It looks like we've hit a severe regression in dask-expr: A number of queries returns wrong results or does not even finish anymore.
First occurrence: #1341
Note
There's a genuine issue with the benchmarks in there as well:
asyncio.exceptions.TimeoutError: Waited for 4 worker(s) to reconnect after restarting, but after 120s, only 0 have returned. Consider a longer timeout, or `wait_for_workers=False`.
cc @phofl
Tracebacks
FAILED tests/tpch/test_correctness.py::test_dask_results[12] - AssertionError: DataFrame are different
DataFrame shape mismatch
[left]: (7, 3)
[right]: (2, 3)
FAILED tests/tpch/test_optimization.py::test_optimization[18] - KeyError: False
FAILED tests/tpch/test_dask.py::test_query_18 - KeyError: False
FAILED tests/tpch/test_correctness.py::test_dask_results[13] - Failed: Timeout >3600.0s
FAILED tests/tpch/test_correctness.py::test_dask_results[16] - AssertionError: DataFrame are different
DataFrame shape mismatch
[left]: (114161, 4)
[right]: (18314, 4)
FAILED tests/tpch/test_correctness.py::test_dask_results[17] - AssertionError: DataFrame.iloc[:, 0] (column name="avg_yearly") are different
DataFrame.iloc[:, 0] (column name="avg_yearly") values are different (100.0 %)
[index]: [0]
[left]: [33330456.87]
[right]: [348406.0542857143]
At positional index 0, first diff: 33330456.87 != 348406.0542857143
FAILED tests/tpch/test_correctness.py::test_dask_results[18] - KeyError: False
FAILED tests/tpch/test_correctness.py::test_dask_results[21] - AssertionError: DataFrame are different
DataFrame shape mismatch
[left]: (0, 2)
[right]: (100, 2)
FAILED tests/tpch/test_correctness.py::test_dask_results[22] - AssertionError: DataFrame.iloc[:, 1] (column name="numcust") are different
DataFrame.iloc[:, 1] (column name="numcust") values are different (100.0 %)
[index]: [0, 1, 2, 3, 4, 5, 6]
[left]: [2680, 2642, 2779, 2697, 2835, 2653, 2739]
[right]: [888, 861, 964, 892, 948, 909, 922]
At positional index 0, first diff: 2680 != 888
ERROR tests/tpch/test_correctness.py::test_dask_results[6] - asyncio.exceptions.TimeoutError: Waited for 4 worker(s) to reconnect after restarting, but after 120s, only 0 have returned. Consider a longer timeout, or `wait_for_workers=False`.
It looks like we've hit a severe regression in
dask-expr
: A number of queries returns wrong results or does not even finish anymore.First occurrence: #1341
Note There's a genuine issue with the benchmarks in there as well:
cc @phofl
Tracebacks
https://github.com/coiled/benchmarks/actions/runs/7763077035/job/21174629659#step:12:1567
1342:
https://github.com/coiled/benchmarks/actions/runs/7770352324/job/21190438203#step:12:1536
1343:
https://github.com/coiled/benchmarks/actions/runs/7763077035/job/21174629659#step:12:1532