Turns out my naive simulation of upsert caused memory usage to balloon, so it isn’t feasible at all. (Refresher: the issue is that currently, upsert is only supported when using Kafka to ingest data into Materialize. I’d rather not have Kafka as a Biff dependency, but the other options only allow appending new records, not updating old records.*) There have been relevant discussions on the Materialize slack, and they’ve mentioned that improving sources is a priority. So I’m putting a pause on the Materialize integration until that happens.
*Actually there is an Avro + Debezium file source that might work, but it’d be inconvenient and I’d rather wait for Materialize to implement more features.
In other news…
Current state of things: the landing page converts well and we have a consistent flow of new users. Unfortunately, retention sucks. My top focus currently is figuring out how to consistently get useful feedback from users so we can improve product-market fit. I came across this post from Superhuman’s founder which I thought was fabulous and (hopefully) exactly what Findka needs. So I’m gonna start doing that. (Short version: ask your users to fill out a short survey with four specific questions, and the answers will tell you everything you need to know).
I’m going to set up a job to email the survey link to anyone who uses Findka at least twice. Out of almost 600 total signups, we have about 100 users who have done so already. I sent the survey to a small handful of them yesterday, and I’ll probably do it for the rest of them today.
Besides the update on Materialize, there was a discussion about Biff on Hacker News today. I’m pleased to announce that Biff has become fairly boring for me. I’ve now been using it in production for about 2.5 months now, and I haven’t made any significant commits in over a month. There are still lots of features I’d like to add (such as Materialize, obviously, and a better deployment story), but all the critical things are already there and working smoothly.