Web Cluster File Synchronization?

**Bunni** · 01-23-12, 02:40 AM

Anyone have any experience with setting up high-availability web clusters? Particularly with synchronizing files altered by apache (aka files altered by the application thats running on the webserver)? Or fuck, even a decent tool to use (NOT RSYNC)?

Seems this is a really specific niche that the few people with experience often setup custom solutions...

My usual uber tech guru forum has failed, irc channels have failed, google has failed, QA's have failed, and contact's has also failed.... And any answers i did get were COMPLETELY different approaches.

Sigh... I'd really, really like to avoid having to trial and error all these approaches. So anyone with any bit or fragment of knowledge, any contacts, or hell even shit you may have over-heard from the guys on the other floor who maintain the companies web cluster; would all be very much appreciated!

**Alundil** · 01-23-12, 11:10 AM

well I am not certain if this is going to be an approach with a high enough "bandwidth" for you or if it's "web cluster" enough.

This was a solution we cooked up to address file change issues across production clinical workstations (and those files are important application configuration files that routinely got "out of sync" with Subversion due to the installation group's poor workflow and adherence to standard policy/protocol

What we have: 400+ workstations (windows) all utilizing a 3rd Party Vendor developed (Off the Shelf) clinical application to record anesthesia documentation (vitals, notes, the whole shebang) in real time and store in a database. Pretty heavy transactional db. If those config files mentioned above got out of sync (meaning someone copied pasted files from an older workstation or directory save) bad things happen up to and including the application not working properly, printing of the reports (records) not happening, or incorrect medical data being stored, or medical data not being stored at all (very bad the last two).

What we did:

Implemented a Subversion instance (actually predates the rest of the solution and was in place for our dev side for quite a while) to maintain, track, version all of the development efforts including application configuration files.
Created a web service that builds a catalog of the current production version of the configuration packages and publishes that catalog
Created a windows service to poll the local workstation and maintain accurate list of configuration versions and hashes that list against the production catalog published by the web service mentioned previously.
The local windows service polls every 30 seconds checking to see if the clinical application we're dealing with is running or not (as the application locks certain files that we might need to update)
If the application is running, the local service sleeps for another 30 seconds and repeats the check
If the application is not running, the local service checks the webservice to see if there are new versions, if there are new versions then the local service pulls the config packages down (as MSI) and installs them
This info is displayed for the installation team via a web console we created that reads the data and displays whether machines are up to date, not up to date, and when they last checked in with the web service.

We've found that this has really helped because we've taken the "manual" aspects complete out of updating hundreds of machines (and hundreds of files per machine).

We're looking to expand this to a clustered setup across two webservers for high availability (though not entirely necessary) and fail over redundancy (more pertinent)