diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
index 6b0fd420b582..25b01b25a065 100644
--- a/.github/workflows/pr.yaml
+++ b/.github/workflows/pr.yaml
@@ -480,17 +480,6 @@ jobs:
       matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
       build_type: pull-request
       script: "ci/test_wheel_cudf_polars.sh"
-  wheel-tests-cudf-polars-with-rapidsmpf:
-    needs: [wheel-build-cudf-polars, changed-files]
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@main
-    if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_wheels && fromJSON(needs.changed-files.outputs.changed_file_groups).neither_cudf_nor_dask_cudf
-    with:
-      # This selects "ARCH=amd64 + the latest supported Python + CUDA" to minimize CI usage.
-      # (rapidsmpf compatibility already validated in rapidsmpf CI)
-      matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
-      build_type: pull-request
-      script: "ci/test_cudf_polars_with_rapidsmpf.sh"
   cudf-polars-polars-tests:
     needs: [wheel-build-cudf-polars, changed-files]
     secrets: inherit
diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
index 51ddac09d4df..b4f672c61f9c 100644
--- a/.github/workflows/test.yaml
+++ b/.github/workflows/test.yaml
@@ -153,19 +153,6 @@ jobs:
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
       script: "ci/test_wheel_cudf_polars.sh"
-  wheel-tests-cudf-polars-with-rapidsmpf:
-    secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@main
-    with:
-      # This selects "ARCH=amd64 + the latest supported Python + CUDA" to minimize CI usage.
-      # (rapidsmpf compatibility already validated in rapidsmpf CI)
-      matrix_filter: map(select(.ARCH == "amd64")) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
-      build_type: ${{ inputs.build_type }}
-      branch: ${{ inputs.branch }}
-      date: ${{ inputs.date }}
-      sha: ${{ inputs.sha }}
-      script: "ci/test_cudf_polars_with_rapidsmpf.sh"
-      continue-on-error: true
   cudf-polars-polars-tests:
     secrets: inherit
     uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@main
diff --git a/ci/run_cudf_polars_pytests.sh b/ci/run_cudf_polars_pytests.sh
index ce30428675e8..a1b06fd2fc56 100755
--- a/ci/run_cudf_polars_pytests.sh
+++ b/ci/run_cudf_polars_pytests.sh
@@ -1,5 +1,5 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION.
 # SPDX-License-Identifier: Apache-2.0
 
 set -euo pipefail
@@ -9,17 +9,32 @@ set -euo pipefail
 # Support invoking run_cudf_polars_pytests.sh outside the script directory
 cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../python/cudf_polars/
 
-# Test the "in-memory" executor
+echo "Test the in-memory executor"
 python -m pytest --cache-clear "$@" tests --executor in-memory
 
-# Test the default "streaming" executor
+echo "Test the default streaming executor"
 python -m pytest --cache-clear "$@" tests --executor streaming
 
-# Test the "streaming" executor with small blocksize
+echo "Test the streaming executor with small blocksize"
 python -m pytest --cache-clear "$@" tests --executor streaming --blocksize-mode small
 
-# Run experimental tests with Distributed cluster
+echo "Test the (future default) streaming executor with rapidsmpf"
+CUDF_POLARS__PARQUET_OPTIONS__USE_RAPIDSMPF_NATIVE=1 CUDF_POLARS__EXECUTOR__SHUFFLE_METHOD=rapidsmpf python -m pytest --cache-clear "$@" tests \
+    --executor streaming \
+    --blocksize-mode small \
+    --cluster single \
+    --runtime rapidsmpf \
+    --cov-fail-under=0  # TODO? missing coverage may be due to old task-based paths
+
+echo "Run experimental tests with Distributed cluster"
 python -m pytest --cache-clear "$@" "tests/experimental" \
     --executor streaming \
     --cluster distributed \
     --cov-fail-under=0  # No code-coverage requirement for these tests.
+
+echo "Run experimental tests with the distributed cluster mode and the rapidsmpf runtime"
+python -m pytest --cache-clear "$@" "tests/experimental" \
+    --executor streaming \
+    --cluster distributed \
+    --runtime rapidsmpf \
+    --cov-fail-under=0  # No code-coverage requirement for these tests(?)
diff --git a/ci/run_cudf_polars_with_rapidsmpf_pytests.sh b/ci/run_cudf_polars_with_rapidsmpf_pytests.sh
deleted file mode 100755
index b172995a214b..000000000000
--- a/ci/run_cudf_polars_with_rapidsmpf_pytests.sh
+++ /dev/null
@@ -1,27 +0,0 @@
-#!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION.
-# SPDX-License-Identifier: Apache-2.0
-
-set -euo pipefail
-
-# Test cudf_polars with rapidsmpf integration
-# This script runs experimental tests with single cluster mode and the rapidsmpf runtime
-
-# It is essential to cd into python/cudf_polars as `pytest-xdist` + `coverage` seem to work only at this directory level.
-
-# Support invoking run_cudf_polars_with_rapidsmpf_pytests.sh outside the script directory
-cd "$(dirname "$(realpath "${BASH_SOURCE[0]}")")"/../python/cudf_polars/
-
-# Run experimental tests with the "single" cluster mode and the "rapidsmpf" runtime
-rapids-logger "Running experimental tests with the 'rapidsmpf' runtime and a 'single' cluster"
-timeout 10m python -m pytest --cache-clear "$@" "tests/experimental" \
-    --executor streaming \
-    --cluster single \
-    --runtime rapidsmpf
-
-# Run experimental tests with the "distributed" cluster mode and the "rapidsmpf" runtime
-rapids-logger "Running experimental tests with the 'rapidsmpf' runtime and a 'distributed' cluster"
-timeout 10m python -m pytest --cache-clear "$@" "tests/experimental" \
-    --executor streaming \
-    --cluster distributed \
-    --runtime rapidsmpf
diff --git a/ci/test_cudf_polars_with_rapidsmpf.sh b/ci/test_cudf_polars_with_rapidsmpf.sh
deleted file mode 100755
index 6e9ad45afe3d..000000000000
--- a/ci/test_cudf_polars_with_rapidsmpf.sh
+++ /dev/null
@@ -1,60 +0,0 @@
-#!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION.
-# SPDX-License-Identifier: Apache-2.0
-
-set -euo pipefail
-
-source rapids-init-pip
-
-rapids-logger "Download wheels"
-
-RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
-CUDF_POLARS_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="cudf_polars_${RAPIDS_PY_CUDA_SUFFIX}" RAPIDS_PY_WHEEL_PURE="1" rapids-download-wheels-from-github python)
-LIBCUDF_WHEELHOUSE=$(RAPIDS_PY_WHEEL_NAME="libcudf_${RAPIDS_PY_CUDA_SUFFIX}" rapids-download-wheels-from-github cpp)
-PYLIBCUDF_WHEELHOUSE=$(rapids-download-from-github "$(rapids-package-name "wheel_python" pylibcudf --stable --cuda "$RAPIDS_CUDA_VERSION")")
-
-rapids-logger "Installing cudf_polars and its dependencies (including rapidsmpf)"
-
-rapids-pip-retry install \
-    -v \
-    --constraint "${PIP_CONSTRAINT}" \
-    "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test,experimental,rapidsmpf]" \
-    "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
-    "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
-
-rapids-logger "Run cudf_polars tests with rapidsmpf"
-
-# Get the latest polars version for testing
-POLARS_VERSION=$(python ci/utils/fetch_polars_versions.py --latest-patch-only dependencies.yaml | awk '{print $NF}')
-
-rapids-logger "Installing polars==${POLARS_VERSION}"
-pip install -U "polars==${POLARS_VERSION}"
-
-# shellcheck disable=SC2317
-function set_exitcode()
-{
-    EXITCODE=$?
-}
-EXITCODE=0
-trap set_exitcode ERR
-set +e
-
-rapids-logger "Running cudf_polars tests with rapidsmpf"
-
-# Run cudf_polars tests with rapidsmpf using dedicated test runner
-timeout 15m ./ci/run_cudf_polars_with_rapidsmpf_pytests.sh \
-    --no-cov \
-    --numprocesses=8 \
-    --dist=worksteal \
-    -v \
-    --junitxml="${RAPIDS_TESTS_DIR}/junit-cudf-polars-rapidsmpf.xml"
-
-trap - ERR
-set -e
-
-if [ ${EXITCODE} != 0 ]; then
-    rapids-logger "cudf_polars + rapidsmpf tests FAILED: exitcode ${EXITCODE}"
-else
-    rapids-logger "cudf_polars + rapidsmpf tests PASSED"
-fi
-exit ${EXITCODE}
diff --git a/ci/test_wheel_cudf_polars.sh b/ci/test_wheel_cudf_polars.sh
index abdaeebd1659..5865468802b6 100755
--- a/ci/test_wheel_cudf_polars.sh
+++ b/ci/test_wheel_cudf_polars.sh
@@ -30,7 +30,7 @@ rapids-pip-retry install \
     -v \
     --constraint ./constraints.txt \
     --constraint "${PIP_CONSTRAINT}" \
-    "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test,experimental]" \
+    "$(echo "${CUDF_POLARS_WHEELHOUSE}"/cudf_polars_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)[test,experimental,rapidsmpf]" \
     "$(echo "${LIBCUDF_WHEELHOUSE}"/libcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)" \
     "$(echo "${PYLIBCUDF_WHEELHOUSE}"/pylibcudf_"${RAPIDS_PY_CUDA_SUFFIX}"*.whl)"
 
diff --git a/dependencies.yaml b/dependencies.yaml
index 70409ea10a14..5dc5eaccb8f1 100644
--- a/dependencies.yaml
+++ b/dependencies.yaml
@@ -115,6 +115,8 @@ files:
       - depends_on_cudf_kafka
       - depends_on_custreamz
       - depends_on_cudf_polars
+      - depends_on_rapidsmpf
+      - cudf_polars_trace
   test_java:
     output: none
     includes:
diff --git a/python/cudf_polars/cudf_polars/experimental/dask_registers.py b/python/cudf_polars/cudf_polars/experimental/dask_registers.py
index 15895e209db0..94334ccbc57f 100644
--- a/python/cudf_polars/cudf_polars/experimental/dask_registers.py
+++ b/python/cudf_polars/cudf_polars/experimental/dask_registers.py
@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES.
 # SPDX-License-Identifier: Apache-2.0
 
 """Dask function registrations such as serializers and dispatch implementations."""
@@ -215,7 +215,7 @@ def _(x: DataFrame) -> int:
         from rapidsmpf.integrations.dask.spilling import register_dask_serialize
 
         register_dask_serialize()  # pragma: no cover; rapidsmpf dependency not included yet
-    except ImportError:
+    except ImportError:  # pragma: no cover
         pass
 
     # Register the tokenizer for NamedExpr and DataType. This is a performance
diff --git a/python/cudf_polars/tests/conftest.py b/python/cudf_polars/tests/conftest.py
index 1a94eb038e13..de923c60cefd 100644
--- a/python/cudf_polars/tests/conftest.py
+++ b/python/cudf_polars/tests/conftest.py
@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION & AFFILIATES.
 # SPDX-License-Identifier: Apache-2.0
 from __future__ import annotations
 
@@ -28,6 +28,11 @@ def clear_memory_resource_cache():
     cudf_polars.callback.default_memory_resource.cache_clear()
 
 
+@pytest.fixture
+def using_rapidsmpf():
+    return cudf_polars.testing.asserts.DEFAULT_RUNTIME == "rapidsmpf"
+
+
 def pytest_addoption(parser):
     parser.addoption(
         "--executor",
diff --git a/python/cudf_polars/tests/experimental/rapidsmpf/test_tracing.py b/python/cudf_polars/tests/experimental/rapidsmpf/test_tracing.py
index 5818609b9a39..1caef3c63f82 100644
--- a/python/cudf_polars/tests/experimental/rapidsmpf/test_tracing.py
+++ b/python/cudf_polars/tests/experimental/rapidsmpf/test_tracing.py
@@ -20,6 +20,7 @@
 @pytest.mark.skipif(DEFAULT_CLUSTER != "single", reason="Requires 'single' cluster.")
 def test_structlog_streaming_node_events():
     """Test that structlog emits 'Streaming Actor' events when tracing is enabled."""
+    pytest.importorskip("structlog")
     # Run in subprocess to control CUDF_POLARS_LOG_TRACES environment variable
     code = textwrap.dedent("""\
     import polars as pl
diff --git a/python/cudf_polars/tests/expressions/test_agg.py b/python/cudf_polars/tests/expressions/test_agg.py
index 5ad6f68cc2ab..e10e1969f9b6 100644
--- a/python/cudf_polars/tests/expressions/test_agg.py
+++ b/python/cudf_polars/tests/expressions/test_agg.py
@@ -194,7 +194,13 @@ def test_implode_agg_unsupported():
     assert_ir_translation_raises(q, NotImplementedError)
 
 
-def test_decimal_aggs(decimal_df: pl.LazyFrame) -> None:
+def test_decimal_aggs(decimal_df: pl.LazyFrame, using_rapidsmpf, request) -> None:
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=using_rapidsmpf,
+            reason="https://github.com/rapidsai/cudf/issues/21645",
+        )
+    )
     q = decimal_df.with_columns(
         sum=pl.col("a").sum(),
         min=pl.col("a").min(),
diff --git a/python/cudf_polars/tests/expressions/test_len.py b/python/cudf_polars/tests/expressions/test_len.py
index ccc903aa99d8..10efadd879a0 100644
--- a/python/cudf_polars/tests/expressions/test_len.py
+++ b/python/cudf_polars/tests/expressions/test_len.py
@@ -30,7 +30,13 @@ def test_len(dtype, empty):
 
 
 @pytest.mark.parametrize("data", [[1, 2, 3], [1, 2, None]])
-def test_col_len(data):
+def test_col_len(data, using_rapidsmpf, request):
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=using_rapidsmpf,
+            reason="https://github.com/rapidsai/cudf/issues/21645",
+        )
+    )
     data = {"a": list("xyz"), "b": data}
     q = pl.LazyFrame(data).select(
         pl.col("a").len().alias("l"),
diff --git a/python/cudf_polars/tests/test_config.py b/python/cudf_polars/tests/test_config.py
index 523f62622827..d369818552ee 100644
--- a/python/cudf_polars/tests/test_config.py
+++ b/python/cudf_polars/tests/test_config.py
@@ -3,6 +3,7 @@
 
 from __future__ import annotations
 
+import os
 import sys
 from typing import Any, cast
 
@@ -381,8 +382,12 @@ def test_scheduler_deprecated() -> None:
 
 def test_validate_shuffle_method_defaults(
     *,
+    monkeypatch: pytest.MonkeyPatch,
     rapidsmpf_distributed_available: bool,
 ) -> None:
+    # TODO: Can remove once removed from ci/run_cudf_polars_pytests.sh
+    # and the default shuffle method changes to "rapidsmpf"
+    monkeypatch.delenv("CUDF_POLARS__EXECUTOR__SHUFFLE_METHOD", raising=False)
     config = ConfigOptions.from_polars_engine(
         pl.GPUEngine(
             executor="streaming",
@@ -804,7 +809,12 @@ def test_cuda_stream_policy_from_env(
     elif runtime == "rapidsmpf":
         with pytest.raises(ValueError, match="The rapidsmpf streaming engine"):
             ConfigOptions.from_polars_engine(engine)
-    else:
+    elif (
+        os.environ.get("CUDF_POLARS__EXECUTOR__SHUFFLE_METHOD", "") != "rapidsmpf"
+        and not rapidsmpf_single_available
+    ):
+        # CUDF_POLARS__EXECUTOR__SHUFFLE_METHOD set in ci/run_cudf_polars_pytests.sh
+        # Would raise since rapidsmpf shuffle requires rapidsmpf to be installed
         config = ConfigOptions.from_polars_engine(engine)
         assert config.cuda_stream_policy == env
 
diff --git a/python/cudf_polars/tests/test_dataframescan.py b/python/cudf_polars/tests/test_dataframescan.py
index e002820fe383..2986c58b5fef 100644
--- a/python/cudf_polars/tests/test_dataframescan.py
+++ b/python/cudf_polars/tests/test_dataframescan.py
@@ -86,7 +86,13 @@ def test_dataframescan_with_decimals():
     POLARS_VERSION_LT_138,
     reason="height parameter added in Polars 1.38",
 )
-def test_dataframescan_zero_width_with_rows():
+def test_dataframescan_zero_width_with_rows(request, using_rapidsmpf):
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf,
+            reason="https://github.com/rapidsai/cudf/issues/21644",
+        )
+    )
     df = pl.LazyFrame(height=5)
     q = df.select(pl.len())
     assert_gpu_result_equal(q)
diff --git a/python/cudf_polars/tests/test_drop_nulls.py b/python/cudf_polars/tests/test_drop_nulls.py
index 0fe9b963eddb..56da63763b5b 100644
--- a/python/cudf_polars/tests/test_drop_nulls.py
+++ b/python/cudf_polars/tests/test_drop_nulls.py
@@ -42,7 +42,15 @@ def test_drop_null(null_data):
     [0, pl.col("a").mean(), pl.col("b")],
     ids=["scalar", "aggregation", "column_expression"],
 )
-def test_fill_null(null_data, value):
+def test_fill_null(null_data, value, using_rapidsmpf, request):
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=using_rapidsmpf
+            and "mean" in str(value)
+            and len(null_data.collect()) == 2,
+            reason="https://github.com/rapidsai/cudf/issues/21721",
+        )
+    )
     q = null_data.select(pl.col("a").fill_null(value))
     assert_gpu_result_equal(q)
 
diff --git a/python/cudf_polars/tests/test_groupby.py b/python/cudf_polars/tests/test_groupby.py
index 8713185f56f5..e9400e22dbd1 100644
--- a/python/cudf_polars/tests/test_groupby.py
+++ b/python/cudf_polars/tests/test_groupby.py
@@ -17,6 +17,8 @@
 )
 from cudf_polars.utils.versions import (
     POLARS_VERSION_LT_132,
+    POLARS_VERSION_LT_133,
+    POLARS_VERSION_LT_134,
     POLARS_VERSION_LT_136,
     POLARS_VERSION_LT_1321,
 )
@@ -126,7 +128,53 @@ def maintain_order(request):
     return request.param
 
 
-def test_groupby(df: pl.LazyFrame, maintain_order, keys, exprs):
+def test_groupby(
+    df: pl.LazyFrame, maintain_order, keys, exprs, using_rapidsmpf, request
+):
+    failing_rapidsmpf_nodeids = {
+        'test_groupby[maintain_order-col("key1")-col("int32").mean()]',
+        'test_groupby[maintain_order-col("key1")-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-col("key2")-col("int32").mean()]',
+        'test_groupby[maintain_order-col("key2")-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-col("key1")-dyn int: 1-col("int32").mean()]',
+        'test_groupby[maintain_order-col("key1")-dyn int: 1-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-[(col("key1")) * (col("key2"))]-col("int32").mean()]',
+        'test_groupby[maintain_order-[(col("key1")) * (col("key2"))]-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-col("key1")-col("key2")-col("int32").mean()]',
+        'test_groupby[maintain_order-col("key1")-col("key2")-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-[(col("key1")) == (col("key2"))]-col("int32").mean()]',
+        'test_groupby[maintain_order-[(col("key1")) == (col("key2"))]-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+        'test_groupby[maintain_order-col("key2")-[(col("key1")) == (1)]-col("int32").mean()]',
+        'test_groupby[maintain_order-col("key2")-[(col("key1")) == (1)]-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]',
+    }
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf and request.node.name in failing_rapidsmpf_nodeids,
+            reason="https://github.com/rapidsai/cudf/issues/21642 and probably mean calculation referenced in https://github.com/rapidsai/cudf/issues/21721",
+        )
+    )
+    failing_rapidsmpf_nodeids_lt_1_36 = {
+        'test_groupby[maintain_order-col("key2")-[(col("key1")) == (dyn int: 1.strict_cast(Int64))]-col("uint16_with_null").sum()-col("uint16_with_null").mean().alias("mean")]'
+    }
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf
+            and POLARS_VERSION_LT_136
+            and request.node.name in failing_rapidsmpf_nodeids_lt_1_36,
+            reason="Type mismatch in columns to concatenate.",
+        )
+    )
+    segfaulting_rapidsmpf_nodeids = {
+        'test_groupby[maintain_order-col("key2")-[(col("key1")) == (dyn int: 1.strict_cast(Int64))]-col("int32").mean()]'
+    }
+    if (
+        using_rapidsmpf
+        and POLARS_VERSION_LT_136
+        and request.node.name in segfaulting_rapidsmpf_nodeids
+    ):
+        pytest.skip(
+            "Usually raises 'Type mismatch in columns to concatenate' but can also segfault."
+        )
     q = df.group_by(*keys, maintain_order=maintain_order).agg(*exprs)
 
     if not maintain_order:
@@ -136,7 +184,14 @@ def test_groupby(df: pl.LazyFrame, maintain_order, keys, exprs):
     assert_gpu_result_equal(q, check_exact=False)
 
 
-def test_groupby_sorted_keys(df: pl.LazyFrame, keys, exprs):
+def test_groupby_sorted_keys(df: pl.LazyFrame, keys, exprs, using_rapidsmpf, request):
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf,
+            strict=False,
+            reason="https://github.com/rapidsai/cudf/issues/21642 -  no deterministic sort for keys",
+        )
+    )
     sorted_keys = [
         key.sort(descending=descending)
         for key, descending in zip(keys, itertools.cycle([False, True]))
@@ -382,7 +437,7 @@ def test_groupby_sum_all_null_group_returns_null():
     ids=["sum", "mean", "median", "quantile-0.5"],
 )
 def test_groupby_aggs_keep_unsupported_as_null(
-    request, df: pl.LazyFrame, agg_expr
+    request, df: pl.LazyFrame, agg_expr, using_rapidsmpf
 ) -> None:
     request.applymarker(
         pytest.mark.xfail(
@@ -390,6 +445,15 @@ def test_groupby_aggs_keep_unsupported_as_null(
             reason="polars raises now",
         )
     )
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=using_rapidsmpf
+            and "quantile" in str(agg_expr)
+            and POLARS_VERSION_LT_134
+            and not POLARS_VERSION_LT_133,
+            reason="decimal precision is 38 instead of 9",
+        )
+    )
     lf = df.filter(pl.col("datetime") == date(2004, 12, 1))
     q = lf.group_by("datetime").agg(agg_expr)
     assert_gpu_result_equal(q)
diff --git a/python/cudf_polars/tests/test_join.py b/python/cudf_polars/tests/test_join.py
index fc1a384957ea..a891d669c901 100644
--- a/python/cudf_polars/tests/test_join.py
+++ b/python/cudf_polars/tests/test_join.py
@@ -70,7 +70,16 @@ def test_join_maintain_order(left, right, maintain_order):
         ["c", "a"],
     ],
 )
-def test_non_coalesce_join(left, right, how, nulls_equal, join_expr):
+def test_non_coalesce_join(
+    left, right, how, nulls_equal, join_expr, using_rapidsmpf, request
+):
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf,
+            strict=False,
+            reason="Non deterministic sort/join on nulls",
+        )
+    )
     query = left.join(
         right, on=join_expr, how=how, nulls_equal=nulls_equal, coalesce=False
     )
diff --git a/python/cudf_polars/tests/test_merge_sorted.py b/python/cudf_polars/tests/test_merge_sorted.py
index 66c3a759f02e..ab654d8445eb 100644
--- a/python/cudf_polars/tests/test_merge_sorted.py
+++ b/python/cudf_polars/tests/test_merge_sorted.py
@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026, NVIDIA CORPORATION & AFFILIATES.
 # SPDX-License-Identifier: Apache-2.0
 from __future__ import annotations
 
@@ -9,17 +9,14 @@
 from cudf_polars.testing.asserts import assert_gpu_result_equal
 
 
-@pytest.mark.parametrize(
-    "descending",
-    [
-        pytest.param(
-            True,
-            marks=pytest.mark.xfail(reason="polars/issues/21511"),
-        ),
-        False,
-    ],
-)
-def test_merge_sorted_without_nulls(descending):
+@pytest.mark.parametrize("descending", [True, False])
+def test_merge_sorted_without_nulls(descending, request, using_rapidsmpf):
+    request.applymarker(
+        pytest.mark.xfail(
+            not using_rapidsmpf and descending,
+            reason="https://github.com/pola-rs/polars/issues/21511",
+        )
+    )
     df0 = pl.LazyFrame(
         {"name": ["steve", "elise", "bob"], "age": [42, 44, 18], "height": [5, 6, 5]}
     ).sort("age", descending=descending)
diff --git a/python/cudf_polars/tests/test_rolling.py b/python/cudf_polars/tests/test_rolling.py
index a9a0617ff719..792d83134e88 100644
--- a/python/cudf_polars/tests/test_rolling.py
+++ b/python/cudf_polars/tests/test_rolling.py
@@ -3,6 +3,7 @@
 
 from __future__ import annotations
 
+import contextlib
 from datetime import datetime
 
 import pytest
@@ -133,7 +134,7 @@ def test_grouped_rolling():
     assert_gpu_result_equal(q)
 
 
-def test_grouped_rolling_unsorted_raises():
+def test_grouped_rolling_unsorted_raises(using_rapidsmpf):
     df = pl.LazyFrame(
         {
             "keys": [1, None, 2, 1, 2, None],
@@ -145,7 +146,19 @@ def test_grouped_rolling_unsorted_raises():
 
     with pytest.raises(pl.exceptions.ComputeError):
         q.collect(engine="in-memory")
-    with pytest.raises(RuntimeError, match="Input for grouped rolling is not sorted"):
+
+    fallback_warning = (
+        pytest.warns(
+            UserWarning,
+            match="shuffle_method=rapidsmpf-single does not support maintain_order=True",
+        )
+        if using_rapidsmpf
+        else contextlib.nullcontext()
+    )
+    with (
+        pytest.raises(RuntimeError, match="Input for grouped rolling is not sorted"),
+        fallback_warning,
+    ):
         q.collect(engine=pl.GPUEngine(raise_on_fail=True))
 
 
diff --git a/python/cudf_polars/tests/test_scan.py b/python/cudf_polars/tests/test_scan.py
index f2ab23468472..66f74b9bce12 100644
--- a/python/cudf_polars/tests/test_scan.py
+++ b/python/cudf_polars/tests/test_scan.py
@@ -702,7 +702,13 @@ def test_scan_tiny_file_not_compressed(tmp_path):
     reason="height parameter added in Polars 1.38",
 )
 @pytest.mark.parametrize("engine", [None, NO_CHUNK_ENGINE])
-def test_scan_parquet_zero_width_with_limit(tmp_path, engine):
+def test_scan_parquet_zero_width_with_limit(tmp_path, engine, request, using_rapidsmpf):
+    request.applymarker(
+        pytest.mark.xfail(
+            using_rapidsmpf and engine is None,
+            reason="https://github.com/rapidsai/cudf/issues/21644",
+        )
+    )
     path = tmp_path / "zero_width.parquet"
     pl.LazyFrame(height=20).sink_parquet(path)
     q = pl.scan_parquet(path).head(5)
diff --git a/python/cudf_polars/tests/test_select.py b/python/cudf_polars/tests/test_select.py
index ac9dbb90d0c0..7f85e233de34 100644
--- a/python/cudf_polars/tests/test_select.py
+++ b/python/cudf_polars/tests/test_select.py
@@ -87,7 +87,13 @@ def test_select_with_cse_no_agg():
     assert_gpu_result_equal(query)
 
 
-def test_select_with_cse_with_agg():
+def test_select_with_cse_with_agg(using_rapidsmpf, request):
+    request.applymarker(
+        pytest.mark.xfail(
+            condition=using_rapidsmpf,
+            reason="https://github.com/rapidsai/cudf/issues/21645",
+        )
+    )
     df = pl.LazyFrame({"a": [1, 2, 3]})
     expr = pl.col("a") + pl.col("a")
     asum = pl.col("a").sum() + pl.col("a").sum()
diff --git a/python/cudf_polars/tests/test_tracing.py b/python/cudf_polars/tests/test_tracing.py
index f67fcda89a45..0cc6ed298d90 100644
--- a/python/cudf_polars/tests/test_tracing.py
+++ b/python/cudf_polars/tests/test_tracing.py
@@ -111,9 +111,9 @@ def test_log_query_plan() -> None:
         [sys.executable, "-c", code], env=env, stderr=subprocess.STDOUT
     )
 
-    # Check for Query Plan event
+    # Check for Query Plan event generated from SerializablePlan
     assert b"Query Plan" in result
     assert b"scope=plan" in result or b"'scope': 'plan'" in result
-    assert b"ir_id" in result
-    assert b"ir_type" in result
-    assert b"children_ir_ids" in result
+    assert b"actor_ir_id" in result
+    assert b"actor_ir_type" in result
+    assert b"children" in result