I am rolling out my app in the next couple months, which is basically a variant of a social network but with a purpose. Now I have not included any high availability factor or any form of clustering or replication. Initially I thought I would start with one server since i would have minimal requests and later expand to a cluster based on the traffic and requests.
But while load testing I saw my database crash due space issues and I spend some time recovering it. So to make my app a going concern I don't want to lose my database or any data for that matter if my database crashes, goes corrupt or for any other reasons.
I am using both MySQL and Neo4j together and my application tries to keep them in sync. I feel tempted to implement a MySQL and Neo4j cluster using Galera MySQL cluster and Neo4j HA cluster respectively over three Amazon instances. It seems to be an overkill though considering I won't have enough traffic in the beginning and most of my resource would be idle (and expensive).
I am ready to put in the extra effort but money-wise I want to be wise! Please let me know if in the initial stage I should go for HA or take a risk and be satisfied with backups and be content with data loss if any due to failures.
High availability is one piece of a reliable infrastructure but I'd put it much lower on my priorities list than investment in automated backup, deployment, and monitoring. For me, the key part of your experience and question is, "I saw my database crash due space issues and I spend some time recovering it." This happened in testing, so I wouldn't expect you to have everything sailing along, but in an ideally configured production environment, you'd know about your space issues ahead of time and address them on your terms. If it did go down, you wouldn't recover, you'd restore, and this would be more of an inconvenience than anything else.
Since you're worried about uptime and not performance, all you really need to focus on now is the uptime expectation of your app. If you only have one server and monitoring informs you of an error that will require brief downtime -- maybe you need to upgrade hardware, move to new storage, install a critical patch -- is it acceptable to be offline for a brief period of time?
If so, skip HA for now and make sure you have all the proactive layers in place. Maybe invest some money that would go towards servers in an HA environment on consultants to make your environment rock solid if there's any concern that you're missing something.
If not, I think you have your answer!