Query planning is hard. Doing so in a massive, sharded database is even harder. The blog we published today at PlanetScale was a labor of love. So many people from across the company helped with technical accuracy, writing, visuals, and making it the best it can be. This was... so much fun to be a part of. Go read. Get a little better at database query planners this evening.show more

Great article! I can't stand when people use I'd as the primary key name Like yeah let's use is but on the fks use customerid

I think the greatest challenge with understanding how it works is that many people never faced problems that you're solving themselves. Hard to go from theory to practice in this field. Great blogpost and visuals! 🚀

gorgeous animations

The animations are so great! Would love to see how you guys build those too

+1 for blog 👍 Neki supports consistent hash as the sharding method only? Do cross-shard queries scale well -- didnt notice much mentioned about it in the blog.

the cross-functional part is the real craft here. i make handoffs explicit and keep the complaint verbatim; a polished query plan is only useful when the next person can see what changed and why.

Have you written/talked about how you cooridate content work like this? It's excellent btw

Went for the cool animations directly and learnt from blog

Looks nice

damn

Following one customer and their orders makes the shard-key decision click. You can see why a perfectly ordinary JOIN suddenly has a travel budget. More database explanations should start with where the rows actually live.

Time to develop Neki, 1 month, the blog post, 2 months

Can u post more on this

查询规划确实越复杂越难搞，尤其是分库场景下坑真多，我晚上去看看这篇博客学习下。

planetscale writeups are consistently worth the read, looking forward to this one.

please keep this comment you have taught me so much in the past three months then I could’ve dreamed of, for years I kind of just put off DBA tasks but I realize now that’s where all the alpha is 🫱🏾‍🫲🏻

This is really amazing work

great talk!

The sharded case is where a planner stops being an optimisation and becomes a correctness surface, because a plan that fans out to the wrong shard still returns rows. What we watch is whether the row count matches across a re-plan. Does the post cover plan stability?