PostgreSQL Uncovered: Internals, Trace Analysis, and Performance

This case shows that the plan citus picked is definitely not the best. After digging through a lot of traces, I can only say citus still isn’t very mature.

Why guess when you can know?

The distributed planner in citus is basically a heuristic planner with simple logic. A real distributed planner is much more complicated than what citus has. I’ve also seen most of the major flaws in the extension approach.

To explain extensions clearly in my postgres course, I spent a lot of time tracing citus, which took at least a full week. I doubt anyone in the future will dig so deeply into postgres just to write a book or teach a course. Would you?

citus handles the A, C, and D of ACID. The paper also says extensions currently don’t support implementing isolation (I).

The 2025 Framer Awards are open. Compete in 10 categories for $100,000 in prizes. New to Framer? Start building today and join the designers creating the internet’s best websites.

I think I’ve got a solid understanding of how citus’s distributed planner and executor work now. Tracing the internals really helped me see how you can build a distributed database just by using extensions.

Once you understand Postgres internals, you will at least realize: (1) There is no perfect database in the world today. (2) Building on MySQL, especially for distributed databases, is so painful. (3) MySQL is weak at handling joins, even with more advanced dynamic programming

Anyone truly passionate about databases should study Postgres internals, especially the planner and extensions. Without this knowledge, a database career is incomplete.

Plans often fall behind reality. Six months into the Postgres course, I still have no idea when I’ll actually finish it, but at least I think I have already mastered the toughest part, the planner.

We are FLYING 9% up Wow in first 45 min OMG ! $NVDA $TSLA $AMC $GME

Owing to the extensive depth of the planner content, including 45 days dedicated to developing the trace tool, 20 days of material preparation, and 10 days of writing, even with full effort I still need five more days for final refinement. I apologize for the delay and appreciate

I really suggest MySQL developers and enthusiasts study Postgres in depth. It might answer a lot of the questions you’ve been wondering about.

Thanks for the reminder. The more accurate terms are the DP-based search (standard_join_search) and GEQO.

Postgres’s DP planner isn’t perfect. Sometimes the genetic algorithm finds a better plan because of row estimate issues. Only real testing shows what actually works, so don’t assume DP is always best.

Sleigh the season with the most personal gift around. Get them a Cameo video!

This might be the most challenging part of Postgres.

I have a Postgres example: same estimated cost, but one query is instant and the other could take forever. Want to see?

My course combines theory with hands-on practice, focusing on three key areas: the storage engine, the optimizer (the deepest and most unique part of the course), and how extensions work. If you complete the course and spend plenty of time practicing with my trace tool, and you

Is there a book written from a DSA perspective? Books like that are by true experts.

You could be here by Friday for a weekend of sun salutations

Why do Postgres internals books feel like something beautiful but all wrapped up and untouchable? It’s much more fun to really dig in and see how things actually work!

These articles are good, but they’re not really for regular users. If you want to become a true master, I’m pretty sure you’ll need three things: trace analysis for hands-on practice, sometimes by reading the source code, research papers to build your theory, and AI tools to