Sat through a two hour meeting on friday about the department’s flagship project and realized two things. One, this product is a 50/50 shot at best of ever actually becoming our new platform, and two, my guys are never going to hit the timeline with the goals as they are now. So what do you do to hurry up a project no one really thinks is ever going to go live? Scrap HA. Currently this project calls for nine different server roles. Three of them are easy to loadbalance: cache, web, and app. One of them is too important for me to scrap HA: database. However the “funky” ones that really do all the grunt work of feeding data into and out of the system as well as what all the backend users connect through? Fuck that. Those things are always increadibly hard to cluster because the vendor never even considered it. You wind up having to reverse engineer the product and custom-hack your way around any application-layer “gotchas”. I fully expected devising an HA scheme for those remaining tiers to be a quarter to a third of the manhours in getting this environment up and running. Now without that on the plate we may actually hit our timeline, and all it cost was a giant risk!
Actually I don’t have the balls to be that risky. I’m counting on VMware HA to be “good enough” for those oddball tiers until we start to hit scaling issues. If it works out it’ll be yet another in the long list of “why vmware is worth every penny I can throw at it”.