[BUG] extract_domain is extremely slow on DuckDB 1.3.2

[ctoprimer]

Describe the bug

netquack’s extract_domain() performance appears to degrade sharply on DuckDB v1.3.2. Even on simple synthetic emails in-memory, runtime grows to ~12.7s for 10k rows, which is unexpectedly slow for basic domain extraction.

Number of Domains	Time to Run (avg of 5 runs)
100	0.133s
1,000	1.332s
10,000	12.700s

To Reproduce

Steps to reproduce the behavior (pure SQL, DuckDB CLI):

1. Start DuckDB v1.3.2 (any environment is fine; the issue reproduces in :memory:).

2. Enable timing in the DuckDB CLI:

   .timer on

3. Install and load the extension:

   SET allow_community_extensions = true;
   INSTALL netquack FROM community;
   LOAD netquack;

4. Confirm DuckDB version:

   SELECT version() AS duckdb_version;

   (Optional) If netquack exposes a version function, please advise the canonical query (e.g. SELECT * FROM netquack_version();).

5. Run the benchmark queries and record the timings shown by .timer on:

   100 rows

   SELECT
     count(*) AS n,
     sum(length(NULLIF(extract_domain(email), ''))) AS total_domain_chars
   FROM (
     SELECT printf('user%08d@sub%d.domain%d.example.com', i, i % 10, i % 100) AS email
     FROM range(100) t(i)
   ) s;

   1,000 rows

   SELECT
     count(*) AS n,
     sum(length(NULLIF(extract_domain(email), ''))) AS total_domain_chars
   FROM (
     SELECT printf('user%08d@sub%d.domain%d.example.com', i, i % 10, i % 100) AS email
     FROM range(1000) t(i)
   ) s;

   10,000 rows

   SELECT
     count(*) AS n,
     sum(length(NULLIF(extract_domain(email), ''))) AS total_domain_chars
   FROM (
     SELECT printf('user%08d@sub%d.domain%d.example.com', i, i % 10, i % 100) AS email
     FROM range(10000) t(i)
   ) s;

6. Observe that runtime increases substantially with row count (e.g., ~12s at 10k rows on DuckDB v1.3.2).

Expected behavior

Domain extraction from simple emails should be much faster (sub-second at 10k rows), especially in-memory on synthetic data. Current performance makes extract_domain() a bottleneck in downstream dbt models.

Versions
Netquack Version: 6a0d651
DuckDB Version: 1.3.2

Additional context

• Reproduces in :memory: and does not rely on file IO.
• Query forces evaluation by aggregating the extract_domain() output.
• Timings reported above were measured from the same workload using DuckDB v1.3.2 (Python package), averaged over 5 runs after warm-up.

[hatamiarash7]

Hi @ctoprimer

Yes, we have a netquack_version() function that helps you with the extension's version. Check it here.

Can you tell me about your hardware or running environment? Because I can't see any performance issue with ~10K records.

Of course, as expected, there is a slowdown at higher values ​​(for example, 10 million records), but it is still not a large number (for example, here it only took 4 seconds).

---

Also, give me a profile output, please.

EXPLAIN ANALYZE SELECT
  count(*) AS n,
  sum(length(NULLIF(extract_domain(email), ''))) AS total_domain_chars
FROM (
  SELECT printf('user%08d@sub%d.domain%d.example.com', i, i % 10, i % 100) AS email
  FROM range(10000) t(i)
) s;

[hatamiarash7]

Do you have any updates? Is it solved on your side?