Postgres from the inside · part 3
Vacuum, seen from the page
Dead tuples, the free space map and why VACUUM almost never gives disk space back.
the notebook of imke vos
Notes from the machine room.
I’m Imke, a backend engineer in Rotterdam. This is where I write down what I learn about databases, build tools and the bugs in between, usually with a terminal still open.
now: reading the Postgres source, one page at a time. What else
Latest posts
The archive-
Reading a B-tree page by hand
Open an index with pageinspect, walk from the root to a single row and see what an index actually promises.
-
The flaky test was a time zone
A test that only failed after 5 pm, and the one-line date parse behind it.
-
What actually happens on INSERT
From the parser to the write-ahead log: following one row into a Postgres table.
-
A build system in 200 lines: the graph
Topological order, cycles that explain themselves and running independent steps in parallel.
-
Make your errors boring
A good error says what happened, where, and what to do next. Everything else is decoration.
Series
All series3 parts · 27 min
Postgres from the inside
A slow walk through the storage layer, with pageinspect open the whole time: heap pages, B-tree indexes and vacuum.
- 01What actually happens on INSERT
- 02Reading a B-tree page by hand
- 03Vacuum, seen from the page
3 parts · 23 min
A build system in 200 lines
Writing a small, correct build tool in Rust, one idea per post: why, content hashing and the dependency graph.
- 01why bother
- 02hashing inputs
- 03the graph
2 parts · 16 min
CSS for backend people
The parts of CSS I wish someone had explained to me in terms I already knew: precedence rules and layout algorithms.
- 01the cascade
- 02layout without tears
newsletter
New posts, by email
One email when a post goes up, about twice a month. Plain text, no tracking pixels, one-click unsubscribe. Prefer a feed reader? There’s RSS.