal and goes for a coffe brake and come back in 5-10 minutes do you think Drupal will pick up no problem?
@mikeytown2: thanks, but this is off topic
alexus.org && alexus.biz
You can do more than one load
You can do more than one load balancer, utilizing HAProxy/Heartbeat:
http://www.howtoforge.com/high-availability-load-balancer-haproxy-heartb...
HollyIT - Grab the Netbeans Drupal Development Tool at GitHub.
Why do I need HAProxy? I'm
Why do I need HAProxy? I'm already using Varnish and it can be configured as Load Balancing. I assume Heartbeat is for floating IP.
alexus.org && alexus.biz
Yes Drupal works just fine
Yes Drupal works just fine for logged in users and multiple servers. Session data is stored in the database rather than the file-system so as long as the web-head can access the database it's all fine. Varnish doesn't really apply to this question since you generally don't use it for authenticated users unless you get into some really advanced stuff using edge-side-includes.
--
Dave Hansen-Lange
Director of Technical Strategy, Advomatic.com
Pronouns: he/him/his
Don't forget the role that
Don't forget the role that Aegir can play in making this happen. Physical infrastructure is one thing, but getting your updates done in a sane way is an entirely different game.
Check out the 2.0 RC version of pfSense for firewalls, routing and ha-proxy. I run it on a couple of old 1u machines at the perimeter and it has never let me down.
i'm talking more into
i'm talking more into enterprise level, where you have separate hardware responsible for firewall, so pfSense isn't something that I'd ever use in my environment
alexus.org && alexus.biz
i've seen hardware solutions
i've seen hardware solutions randomly lag/fail in unexpected ways and there is no way to fix them.
HA on MySQL, InnoDB
You may want to look into SchoonerSQL from Schooner Information Technology. It's a commercially licensed HA release of MySQL, 100% compatible with MySQL 5.1 & InnoDB. SchoonerSQL is a shared-nothing cluster incorporating synchronous multi-threaded replication between the Master and the Read-only Master (Slave), 4x faster than other MySQL builds.
http://www.schoonerinfotech.com/products/schoonersql/index.php
HowTo: Scaling Drupal on Multiple Servers
(disclosure: I work for Severalnines)
A number of our users are running a clustered Drupal/MySQL setup to load-balance traffic on multiple servers, and achieve High Availability.
http://www.severalnines.com/blog/scaling-drupal-multiple-servers-galera-...
The blog post goes through:
- setting up multiple web servers + HAProxy for load-balancing
- how to cluster mysql (multi-master cluster for InnoDB)
- how to set up file system clustering
- migrating data from your single instance drupal to the clustered environment
Drupal Run Book
I'd just like to raise my hand in this old issue, to share a key old Unix Sys Adminism on the subject. High Availability on LAMP is doable, but don't do it without writing a Run Book for yourself and colleagues, all along the way.
We're not just talking git, proper code commenting, config check-ins, Network maps and tools here. Write yourself a book to tie it all together. Or, at very least, implement rigorous Confluence requirements and key areas of expertise within your team.
When you start to get older, you'll be glad you did.
http://drupalcreations.com/Drupal_HIgh_availability
Robbie Jackson
www.drupalcreations.com
robbiejackson@drupalcreations.com