Hey Medplum team! Wanted to understand more the rationale for storing all the data as text fields rather than jsonb inside Postgres and whether or not there is any discussion towards using jsonb. We've been hitting quite a few performance bottlenecks where the write throughput is quite poor (ie. <50 / second). Our server and database are running on fairly beefy instances. I have yet to run a very comprehensive performance profiling but I've attempted to scale up medplum-server replications and it doesn't seem to improve performance by much Thanks!

Hi @Generaleoley it's a good question. There are a lot of different factors that can affect write performance, but we'll try to provide some insight.

Thank you!

db size would also be helpful

The fork was taken off 3.1.6 so it's over 1 major version behind, I wouldn't be surprised if there was a significant amount of improvement since then. The main operations that I've ran that affected over 100k+ rows was: 1. Adding displays to the interperatations of around 250k Observations 2. Ingesting around 175k Communication resources All of these were done in large batch operations where I manually bumped up the max_json_size and disabled all rate-limiting I think for a proper profile I'm going to create ~100k dummy resources and ingest them both on our fork and latest upstream with the node profiler + pg profiler on and see what I find.

THanks for the detailed readout. It's hard to diagnose these perf bottlenecks remotely, as there can be many different underlying causes, but I can point you in some directions to investgate