This comment makes me unreasonably giddy. The replies are focused on MySQL (and to a lesser extent, PHP). None of that matters. The hero here is the logging s

This comment makes me unreasonably giddy. The replies are focused on MySQL (and to a lesser extent, PHP). None of that matters. The hero here is the logging system serving as seekable stream of collected events. (I’m trying hard here to not say K-fka). By consuming events from that source and being able to perform time-based seeking, you’ve offloaded (or dare I say, eliminated) your scaling pressures. Even if you commit the “sins” of no indexes on heavy queries, using MyISAM, or building a giant, memory-intensive associative array… none of it matters. You can choose to wipe your local dataset, start again in a different language, switch to another data store model, and none of your downstream consumers will know nor care (unless you blog about it). *This* type of design, that prioritizes consumer-facing resiliency and continuity, is what I look for in systems that scale.