pytest parametrize id collision in xdist: how to fix

TL;DR

Give every parametrized case a unique id with pytest.param(..., id=...) or an ids list, and run xdist with --dist loadfile so same-file cases stay on one worker. Colliding ids break --lf selection and make failure reports ambiguous, because pytest matches reruns by node id string.

Problem

Two or more cases in a @pytest.mark.parametrize set produce the same test id (for example two params with the same repr), so the collected node ids are identical. Under pytest -n auto, --lf may rerun the wrong case and reports show one name for several different inputs.

Steps

1. Reproduce the collision: run pytest --collect-only -q and look for duplicate node ids such as test_login[user0] appearing twice.

Expected: you see the same id string listed more than once.

1. Assign explicit unique ids with pytest.param:

   import pytest

   @pytest.mark.parametrize(
       "user",
       [
           pytest.param("ann", id="ann-first"),
           pytest.param("ann", id="ann-second"),
           pytest.param("bob", id="bob"),
       ],
   )
   def test_login(user):
       ...

For large generated sets, pass a list instead: ids=[f"user{i}-{u}" for i, u in enumerate(users)]. Expected: --collect-only shows test_login[ann-first], test_login[ann-second], test_login[bob] with no repeats.

1. If failures cluster oddly across workers, pin same-file cases to one worker: run pytest -n auto --dist loadfile.

Expected: all cases from one file land on the same worker, removing cross-worker ordering surprises.

1. Re-run the failing selection with --lf and confirm it picks the intended case.

Expected: only the previously failed unique id is rerun.

When to use

• Duplicate ids in --collect-only -q output.
• --lf or --deselect picks the wrong parametrized case under -n.
• Failure reports where you cannot tell which parameter value failed.

When not to use

• You just want prettier ids (use ids= freely, no xdist needed).
• The collision is between two different test functions (rename one).
• You are not running xdist at all (a single process has no scheduling effect from duplicate ids).

Tool compatibility

• pytest 8.x with pytest-xdist 3.x (-n auto, --dist loadfile).
• The ids list and pytest.param(id=...) API is stable since pytest 4, so this works on pytest 7 too.

Variant phrasings

pytest duplicate test ids with -n workers

Same root cause: non-unique ids confuse worker scheduling and rerun filters. Fix with unique ids as above.

xdist --lf reruns the wrong parametrized case

--lf matches by node id string, so duplicates resolve to the first match. Unique ids fix it.

Why it happens

pytest builds node ids from the repr of parameter values; two equal (or equal-repr) values produce the same id. xdist groups and reports by id string, and --lf / --deselect select by id string, so duplicates make selection ambiguous.

Edge cases

• Non-string params with identical reprs (for example two dicts that print the same): always use explicit id= via pytest.param.
• Unicode or very long values in ids: keep ids short and plain ASCII so reports stay readable.
• --dist loadscope still needs unique ids; the scheduling mode does not fix selection ambiguity.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_ny2GGdQKSkugp2ToNaGEJQ