[kevrone]

At Timehop we currently work with a single instance AWS Aurora (MySQL-ish) database with over 40TB of data (plus a read-only replica on a smaller instance). Some stats: 1.5MB/sec receive throughput, 10MB/sec transmit throughput, commit latency around 3-4ms (with very occasional spikes to 10-20ms), select & commit counts are about 300/s, and select latency hovers around 35ms (we do about a dozen unions per query though).

All in all it's the easiest relational database I've ever worked with in terms of stability, speed, and scalability. I know this sounds like an ad for Aurora, but I just really like it.

[dirtyaura]

I'm curious, how does the replication work if the replica instance is smaller (I assume smaller in disk space)? Is is automatically removing some of the data from the replica based on a heuristic rule?

[kevrone]

Smaller instances are smaller in memory/compute power only. Storage is charged separately and the implementation details are unknown to me.

[kbyatnal]

Bit of an aside, but why haven't you guys listened to your users yet? Based on all the negative feedback about the recent updates, what are you guys doing?

[kevrone]

Those decisions are above my pay grade mate ;)

[WrtCdEvrydy]

What's the backing DB? Did you use MariaDB, or did you end up using Postgres?

[kevrone]

Aurora is it's own database type. The interface is MySQL-compatible, but it's not a perfect match.