CTE stands for Common Table Expressions in SQL. They are temporary result sets defined within a single query using the WITH clause, acting like named subqueries to improve readability and structure.
Eh, almost every link on the homepage has an initialism or acronym in the title, and roughly none of them are actually defining the term they're using. Indeed, not to point fingers, your own submissions make the same mistake.
Sure, yes, OP should (and now has) defined the term. But at the same time it's reasonable to expect that someone reading a blog post on BoringSQL.com would already know the term just as much as we could expect people interested in Clojure would know what a REPL is.
I've always thought of CTEs as a code organisation tool, not an optimisation tool. The fact the some rdbms treats them as an optimisation fence was a bug, not a feature.
> Recursive CTEs use an iterative working-table mechanism. Despite the name, they aren't truly recursive. PostgreSQL doesn't "call itself" by creating a nested stack of unfinished queries.
If you want something that is more like actual recursion (I.e., depth-first), Oracle has CONNECT BY which does not require the same kind of tracking. It also comes with extra features to help with cycle detection, stack depth reflection, etc.
If your problem is aligned with the DFS model, the oracle technique can run circles around recursive CTEs. Anything with a deep hierarchy and early termination conditions is a compelling candidate.
Great article, I always like to structure my queries with CTEs and I was (wrongly) assuming it all gets inlined at the end.
Sometimes it also gets complicated since these intermediate results can't be easily seen in a SQL editor. I was working on a UI to parse CTE queries and then execute them step by step to show the results of all the CTEs for easier understanding of the query (as part of this project https://github.com/sqg-dev/sqg/)
To the author of the article. This was a really nice and educating read. You made me finally understand recursive CTEs, the org chart was a really good example. Thank you.
Not to sound like a broken record, but I need to echo what many other replies have already said: great article.
Embarassingly, despite thinking of myself as pretty knowledgeable with SQL, I had no idea you could nest DML statements inside a CTE. I always assumed/used DML statements as the final statement after a CTE was defined. I'm not sure if or when I might use this in the future, but it's neat to learn something new (and to be humbled at the same time).
Sometimes easy performance trick is to split the CTE to separate queries, put the results to unlogged temporary tables and add whatever indexes the next step needs.
Obviously makes only sense for stuff like analytical queries that are not running constantly.
Kind of a weird flex - you're basically saying that you aren't any better than the Indian teams at writing efficient queries or optimizing existing ones, the only talent you've demonstrated is ability to ask Claude Code. It's like you are announcing your own obsolescence.
I mostly use CTEs for organization these days, and in rare cases to express queries which cannot be written without them.
These days I often write queries like this (especially when doing exploratory ad hoc queries, but also in apps) even when it's not necessary to use a CTE:
WITH
a AS (
SELECT ... FROM ... WHERE ... etc.
)
SELECT * FROM a
The first CTE query defines the input, and the main query just reads from it. Subsequent subqueries invoke steps on that input to group, filter, join, and so on.
This has a bunch of nice benefits. For example, it allows me to add steps incrementally, and to "comment out" a step I can simply change the next step's input to read from the preceding step. Each step can be read and understood in isolation.
I work a lot with Postgres, ClickHouse, and SQLite, and generally find that the database inlines and optimizes CTEs, and challenges mostly concern performance traps (like IN or EXISTS) that allly to non-CTE situations as well.
17 comments
[ 2.7 ms ] story [ 45.8 ms ] threadCTE stands for Common Table Expressions in SQL. They are temporary result sets defined within a single query using the WITH clause, acting like named subqueries to improve readability and structure.
Sure, yes, OP should (and now has) defined the term. But at the same time it's reasonable to expect that someone reading a blog post on BoringSQL.com would already know the term just as much as we could expect people interested in Clojure would know what a REPL is.
If you want something that is more like actual recursion (I.e., depth-first), Oracle has CONNECT BY which does not require the same kind of tracking. It also comes with extra features to help with cycle detection, stack depth reflection, etc.
If your problem is aligned with the DFS model, the oracle technique can run circles around recursive CTEs. Anything with a deep hierarchy and early termination conditions is a compelling candidate.
Regarding recursive CTEs, you might be interested in how DuckDb evolved them with USING KEY: https://duckdb.org/2025/05/23/using-key
Embarassingly, despite thinking of myself as pretty knowledgeable with SQL, I had no idea you could nest DML statements inside a CTE. I always assumed/used DML statements as the final statement after a CTE was defined. I'm not sure if or when I might use this in the future, but it's neat to learn something new (and to be humbled at the same time).
Obviously makes only sense for stuff like analytical queries that are not running constantly.
I'll write some nice clean CTEs and then have to refactor it to temp tables due to the lack of materialization.
I just rewrote all queries with claude code and next day and compute decreased to 1/5th.
These days I often write queries like this (especially when doing exploratory ad hoc queries, but also in apps) even when it's not necessary to use a CTE:
The first CTE query defines the input, and the main query just reads from it. Subsequent subqueries invoke steps on that input to group, filter, join, and so on.This has a bunch of nice benefits. For example, it allows me to add steps incrementally, and to "comment out" a step I can simply change the next step's input to read from the preceding step. Each step can be read and understood in isolation.
I work a lot with Postgres, ClickHouse, and SQLite, and generally find that the database inlines and optimizes CTEs, and challenges mostly concern performance traps (like IN or EXISTS) that allly to non-CTE situations as well.