Vitess magic: reshard MySQL without touching application code. Slack moved from team sharding to channel sharding transparently. Same MySQL, better distribution. Full episode →

Madeline's team saved Slack during COVID by breaking the rules. Flipped IBM to Vitess early. 650→250 QPS. Instant fix. "If we hadn't been in position to do this, I don't know what we would have done.". Full episode →

IBM had 300,000 employees on Slack. All on one MySQL shard. When you give infrastructure a pet name (542), you know you're in trouble. Full episode →

Madeline moved every Slack message to a new database weeks before COVID would have killed the old one. 4 engineers. 10 months. 0 incidents. The secret? Do less. Full story →

Yext's infrastructure team had to edit code to move services off servers. Do the upgrade. Edit code again to move them back. Every. Single. Time. That's what happens when 1-2 people build your orchestrator. Full episode →

Yext's infrastructure team had to edit code to move services off servers. Do the upgrade. Edit code again to move them back. Every. Single. Time. That's what happens when 1-2 people build your orchestrator. Full episode →

300 devs. Manual deploys from desktops. No CI. And everyone had to share those trashcan Macs. This was Yext in 2015—right before they tried to scale from 10 to 80 servers. Full episode →

Yext had 100 services per team. They couldn't run migration scripts 200 times. Tom's fix: batch 5 services per commit. Big enough to review, small enough to rollback. Sometimes the boring answer is the right answer. Full episode →

Tom Elliott joined Yext to build iOS apps. Both projects got canceled in a year. Within a year, he was leading a migration of 2,000 services from a home-built version of Google's Borg to Nomad. Full episode →

"What's good for humans is good for Claude.". Andy generates types from Postgres, not the other way around. The database is the source of truth. Clear schemas help everyone—human and machine. →

Migration scripts drift from production code. Always. Andy's insight: make them the same code. One system for old data and new writes. "You guarantee you have the same logic in your migration code as your dual write code.". →

They built their entire platform on a document database doing relational work. "We have foreign keys between records. We do effectively a lot of joins.".Andy knew Postgres was inevitable. →

"We finally got the notification we were dreading.". IBM bought killed RethinkDB support. No alternatives. Just a deadline. Corporate M&A creating technical debt since forever. →

IBM bought the last RethinkDB hosting provider and pulled the plug. 30 days notice. Andy Monroe's 5-person team built one system for migration AND production—zero incidents. Full episode →

PagerDuty built their queue on Cassandra. Every maintainer: "Don't do this".Every manual: "DON'T DO THIS".Arup: "Best decision we made". Sometimes good enough beats technically correct.

How PagerDuty knew MySQL was dying:. They were using DRBD (disk-level replication) for their queue. At tens of transactions per second, the disk was already screaming. The clever hack had become the bottleneck.

PagerDuty at 200 customers wasn't optimizing for millions of requests. They were optimizing for one promise: every 200 response meant data was replicated across regions. Sometimes the constraint that matters isn't scale.

Moving from MySQL to Cassandra at PagerDuty:. "Your joins were in your code now". The database stopped doing the work. The complexity didn't disappear—it just moved into their application.

"Don't use Cassandra as a queue"."Don't run it across regions" ."Don't use it in high-latency networks". PagerDuty: *does all three*. Sometimes the wrong technical decision is the right business decision. Here's how they made it work:

The killer feature?. “Send me an email when this magic condition happens.”. Bobby Grayson on real workflows—and why utility > UI. 🎧 Full ep →