diff --git a/.github/workflows/tests.yml b/.github/workflows/tests.yml
index 307a615f1f..194f1e64e3 100644
--- a/.github/workflows/tests.yml
+++ b/.github/workflows/tests.yml
@@ -4,7 +4,7 @@ on:
     branches:
       - main
     tags:
-      - '*'
+      - "*"
   pull_request:
   schedule:
     # Runs "At 00:01" (see https://crontab.guru)
@@ -22,96 +22,8 @@ defaults:
     shell: bash -l {0}
 
 jobs:
-  software:
-    name: Setup
-    runs-on: ubuntu-latest
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.8", "3.9", "3.10"]
-
-    steps:
-      - uses: actions/checkout@v2
-        with:
-          fetch-depth: 0
-
-      - name: Set up environment
-        uses: conda-incubator/setup-miniconda@v2
-        with:
-          miniforge-variant: Mambaforge
-          use-mamba: true
-          condarc-file: ci/condarc
-          python-version: ${{ matrix.python-version }}
-          environment-file: ci/environment.yml
-
-      - name: Check upstream
-        run: |
-          if [[ ${{ github.event_name }} == "pull_request" ]]
-          then
-              REF="HEAD^2"
-          else
-              REF="HEAD"
-          fi
-
-          COMMIT="$(git log -n 1 --pretty=format:%s $REF)"
-          if [[ "$COMMIT" == *"test-upstream"* || ${{ github.event_name }} == "schedule" ]]
-          then
-            TEST_UPSTREAM="true"
-            AB_BASELINE="coiled-upstream-py3.9 coiled-0.1.0-py3.9"
-          else
-            TEST_UPSTREAM="false"
-            AB_BASELINE="coiled-latest-py3.9 coiled-0.1.0-py3.9"
-          fi
-
-          # Put TEST_UPSTREAM into $GITHUB_ENV so it can be used in subsequent workflow steps
-          echo $TEST_UPSTREAM
-          echo TEST_UPSTREAM=$TEST_UPSTREAM >> $GITHUB_ENV
-
-          # Put env variables into files so it can be downloaded in subsequent workflow jobs
-          echo $TEST_UPSTREAM > test_upstream.txt
-          echo $AB_BASELINE > ab_baseline.txt
-
-      - name: Build Coiled Software Environment
-        env:
-          DASK_COILED__TOKEN: ${{ secrets.COILED_BENCHMARK_BOT_TOKEN }}
-        run: |
-          PYTHON_VERSION_FORMATTED=$(echo "${{ matrix.python-version }}" | sed 's/\.//g' )
-          NAME_HEAD=dask-engineering/coiled-runtime-${{ github.event_name }}
-          NAME_TAIL=$GITHUB_RUN_ID-py$PYTHON_VERSION_FORMATTED
-          if [[ ${{ github.event_name }} = 'pull_request' ]]
-          then
-            NAME_MID=${{ github.event.number }}
-          else
-            NAME_MID=$GITHUB_REF_TYPE-$(echo "$GITHUB_REF_NAME" | sed 's/\./-/g' )
-          fi
-          # env name can only contain lowercase ASCII letters, numbers, hyphens and underscores
-          COILED_SOFTWARE_NAME=$NAME_HEAD-$(echo $NAME_MID-$NAME_TAIL | tr 'A-Z' 'a-z' | sed -r 's/[^a-z0-9_-]/_/g')
-
-          # Create conda environment.yaml file for the latest software environment
-          python ci/create_latest_runtime_meta.py
-          ENV_FILE=coiled_software_environment.yaml
-          cat $ENV_FILE
-
-          mamba install coiled
-          echo "Creating Coiled software environment for $COILED_SOFTWARE_NAME"
-          coiled env create --name $COILED_SOFTWARE_NAME --conda $ENV_FILE
-
-          # Put COILED_SOFTWARE_NAME into a file so it can be downloaded in subsequent workflow jobs
-          echo $COILED_SOFTWARE_NAME > software_name.txt
-
-      - name: Upload environment file
-        uses: actions/upload-artifact@v3
-        with:
-          name: software-environment-py${{ matrix.python-version }}
-          path: |
-            coiled_software_environment.yaml
-            software_name.txt
-            test_upstream.txt
-            ab_baseline.txt
-
   runtime:
     name: Runtime - ${{ matrix.os }}, Python ${{ matrix.python-version }}, Runtime ${{ matrix.runtime-version }}
-    needs: software
     runs-on: ${{ matrix.os }}
     timeout-minutes: 120
     strategy:
@@ -119,7 +31,7 @@ jobs:
       matrix:
         os: ["ubuntu-latest"]
         python-version: ["3.9"]
-        runtime-version: ["latest", "0.0.4", "0.1.0"]
+        runtime-version: ["upstream", "latest", "0.0.4", "0.1.0"]
 
     steps:
       - uses: actions/checkout@v2
@@ -135,16 +47,12 @@ jobs:
           python-version: ${{ matrix.python-version }}
           environment-file: ci/environment.yml
 
-      - name: Download software environment assets
-        if: matrix.runtime-version == 'latest'
-        uses: actions/download-artifact@v3
-        with:
-          name: software-environment-py${{ matrix.python-version }}
-
       - name: Install coiled-runtime
         env:
           COILED_RUNTIME_VERSION: ${{ matrix.runtime-version }}
-        run: source ci/scripts/install_coiled_runtime.sh
+        run: |
+          python ci/create_runtime_meta.py
+          source ci/scripts/install_coiled_runtime.sh
 
       - name: Run Coiled Runtime Tests
         id: test
@@ -166,7 +74,6 @@ jobs:
 
   benchmarks:
     name: Benchmarks - ${{ matrix.os }}, Python ${{ matrix.python-version }}, Runtime ${{ matrix.runtime-version }}
-    needs: software
     runs-on: ${{ matrix.os }}
     timeout-minutes: 120
     strategy:
@@ -174,7 +81,7 @@ jobs:
       matrix:
         os: ["ubuntu-latest"]
         python-version: ["3.9"]
-        runtime-version: ["latest", "0.0.4", "0.1.0"]
+        runtime-version: ["upstream", "latest", "0.0.4", "0.1.0"]
 
     steps:
       - uses: actions/checkout@v2
@@ -190,16 +97,12 @@ jobs:
           python-version: ${{ matrix.python-version }}
           environment-file: ci/environment.yml
 
-      - name: Download software environment assets
-        if: matrix.runtime-version == 'latest'
-        uses: actions/download-artifact@v3
-        with:
-          name: software-environment-py${{ matrix.python-version }}
-
       - name: Install coiled-runtime
         env:
           COILED_RUNTIME_VERSION: ${{ matrix.runtime-version }}
-        run: source ci/scripts/install_coiled_runtime.sh
+        run: |
+          python ci/create_runtime_meta.py
+          source ci/scripts/install_coiled_runtime.sh
 
       - name: Run benchmarking tests
         id: benchmarking_tests
@@ -221,7 +124,6 @@ jobs:
 
   stability:
     name: Stability - ${{ matrix.os }}, Python ${{ matrix.python-version }}, Runtime ${{ matrix.runtime-version }}
-    needs: software
     runs-on: ${{ matrix.os }}
     timeout-minutes: 120
     strategy:
@@ -229,7 +131,7 @@ jobs:
       matrix:
         os: ["ubuntu-latest"]
         python-version: ["3.8", "3.9", "3.10"]
-        runtime-version: ["latest", "0.0.4", "0.1.0"]
+        runtime-version: ["upstream", "latest", "0.0.4", "0.1.0"]
         include:
           - python-version: "3.9"
             runtime-version: "latest"
@@ -252,16 +154,12 @@ jobs:
           python-version: ${{ matrix.python-version }}
           environment-file: ci/environment.yml
 
-      - name: Download software environment assets
-        if: matrix.runtime-version == 'latest'
-        uses: actions/download-artifact@v3
-        with:
-          name: software-environment-py${{ matrix.python-version }}
-
       - name: Install coiled-runtime
         env:
           COILED_RUNTIME_VERSION: ${{ matrix.runtime-version }}
-        run: source ci/scripts/install_coiled_runtime.sh
+        run: |
+          python ci/create_runtime_meta.py
+          source ci/scripts/install_coiled_runtime.sh
 
       - name: Run stability tests
         id: stability_tests
@@ -282,39 +180,6 @@ jobs:
           name: stability-${{ matrix.os }}-${{ matrix.runtime-version }}-py${{ matrix.python-version }}
           path: stability-${{ matrix.os }}-${{ matrix.runtime-version }}-py${{ matrix.python-version }}.db
 
-  cleanup:
-    needs: [software, runtime, benchmarks, stability]
-    if: always()
-    name: Cleanup
-    runs-on: ubuntu-latest
-    strategy:
-      fail-fast: false
-      matrix:
-        python-version: ["3.8", "3.9", "3.10"]
-    steps:
-      - uses: actions/checkout@v2
-
-      - name: Set up Python
-        uses: actions/setup-python@v4
-        with:
-          python-version: ${{ matrix.python-version }}
-
-      - name: Install coiled
-        run: python -m pip install coiled
-
-      - name: Download software environment assets
-        uses: actions/download-artifact@v3
-        with:
-          name: software-environment-py${{ matrix.python-version }}
-
-      - name: Remove Coiled software environment
-        env:
-          DASK_COILED__TOKEN: ${{ secrets.COILED_BENCHMARK_BOT_TOKEN }}
-        run: |
-          SOFTWARE_NAME=$(cat software_name.txt)
-          echo "Deleting $SOFTWARE_NAME"
-          coiled env delete $SOFTWARE_NAME
-
   process-results:
     needs: [runtime, benchmarks, stability]
     name: Combine separate benchmark results
@@ -341,7 +206,7 @@ jobs:
         env:
           AWS_ACCESS_KEY_ID: ${{ secrets.RUNTIME_CI_BOT_AWS_ACCESS_KEY_ID }}
           AWS_SECRET_ACCESS_KEY: ${{ secrets.RUNTIME_CI_BOT_AWS_SECRET_ACCESS_KEY }}
-          AWS_DEFAULT_REGION: us-east-2  # this is needed for boto for some reason
+          AWS_DEFAULT_REGION: us-east-2 # this is needed for boto for some reason
           DB_NAME: benchmark.db
         run: |
           aws s3 cp s3://coiled-runtime-ci/benchmarks/$DB_NAME . || true
@@ -356,7 +221,7 @@ jobs:
         env:
           AWS_ACCESS_KEY_ID: ${{ secrets.RUNTIME_CI_BOT_AWS_ACCESS_KEY_ID }}
           AWS_SECRET_ACCESS_KEY: ${{ secrets.RUNTIME_CI_BOT_AWS_SECRET_ACCESS_KEY }}
-          AWS_DEFAULT_REGION: us-east-2  # this is needed for boto for some reason
+          AWS_DEFAULT_REGION: us-east-2 # this is needed for boto for some reason
           DB_NAME: benchmark.db
         run: |
           aws s3 cp $DB_NAME s3://coiled-runtime-ci/benchmarks/
@@ -364,7 +229,7 @@ jobs:
       - name: Upload benchmark results as artifact
         uses: actions/upload-artifact@v3
         with:
-          name: benchmark.db
+          name: benchmark
           path: benchmark.db
 
   regressions:
@@ -380,7 +245,7 @@ jobs:
 
       - uses: actions/download-artifact@v3
         with:
-          name: benchmark.db
+          name: benchmark
 
       - name: Set up environment
         uses: conda-incubator/setup-miniconda@v2
@@ -451,7 +316,7 @@ jobs:
 
       - uses: actions/download-artifact@v3
         with:
-          name: benchmark.db
+          name: benchmark
 
       - name: Set up environment
         uses: conda-incubator/setup-miniconda@v2
@@ -461,13 +326,8 @@ jobs:
           python-version: "3.9"
           environment-file: ci/environment-dashboard.yml
 
-      - name: Download software environment assets
-        uses: actions/download-artifact@v3
-        with:
-          name: software-environment-py3.9
-
       - name: Generate dashboards
-        run: python dashboard.py -d benchmark.db -o static -b $(cat ab_baseline.txt)
+        run: python dashboard.py -d benchmark.db -o static -b coiled-latest-py3.9 coiled-upstream-py3.9 
 
       - name: Upload artifact
         uses: actions/upload-artifact@v3
diff --git a/README.md b/README.md
index 84c5cc6779..100ca01131 100644
--- a/README.md
+++ b/README.md
@@ -57,10 +57,11 @@ The `coiled-runtime` test suite can be run locally with the following steps:
    the Coiled Dask Engineering AWS S3 account.
 2. Create a Python environment and install development dependencies as
    specified in `ci/environment.yml`.
-3. (Optional) If testing against an unreleased version of `coiled-runtime`,
-   create a Coiled software environment with the unreleased `coiled-runtime` installed
-   and set a local `COILED_SOFTWARE_NAME` environment variable to the name
-   of the software environment (e.g. `export COILED_SOFTWARE_NAME="account/software-name"`)
+3. Install a coiled runtime environment. This might be from one of the environments
+   listed in ``environments/``, or it could be a development environment if you are
+   testing feature branches of dask or distributed. This test suite is configured
+   to run Coiled's ``package_sync`` feature, so your local environment should be copied
+   to the cluster.
 4. Run tests with `python -m pytest tests`
 
 Additionally, tests are automatically run on pull requests to this repository.
diff --git a/alembic/versions/2764a4f5582b_declare_bankruptcy_for_cluster_startup_.py b/alembic/versions/2764a4f5582b_declare_bankruptcy_for_cluster_startup_.py
new file mode 100644
index 0000000000..1f3df6c7c5
--- /dev/null
+++ b/alembic/versions/2764a4f5582b_declare_bankruptcy_for_cluster_startup_.py
@@ -0,0 +1,30 @@
+"""Declare bankruptcy for cluster startup time
+
+Revision ID: 2764a4f5582b
+Revises: 924e9b1430e1
+Create Date: 2022-09-14 11:45:46.024184
+
+"""
+from alembic import op
+import sqlalchemy as sa
+
+
+# revision identifiers, used by Alembic.
+revision = "2764a4f5582b"
+down_revision = "924e9b1430e1"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.execute(
+        """
+        delete from test_run
+        where originalname = 'test_default_cluster_spinup_time'
+        and path = 'benchmarks/test_coiled.py';
+        """
+    )
+
+
+def downgrade() -> None:
+    pass
diff --git a/ci/create_latest_runtime_meta.py b/ci/create_runtime_meta.py
similarity index 75%
rename from ci/create_latest_runtime_meta.py
rename to ci/create_runtime_meta.py
index 4a7fa2680e..d9fbf7fe9c 100644
--- a/ci/create_latest_runtime_meta.py
+++ b/ci/create_runtime_meta.py
@@ -6,7 +6,6 @@
 import shlex
 import subprocess
 import sys
-from distutils.util import strtobool
 
 import yaml
 from jinja2 import Environment, FileSystemLoader, select_autoescape
@@ -38,15 +37,20 @@ def main():
         if package_name == "python":
             requirements[idx] = f"python =={python_version}"
 
-    # Optionally use the development version of `dask` and `distributed`
-    # from `dask/label/dev` conda channel
-    upstream = strtobool(os.environ.get("TEST_UPSTREAM", "false"))
-    if upstream:
-        upstream_packages = {"dask", "distributed"}
-        for idx, req in enumerate(requirements):
-            package_name = Requirement(req).name
-            if package_name in upstream_packages:
-                requirements[idx] = get_latest_conda_build(package_name)
+    if os.environ.get("COILED_RUNTIME_VERSION", "unknown") == "upstream":
+        requirements = [
+            r
+            for r in requirements
+            if Requirement(r).name not in {"dask", "distributed"}
+        ]
+        requirements.append(
+            {
+                "pip": [
+                    "git+https://github.com/dask/dask@main",
+                    "git+https://github.com/dask/distributed@main",
+                ]
+            }
+        )
 
     # File compatible with `mamba env create --file <...>`
     env = {
diff --git a/ci/scripts/combine-dbs.sh b/ci/scripts/combine-dbs.sh
index 459fe2f86c..976e43ea42 100644
--- a/ci/scripts/combine-dbs.sh
+++ b/ci/scripts/combine-dbs.sh
@@ -16,6 +16,11 @@ DB_NAME=benchmark.tmp.db alembic upgrade head
 # Merge in the individual job dbs into our working copy
 for FILE in $(find benchmarks -name "*.db")
 do
+  # Skip the output DB if we see it
+  if [ ${FILE##*/} == $DB_NAME ]; then
+    echo "Skipping $FILE"
+    continue
+  fi
   echo "Processing $FILE"
   # Copy the individual table into the primary one. We make an intermediate
   # temp table so that we can null out the primary keys and reset the
diff --git a/ci/scripts/install_coiled_runtime.sh b/ci/scripts/install_coiled_runtime.sh
index a98a155bb4..a1044279cc 100644
--- a/ci/scripts/install_coiled_runtime.sh
+++ b/ci/scripts/install_coiled_runtime.sh
@@ -5,7 +5,7 @@ set -o errexit
 set -o nounset
 set -o xtrace
 
-if [[ "$COILED_RUNTIME_VERSION" =~ latest|AB_ ]]
+if [[ "$COILED_RUNTIME_VERSION" =~ upstream|latest|AB_ ]]
 then
   cat coiled_software_environment.yaml
   mamba env update --file coiled_software_environment.yaml
@@ -15,4 +15,4 @@ fi
 
 # For debugging
 echo -e "--\n--Conda Environment (re-create this with \`conda env create --name <name> -f <output_file>\`)\n--"
-mamba env export | grep -E -v '^prefix:.*$'
\ No newline at end of file
+mamba env export | grep -E -v '^prefix:.*$'
diff --git a/ci/scripts/run_tests.sh b/ci/scripts/run_tests.sh
index b6676e1887..6b3b53ecaf 100644
--- a/ci/scripts/run_tests.sh
+++ b/ci/scripts/run_tests.sh
@@ -5,19 +5,11 @@ set -o xtrace
 BENCHMARK="${BENCHMARK:-false}"
 
 # Ensure we run additional tests when testing the latest coiled-runtime
-if [[ "$COILED_RUNTIME_VERSION" = 'latest' ]]
+if [[ $COILED_RUNTIME_VERSION = 'latest' || $COILED_RUNTIME_VERSION = 'upstream' ]]
 then
   EXTRA_OPTIONS="--run-latest"
-  export COILED_SOFTWARE_NAME=$(cat software_name.txt)
-  export TEST_UPSTREAM=$(cat test_upstream.txt)
-elif [[ "$COILED_RUNTIME_VERSION" =~ AB_ ]]
-then
-  EXTRA_OPTIONS=""
-  export COILED_SOFTWARE_NAME=$(cat software_name.txt)
-  export TEST_UPSTREAM=$(cat test_upstream.txt)
 else
   EXTRA_OPTIONS=""
-  unset COILED_SOFTWARE_NAME
 fi
 if [[ $BENCHMARK = 'true' ]]
 then
diff --git a/conftest.py b/conftest.py
index c574252630..3611aa0005 100644
--- a/conftest.py
+++ b/conftest.py
@@ -65,8 +65,6 @@ def pytest_collection_modifyitems(config, items):
 
 
 def get_coiled_runtime_version():
-    if strtobool(os.environ.get("TEST_UPSTREAM", "false")):
-        return "upstream"
     try:
         return os.environ["COILED_RUNTIME_VERSION"]
     except KeyError:
@@ -78,37 +76,13 @@ def get_coiled_runtime_version():
         if runtime_info:
             return runtime_info[0]["version"]
         else:
-            return "latest"
+            return "unknown"
 
 
-def get_coiled_software_name():
-    try:
-        return os.environ["COILED_SOFTWARE_NAME"]
-    except KeyError:
-        # Determine software environment from local `coiled-runtime` version (in installed)
-        out = subprocess.check_output(
-            shlex.split("conda list --json coiled-runtime"), text=True
-        ).rstrip()
-        runtime_info = json.loads(out)
-        if runtime_info:
-            version = runtime_info[0]["version"].replace(".", "-")
-            py_version = f"{sys.version_info[0]}{sys.version_info[1]}"
-            return f"coiled/coiled-runtime-{version}-py{py_version}"
-        else:
-            raise RuntimeError(
-                "Must either specific `COILED_SOFTWARE_NAME` environment variable "
-                "or have `coiled-runtime` installed"
-            )
-
-
-dask.config.set(
-    {
-        "coiled.account": "dask-engineering",
-        "coiled.software": get_coiled_software_name(),
-    }
-)
+dask.config.set({"coiled.account": "dask-engineering"})
 
 COILED_RUNTIME_VERSION = get_coiled_runtime_version()
+COILED_SOFTWARE_NAME = "package_sync"
 
 
 # ############################################### #
@@ -213,7 +187,7 @@ def test_run_benchmark(benchmark_db_session, request, testrun_uid):
             dask_version=dask.__version__,
             distributed_version=distributed.__version__,
             coiled_runtime_version=COILED_RUNTIME_VERSION,
-            coiled_software_name=dask.config.get("coiled.software"),
+            coiled_software_name=COILED_SOFTWARE_NAME,
             python_version=".".join(map(str, sys.version_info)),
             platform=sys.platform,
             ci_run_url=WORKFLOW_URL,
@@ -441,6 +415,7 @@ def small_cluster(request):
         worker_vm_types=["t3.large"],  # 2CPU, 8GiB
         scheduler_vm_types=["t3.large"],
         backend_options=backend_options,
+        package_sync=True,
     ) as cluster:
         yield cluster
 
diff --git a/tests/benchmarks/test_coiled.py b/tests/benchmarks/test_coiled.py
index dc6b064229..bb39d95fa5 100644
--- a/tests/benchmarks/test_coiled.py
+++ b/tests/benchmarks/test_coiled.py
@@ -5,5 +5,7 @@
 
 def test_default_cluster_spinup_time(request, auto_benchmark_time):
 
-    with Cluster(name=f"{request.node.originalname}-{uuid.uuid4().hex[:8]}"):
+    with Cluster(
+        name=f"{request.node.originalname}-{uuid.uuid4().hex[:8]}", package_sync=True
+    ):
         pass
diff --git a/tests/benchmarks/test_parquet.py b/tests/benchmarks/test_parquet.py
index 5e0d499371..feeaa7666b 100644
--- a/tests/benchmarks/test_parquet.py
+++ b/tests/benchmarks/test_parquet.py
@@ -21,6 +21,7 @@ def parquet_cluster():
         n_workers=N_WORKERS,
         worker_vm_types=["m5.xlarge"],
         scheduler_vm_types=["m5.xlarge"],
+        package_sync=True,
     ) as cluster:
         yield cluster
 
diff --git a/tests/runtime/test_build.py b/tests/runtime/test_build.py
index 6c7b964af9..96ff838b26 100644
--- a/tests/runtime/test_build.py
+++ b/tests/runtime/test_build.py
@@ -7,10 +7,8 @@
 import shlex
 import subprocess
 import sys
-from distutils.util import strtobool
 
 import coiled
-import dask
 import pytest
 import yaml
 from distributed import Client
@@ -77,8 +75,10 @@ def test_install_dist():
     # Test that versions of packages installed are consistent with those
     # specified in `meta.yaml`
 
-    if Version(dask.__version__).local or strtobool(
-        os.environ.get("TEST_UPSTREAM", "false")
+    if os.environ.get("COILED_RUNTIME_VERSION", "unknown") in (
+        "upstream",
+        "latest",
+        "unknown",
     ):
         pytest.skip("Not valid on upstream build")
 
diff --git a/tests/stability/test_deadlock.py b/tests/stability/test_deadlock.py
index ff72db95ca..31c07b30a5 100644
--- a/tests/stability/test_deadlock.py
+++ b/tests/stability/test_deadlock.py
@@ -17,8 +17,10 @@ def test_repeated_merge_spill(upload_cluster_dump, benchmark_all):
     with Cluster(
         name=f"test_deadlock-{uuid.uuid4().hex}",
         n_workers=20,
-        worker_vm_types=["t3.medium"],
+        worker_vm_types=["t3.large"],
+        scheduler_vm_types=["t3.large"],
         wait_for_workers=True,
+        package_sync=True,
     ) as cluster:
         with Client(cluster) as client:
             with upload_cluster_dump(client, cluster), benchmark_all(client):
diff --git a/tests/stability/test_spill.py b/tests/stability/test_spill.py
index 45825a5f17..e7a10ba1a5 100644
--- a/tests/stability/test_spill.py
+++ b/tests/stability/test_spill.py
@@ -6,42 +6,15 @@
 from dask.distributed import Client, wait
 
 
-@pytest.mark.stability
-@pytest.mark.parametrize("keep_around", (True, False))
-def test_spilling(keep_around, upload_cluster_dump, benchmark_all):
+@pytest.fixture(scope="module")
+def spill_cluster():
     with Cluster(
-        name=f"test_spill-{uuid.uuid4().hex}",
+        f"spill-{uuid.uuid4().hex[:8]}",
         n_workers=5,
-        worker_disk_size=55,
-        worker_vm_types=["t3.medium"],
-        wait_for_workers=True,
-        environ={
-            # Note: We set allowed-failures to ensure that no tasks are not retried
-            #  upon ungraceful shutdown behavior during adaptive scaling
-            #  but we receive a KilledWorker() instead.
-            "DASK_DISTRIBUTED__SCHEDULER__ALLOWED_FAILURES": "0"
-        },
-    ) as cluster:
-        with Client(cluster) as client:
-            with upload_cluster_dump(client, cluster), benchmark_all(client):
-                arr = da.random.random((200, 2**27)).persist()  # 200 GiB
-                wait(arr)
-                fut = client.compute(arr.sum())
-                if not keep_around:
-                    del arr
-                wait(fut)
-
-
-@pytest.mark.skip(
-    reason="Skip until https://github.com/coiled/feedback/issues/185 is resolved."
-)
-@pytest.mark.stability
-def test_tensordot_stress(upload_cluster_dump, benchmark_all):
-    with Cluster(
-        name=f"test_spill-{uuid.uuid4().hex}",
-        n_workers=5,
-        worker_disk_size=55,
-        worker_vm_types=["t3.medium"],
+        package_sync=True,
+        worker_disk_size=64,
+        worker_vm_types=["t3.large"],
+        scheduler_vm_types=["t3.large"],
         wait_for_workers=True,
         environ={
             # Note: We set allowed-failures to ensure that no tasks are not retried
@@ -55,10 +28,37 @@ def test_tensordot_stress(upload_cluster_dump, benchmark_all):
             "DASK_DISTRIBUTED__WORKER__CONNECTIONS__OUTGOING": "1",
         },
     ) as cluster:
-        with Client(cluster) as client:
-            with upload_cluster_dump(client, cluster), benchmark_all(client):
-                a = da.random.random((48 * 1024, 48 * 1024))  # 18 GiB
-                b = (a @ a.T).sum().round(3)
-                fut = client.compute(b)
-                wait(fut)
-                assert fut.result()
+        yield cluster
+
+
+@pytest.fixture
+def spill_client(spill_cluster, upload_cluster_dump, benchmark_all):
+    with Client(spill_cluster) as client:
+        spill_cluster.scale(5)
+        client.wait_for_workers(5)
+        client.restart()
+        with upload_cluster_dump(client, spill_cluster), benchmark_all(client):
+            yield client
+
+
+@pytest.mark.stability
+@pytest.mark.parametrize("keep_around", (True, False))
+def test_spilling(spill_client, keep_around):
+    arr = da.random.random((64, 2**27)).persist()  # 64 GiB, ~2x memory
+    wait(arr)
+    fut = spill_client.compute(arr.sum())
+    if not keep_around:
+        del arr
+    wait(fut)
+
+
+@pytest.mark.skip(
+    reason="Skip until https://github.com/coiled/feedback/issues/185 is resolved."
+)
+@pytest.mark.stability
+def test_tensordot_stress(spill_client):
+    a = da.random.random((48 * 1024, 48 * 1024))  # 18 GiB
+    b = (a @ a.T).sum().round(3)
+    fut = spill_client.compute(b)
+    wait(fut)
+    assert fut.result()