When I left Fitbit we were still using MySQL as the primary datastore for all tracker data. Admittedly only in the 10s of millions of users, but tracker data wa

When I left Fitbit we were still using MySQL as the primary datastore for all tracker data. Admittedly only in the 10s of millions of users, but tracker data was stored at sub second precision and indefinitely. It was definitely up in the 10s or 100s of TB. Unlike most DBs it was more write heavy than read heavy. Our scale has handled by sharding based on userId and geo. We could dynamically shift the shards when scaling the cluster up and down. I wish we had open sourced that sharding layer, it was pretty nifty.