Weekly PlanetScale health digest with Linear backlog items

Every Monday at 8am, on a cron schedule, review the health of our PlanetScale databases and turn the findings into work we will actually do.

Start in PlanetScale with List databases to get every database in our organization. Work out which ones are production databases and review only those, skipping development and sandbox databases.

For each production database, pull two things from PlanetScale. First, List schema recommendations, which returns tailored recommendations in the form of DDL statements generated from query level telemetry, system tables, and the database schema. Second, List branch queries for its production branch, so you have recent query activity to reason about.

Group the recommendations by type: add an index, drop an unused index, drop an unused table, and primary key ID exhaustion risk. Then rank them by likely impact. Use the recent branch query data as your evidence, so a recommendation touching a table or query pattern with high call volume or slow response times ranks above one touching something barely used. Remember that unnecessary indexes are not free: while indexes can drastically improve query performance, having unnecessary indexes slows down writes and consumes additional storage and memory.

Post a single digest to our engineering channel in Slack using Send a Message. One digest per run, not one message per database. Lead with the top items across all databases, and give each one a single line of rationale naming the database and table it affects and why it is worth doing. Group the remaining items by recommendation type underneath, and keep the whole thing skimmable.

The digest must carry this caveat prominently, every single week and not just the first time: once a drop unused index recommendation is opened, it will remain open even if the index is subsequently used. Tell the reader to check current Insights usage data to verify that the index is still unused before permanently dropping it. Never present a drop index item as safe to action on the strength of the recommendation alone.

For anything judged high impact, create a Linear issue with Create Issue so it lands in the backlog instead of getting lost in chat. Put the suggested DDL statement in the issue description, together with the database and branch it applies to, the recommendation type, and the rationale for the ranking. Reference the created issue from its line in the Slack digest so the two do not drift apart. Do not create issues for low impact items, since the point is a backlog people trust rather than an exhaustive dump.

This workflow must never apply DDL itself. It reads, summarizes, notifies, and files tickets. Every schema change stays a human decision, executed by a person after review.

Keep the digest scoped to schema recommendations and query metadata. The PlanetScale API manages platform resources only and cannot read the data inside the databases, so never attempt to query table contents or report on customer data. If a week produces no recommendations, still post a short digest confirming the check ran and everything looks clean.