On 10/22/2014 04:04 AM, Rich Kulawiec wrote:
I've seen similar tactical mistakes when developers insist that information *must* be stored in a relational database -- even though plain old ordinary text files are perfectly adequate for the task, are easier to debug, are easier to fix, and easier to maintain.
Right on, bro. I'm in the middle of a "system" where the architect insists on just that. Never mind that I brought the relational database server to its knees (even flat on its back) because of the sheer number of processes I'm running, and the load on the engine that causes. My *ONLY* option was to replicate the information in the database and store it in files on the servers I'm running on -- semi-permanent data on the hard drives, working data in a RAM disk system. "But updates don't happen instantly, we have to wait for your replication daemon to run!" Yep. Amazing how that makes the entire system considerably more stable when you don't change horses mid-flight. Not to mention the considerable reduction in inter-server network traffic. How did this discussion get into NANOG? :)