Just want to share an exciting development with the community - we have managed to get support for postgres clusters working in openbalena, which provides a significant performance improvement. We are running postgers as a standalone percona cluster, which is installed and managed via Percona’s postgresql operator. The cluster provides both primary and replica servers, however up until now openbalena has been configured using the standard method, which entails setting the environment variables DB_HOST, DB_NAME, DB_USER and DB_PASSWORD in the open-balena-api container. This static configuration only supports a single database server, so all calls were directed to the single primary postgres server. We found that pinejs has internal support for separating out read and write transactions and funneling them to primary or replica servers accordingly. To enable this, we had to pass an additional databaseOptions parameter to pinejs, which was not being provided by open-balena-api in the main init.ts entrypoint.
To get around this, we used a kubernetes init container which prepares the databaseOptions parameter based on settings you can configure in the helm chart, including settings for both the primary and replica databases. Then we use a startup script for the main open-balena-api pod to apply this additional setting to init.ts and voila - it works with our postgres cluster! Overall performance is dramatically better, i.e. in firing off 100 parallel api calls to the old vs. new setup has resulted in a >10x performance improvement - so thought I’d share with the community. You can find the relevant helm charts here if you’re interested in playing with this or other features included in there such as a build server, delta server etc. I know many of you run openbalena with docker compose so this same methodology could be adapted for that as well. Hope this helps!