Battle with a Phantom Postgres WAL Segment

blog.keikooda.net

38 points·onderkalaci··8 comments

The follower hole was a long running bug for us back when I was at Heroku, it existed for at least 5 years. In fact it'd become a bit of a joke so much so that when we'd start our planning process each time (http://www.craigkerstiens.com/2013/03/13/planning-and-priori...) we'd start by placing "fix the follower hole" in the high on effort and low on impact to use as an anchor for everything else.

Congrats to the team for digging into a hard problem which Daniel never solved ;), squashing it, and then the great write-up about it.

Thanks for sharing that link, and your story about your time at Heroku, it is very inspiring. But I have one quesiton, who is Daniel? :)

Ha. Daniel was the first engineer hired on the Heroku Postgres team. He authored wal-e (https://github.com/wal-e/wal-e) which is the foundation for much of the fork/follower functionality and thus the follower-hole itself. He's now running databases with me at Citus.

and my job here at Heroku Postgres is to delete Daniel’s code

Destroying is so much more work than creating.

Off-topic question: Does anyone know which software the authors used to create the illustrations?

It's a paper and pencil :P I used Ponky (https://www.mpuni.co.jp/products/pencils/color/school/ponky/...) to color and scan. After I realized that my handwriting is terrible, I replaced several of my handwriting letters to fonts using Pixelmator ;)

Headers on the JPGs say Pixelmator 2, so looks like they were done at least partly by hand.

Reply on news.ycombinator.com