From 4de9b01f87f96715ccae19427529a7e39638514e Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 16:33:25 +0200
Subject: [PATCH 1/9] Runs the SDK E2E tests concurrently per test case (rather
 than per test file)

---
 .github/workflows/check-sdk-tests.yml | 87 +++++++--------------------
 1 file changed, 23 insertions(+), 64 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index 632f86b..99fac81 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -91,7 +91,7 @@ jobs:
     if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
     runs-on: ubuntu-latest
     outputs:
-      test-files: ${{ steps.get-tests.outputs.test-files }}
+      test-cases: ${{ steps.get-tests.outputs.test-cases }}
     steps:
       - name: Research preparation
         working-directory: ${{ github.workspace }}
@@ -110,11 +110,26 @@ jobs:
       - name: Install dependencies
         run: sudo apt-get install -y jq
 
-      - name: Find e2e test files
+      - name: Find e2e test cases
         id: get-tests
         run: |
-          test_files=$(find ${{ github.workspace }}/bittensor/tests/e2e_tests -name "test*.py" | jq -R -s -c 'split("\n") | map(select(. != ""))')
-          echo "test-files=$test_files" >> $GITHUB_OUTPUT
+          # Emit one matrix entry per individual test function (as a `path::test_name`
+          # node id) rather than per file, so each test gets its own concurrent job.
+          # The e2e suite has no test classes, so grepping top-level `def test_*` is
+          # sufficient and keeps this job dependency-free (no pytest collection needed).
+          # Node ids are relative to the bittensor repo root so they resolve inside the
+          # prebuilt image (where bittensor lives at /opt/bittensor), not the workspace.
+          cd "${{ github.workspace }}/bittensor"
+          test_cases=$(
+            while IFS= read -r f; do
+              grep -oE '^(async[[:space:]]+)?def[[:space:]]+test_[A-Za-z0-9_]+' "$f" \
+                | grep -oE 'test_[A-Za-z0-9_]+' \
+                | while IFS= read -r name; do printf '%s::%s\n' "$f" "$name"; done
+            done < <(find tests/e2e_tests -name "test*.py") \
+              | jq -R -s -c 'split("\n") | map(select(. != ""))'
+          )
+          echo "Collected $(echo "$test_cases" | jq 'length') test cases"
+          echo "test-cases=$test_cases" >> $GITHUB_OUTPUT
         shell: bash
 
   pull-docker-image:
@@ -152,13 +167,7 @@ jobs:
       fail-fast: false
       max-parallel: 16
       matrix:
-        rust-branch:
-          - stable
-        rust-target:
-          - x86_64-unknown-linux-gnu
-        os:
-          - ubuntu-latest
-        test-file: ${{ fromJson(needs.find-e2e-tests.outputs.test-files) }}
+        test-case: ${{ fromJson(needs.find-e2e-tests.outputs.test-cases) }}
 
     env:
       RELEASE_NAME: development
@@ -168,59 +177,8 @@ jobs:
       TARGET: ${{ matrix.rust-target }}
 
     timeout-minutes: 60
-    name: "e2e tests: ${{ matrix.test-file }}"
+    name: "e2e tests: ${{ matrix.test-case }}"
     steps:
-      - name: Check-out repository
-        uses: actions/checkout@v6
-
-      - name: Install system dependencies
-        run: |
-          sudo apt-get update &&
-          sudo apt-get install -y clang curl libssl-dev llvm libudev-dev protobuf-compiler
-
-      - name: Set up Python
-        uses: actions/setup-python@v6
-        with:
-          python-version: '3.13'
-
-      - name: Install uv
-        uses: astral-sh/setup-uv@v4
-
-      - name: Clone Bittensor SDK repo
-        run: git clone https://github.com/latent-to/bittensor.git
-
-      - name: Checkout Bittensor branch
-        working-directory: ${{ github.workspace }}/bittensor
-        run: |
-          if ! git fetch origin $BITTENSOR_BRANCH; then
-            echo "❌ Error: Branch '$BITTENSOR_BRANCH' does not exist in latent-to/bittensor."
-            exit 1
-          fi
-          git checkout FETCH_HEAD
-          echo "✅ Using Bittensor branch: $BITTENSOR_BRANCH"
-
-      - name: Install Bittensor SDK dependencies
-        working-directory: ${{ github.workspace }}/bittensor
-        run: uv pip install --system '.[dev]'
-
-      - name: Clone async-substrate-interface repo
-        run: git clone https://github.com/latent-to/async-substrate-interface.git
-
-      - name: Checkout PR branch in async-substrate-interface
-        working-directory: ${{ github.workspace }}/async-substrate-interface
-        run: |
-          BRANCH="${{ github.event.pull_request.head.ref || github.ref_name }}"
-          REPO_URL="${{ github.event.pull_request.head.repo.clone_url || 'https://github.com/latent-to/async-substrate-interface.git' }}"
-          git fetch $REPO_URL $BRANCH
-          git checkout FETCH_HEAD
-          echo "Current SHA: $(git rev-parse HEAD)"
-
-      - name: Install async-substrate-interface with dev dependencies
-        working-directory: ${{ github.workspace }}/async-substrate-interface
-        run: |
-          uv pip uninstall --system async-substrate-interface || true
-          uv pip install --system '.[dev]'
-
       - name: Download Cached Docker Image
         uses: actions/download-artifact@v8
         with:
@@ -230,7 +188,8 @@ jobs:
         run: docker load -i subtensor-localnet.tar
 
       - name: Run e2e tests
-        run: pytest ${{ matrix.test-file }} -s
+        working-directory: /opt/bittensor
+        run: pytest "${{ matrix.test-case }}" -s
 
 
   run-integration-and-unit-test:

From 6ca701138cfb5d6a2016db98fb6feb879b5e02f1 Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 16:33:50 +0200
Subject: [PATCH 2/9] Builds a reusable e2e tester docker image that is used to
 execute the test cases

---
 .github/workflows/check-sdk-tests.yml | 99 +++++++++++++++++++++++++--
 1 file changed, 94 insertions(+), 5 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index 99fac81..c176cf0 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -3,6 +3,7 @@ name: Bittensor SDK Test
 permissions:
   pull-requests: write
   contents: read
+  packages: write
 
 concurrency:
   group: e2e-sdk-${{ github.ref }}
@@ -155,14 +156,104 @@ jobs:
           name: subtensor-localnet
           path: subtensor-localnet.tar
 
+  build-e2e-image:
+    # Build the test environment (system deps + bittensor[dev] + this PR's
+    # async-substrate-interface) into a single image ONCE per run, so the e2e
+    # matrix jobs below just pull it and run pytest — no per-test setup.
+    needs:
+      - check-labels
+    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    runs-on: ubuntu-latest
+    outputs:
+      image: ${{ steps.meta.outputs.image }}
+    steps:
+      - name: Clone Bittensor SDK repo
+        run: git clone https://github.com/latent-to/bittensor.git
+
+      - name: Checkout Bittensor branch
+        working-directory: ${{ github.workspace }}/bittensor
+        run: |
+          if ! git fetch origin $BITTENSOR_BRANCH; then
+            echo "❌ Error: Branch '$BITTENSOR_BRANCH' does not exist in latent-to/bittensor."
+            exit 1
+          fi
+          git checkout FETCH_HEAD
+          echo "✅ Using Bittensor branch: $BITTENSOR_BRANCH"
+
+      - name: Clone async-substrate-interface repo
+        run: git clone https://github.com/latent-to/async-substrate-interface.git
+
+      - name: Checkout PR branch in async-substrate-interface
+        working-directory: ${{ github.workspace }}/async-substrate-interface
+        run: |
+          BRANCH="${{ github.event.pull_request.head.ref || github.ref_name }}"
+          REPO_URL="${{ github.event.pull_request.head.repo.clone_url || 'https://github.com/latent-to/async-substrate-interface.git' }}"
+          git fetch $REPO_URL $BRANCH
+          git checkout FETCH_HEAD
+          echo "Current SHA: $(git rev-parse HEAD)"
+
+      - name: Write Dockerfile
+        run: |
+          # Exclude the (large) .git dirs from the build context.
+          printf '**/.git\n' > .dockerignore
+          cat > Dockerfile <<'DOCKERFILE'
+          FROM python:3.13-slim
+
+          # System deps: build toolchain for any source wheels, the docker CLI so the
+          # e2e fixtures can launch the subtensor-localnet container (docker-out-of-docker),
+          # and git for any VCS-based installs.
+          RUN apt-get update && apt-get install -y --no-install-recommends \
+                clang curl libssl-dev llvm libudev-dev protobuf-compiler git docker.io \
+              && rm -rf /var/lib/apt/lists/*
+
+          RUN pip install --no-cache-dir uv
+
+          COPY bittensor /opt/bittensor
+          COPY async-substrate-interface /opt/async-substrate-interface
+
+          # Install the SDK with dev extras, then overlay this PR's build of
+          # async-substrate-interface so the tests exercise the PR code.
+          RUN uv pip install --system '/opt/bittensor[dev]' \
+              && (uv pip uninstall --system async-substrate-interface || true) \
+              && uv pip install --system /opt/async-substrate-interface
+
+          # pytest is invoked from here so the relative `tests/e2e_tests/...` node ids resolve.
+          WORKDIR /opt/bittensor
+          DOCKERFILE
+
+      - name: Set image ref
+        id: meta
+        run: echo "image=ghcr.io/${{ github.repository }}/e2e:${{ github.run_id }}-${{ github.run_attempt }}" >> $GITHUB_OUTPUT
+
+      - name: Log in to GitHub Container Registry
+        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u $GITHUB_ACTOR --password-stdin
+
+      - name: Build and push image
+        run: |
+          docker build -t "${{ steps.meta.outputs.image }}" .
+          docker push "${{ steps.meta.outputs.image }}"
+
   run-e2e-tests:
     needs:
       - check-labels
       - find-e2e-tests
       - pull-docker-image
+      - build-e2e-image
 
     if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
     runs-on: ubuntu-latest
+
+    # Run each test inside the prebuilt image: all deps are already installed, so
+    # there is no per-test apt/clone/install. Docker-out-of-docker (mounted host
+    # socket) + host networking let the e2e fixtures launch the localnet container
+    # and reach it at ws://localhost:9944, just as on a bare runner.
+    container:
+      image: ${{ needs.build-e2e-image.outputs.image }}
+      credentials:
+        username: ${{ github.actor }}
+        password: ${{ secrets.GITHUB_TOKEN }}
+      options: --network host -v /var/run/docker.sock:/var/run/docker.sock
+
     strategy:
       fail-fast: false
       max-parallel: 16
@@ -170,11 +261,9 @@ jobs:
         test-case: ${{ fromJson(needs.find-e2e-tests.outputs.test-cases) }}
 
     env:
-      RELEASE_NAME: development
-      RUSTV: ${{ matrix.rust-branch }}
-      RUST_BACKTRACE: full
-      RUST_BIN_DIR: target/${{ matrix.rust-target }}
-      TARGET: ${{ matrix.rust-target }}
+      # The localnet image is preloaded from the cached tar below; skip the
+      # fixture's own `docker pull`.
+      SKIP_PULL: "1"
 
     timeout-minutes: 60
     name: "e2e tests: ${{ matrix.test-case }}"

From 580d7b2e18b36a48a4a802c01016e9d2e7ebe36d Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 16:40:10 +0200
Subject: [PATCH 3/9] Adjust for github runner

---
 .github/workflows/check-sdk-tests.yml | 38 ++++++++++++++-------------
 1 file changed, 20 insertions(+), 18 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index c176cf0..df66ba1 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -243,31 +243,21 @@ jobs:
     if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
     runs-on: ubuntu-latest
 
-    # Run each test inside the prebuilt image: all deps are already installed, so
-    # there is no per-test apt/clone/install. Docker-out-of-docker (mounted host
-    # socket) + host networking let the e2e fixtures launch the localnet container
-    # and reach it at ws://localhost:9944, just as on a bare runner.
-    container:
-      image: ${{ needs.build-e2e-image.outputs.image }}
-      credentials:
-        username: ${{ github.actor }}
-        password: ${{ secrets.GITHUB_TOKEN }}
-      options: --network host -v /var/run/docker.sock:/var/run/docker.sock
-
     strategy:
       fail-fast: false
       max-parallel: 16
       matrix:
         test-case: ${{ fromJson(needs.find-e2e-tests.outputs.test-cases) }}
 
-    env:
-      # The localnet image is preloaded from the cached tar below; skip the
-      # fixture's own `docker pull`.
-      SKIP_PULL: "1"
-
     timeout-minutes: 60
     name: "e2e tests: ${{ matrix.test-case }}"
     steps:
+      - name: Log in to GitHub Container Registry
+        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u $GITHUB_ACTOR --password-stdin
+
+      - name: Pull prebuilt test image
+        run: docker pull "${{ needs.build-e2e-image.outputs.image }}"
+
       - name: Download Cached Docker Image
         uses: actions/download-artifact@v8
         with:
@@ -276,9 +266,21 @@ jobs:
       - name: Load Docker Image
         run: docker load -i subtensor-localnet.tar
 
+      # The job stays on the bare runner (no `container:`), so we can launch the test
+      # container ourselves with `--network host` — GitHub injects its own network into
+      # job containers, which conflicts with host mode. Host networking + the mounted
+      # docker socket let the e2e fixtures start the localnet (published on the host's
+      # :9944) and reach it at ws://localhost:9944. All deps are baked into the image,
+      # so there is no per-test apt/clone/install.
       - name: Run e2e tests
-        working-directory: /opt/bittensor
-        run: pytest "${{ matrix.test-case }}" -s
+        run: |
+          docker run --rm \
+            --network host \
+            -v /var/run/docker.sock:/var/run/docker.sock \
+            -w /opt/bittensor \
+            -e SKIP_PULL=1 \
+            "${{ needs.build-e2e-image.outputs.image }}" \
+            pytest "${{ matrix.test-case }}" -s
 
 
   run-integration-and-unit-test:

From 395530ee9407b531d2149a6c2e5aa8431c1a4b1b Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 16:51:13 +0200
Subject: [PATCH 4/9] Troubleshooting

---
 .github/workflows/check-sdk-tests.yml | 32 +++++++++++++++++++++++----
 1 file changed, 28 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index df66ba1..2a09ec2 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -199,13 +199,22 @@ jobs:
           cat > Dockerfile <<'DOCKERFILE'
           FROM python:3.13-slim
 
-          # System deps: build toolchain for any source wheels, the docker CLI so the
-          # e2e fixtures can launch the subtensor-localnet container (docker-out-of-docker),
-          # and git for any VCS-based installs.
+          # System deps: build toolchain for any source wheels, curl for the docker CLI
+          # download below, and git for any VCS-based installs.
           RUN apt-get update && apt-get install -y --no-install-recommends \
-                clang curl libssl-dev llvm libudev-dev protobuf-compiler git docker.io \
+                clang curl libssl-dev llvm libudev-dev protobuf-compiler git \
               && rm -rf /var/lib/apt/lists/*
 
+          # Static Docker CLI (client only). The e2e fixtures shell out to `docker` to
+          # launch the localnet via the mounted host socket; installing the static binary
+          # to /usr/local/bin guarantees it is on PATH regardless of the base distro's
+          # packaging (the `docker.io` apt package did not reliably provide it). The final
+          # `docker --version` fails the build early if the CLI is ever missing.
+          ARG DOCKER_CLI_VERSION=27.5.1
+          RUN curl -fsSL "https://download.docker.com/linux/static/stable/x86_64/docker-${DOCKER_CLI_VERSION}.tgz" \
+                | tar -xz -C /usr/local/bin --strip-components=1 docker/docker \
+              && docker --version
+
           RUN pip install --no-cache-dir uv
 
           COPY bittensor /opt/bittensor
@@ -272,6 +281,21 @@ jobs:
       # docker socket let the e2e fixtures start the localnet (published on the host's
       # :9944) and reach it at ws://localhost:9944. All deps are baked into the image,
       # so there is no per-test apt/clone/install.
+      # Fail loudly (instead of letting the e2e fixture silently skip) if the docker
+      # CLI isn't on PATH, the host daemon isn't reachable through the mounted socket,
+      # or the localnet image didn't load — the three things the `local_chain` fixture
+      # needs before it will actually start a chain.
+      - name: Preflight - verify docker is reachable inside the test container
+        run: |
+          docker run --rm \
+            -v /var/run/docker.sock:/var/run/docker.sock \
+            "${{ needs.build-e2e-image.outputs.image }}" \
+            sh -euxc '
+              command -v docker
+              docker version
+              docker image inspect ghcr.io/opentensor/subtensor-localnet:devnet-ready >/dev/null
+            '
+
       - name: Run e2e tests
         run: |
           docker run --rm \

From 08ad1447366e6ebfd580b69da09e7a306adc6970 Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 16:56:52 +0200
Subject: [PATCH 5/9] Remove debug

---
 .github/workflows/check-sdk-tests.yml | 17 +----------------
 1 file changed, 1 insertion(+), 16 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index 2a09ec2..3376eea 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -254,7 +254,7 @@ jobs:
 
     strategy:
       fail-fast: false
-      max-parallel: 16
+      max-parallel: 64
       matrix:
         test-case: ${{ fromJson(needs.find-e2e-tests.outputs.test-cases) }}
 
@@ -281,21 +281,6 @@ jobs:
       # docker socket let the e2e fixtures start the localnet (published on the host's
       # :9944) and reach it at ws://localhost:9944. All deps are baked into the image,
       # so there is no per-test apt/clone/install.
-      # Fail loudly (instead of letting the e2e fixture silently skip) if the docker
-      # CLI isn't on PATH, the host daemon isn't reachable through the mounted socket,
-      # or the localnet image didn't load — the three things the `local_chain` fixture
-      # needs before it will actually start a chain.
-      - name: Preflight - verify docker is reachable inside the test container
-        run: |
-          docker run --rm \
-            -v /var/run/docker.sock:/var/run/docker.sock \
-            "${{ needs.build-e2e-image.outputs.image }}" \
-            sh -euxc '
-              command -v docker
-              docker version
-              docker image inspect ghcr.io/opentensor/subtensor-localnet:devnet-ready >/dev/null
-            '
-
       - name: Run e2e tests
         run: |
           docker run --rm \

From f079bb5a7020f6575e7ce3a073d05126334b76c2 Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 17:15:19 +0200
Subject: [PATCH 6/9] Change to GH artifact rather than registry

---
 .github/workflows/check-sdk-tests.yml | 34 +++++++++++++++++----------
 1 file changed, 22 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index 3376eea..a978d05 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -3,7 +3,6 @@ name: Bittensor SDK Test
 permissions:
   pull-requests: write
   contents: read
-  packages: write
 
 concurrency:
   group: e2e-sdk-${{ github.ref }}
@@ -232,15 +231,24 @@ jobs:
 
       - name: Set image ref
         id: meta
-        run: echo "image=ghcr.io/${{ github.repository }}/e2e:${{ github.run_id }}-${{ github.run_attempt }}" >> $GITHUB_OUTPUT
+        # A plain local tag (no registry): the image is shipped to the matrix jobs as a
+        # `docker save` tarball artifact, not via a registry push.
+        run: echo "image=async-substrate-interface-e2e:${{ github.run_id }}-${{ github.run_attempt }}" >> $GITHUB_OUTPUT
 
-      - name: Log in to GitHub Container Registry
-        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u $GITHUB_ACTOR --password-stdin
+      - name: Build image
+        run: docker build -t "${{ steps.meta.outputs.image }}" .
 
-      - name: Build and push image
-        run: |
-          docker build -t "${{ steps.meta.outputs.image }}" .
-          docker push "${{ steps.meta.outputs.image }}"
+      - name: Save image to tarball
+        # Distribute via a GitHub artifact (fast internal transport, same as the localnet
+        # image) instead of GHCR, which avoids slow registry pulls across all matrix jobs,
+        # registry rate-limits, and the fork-PR push-permission problem.
+        run: docker save -o e2e-image.tar "${{ steps.meta.outputs.image }}"
+
+      - name: Upload image as artifact
+        uses: actions/upload-artifact@v7
+        with:
+          name: e2e-image
+          path: e2e-image.tar
 
   run-e2e-tests:
     needs:
@@ -261,11 +269,13 @@ jobs:
     timeout-minutes: 60
     name: "e2e tests: ${{ matrix.test-case }}"
     steps:
-      - name: Log in to GitHub Container Registry
-        run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u $GITHUB_ACTOR --password-stdin
+      - name: Download prebuilt test image
+        uses: actions/download-artifact@v8
+        with:
+          name: e2e-image
 
-      - name: Pull prebuilt test image
-        run: docker pull "${{ needs.build-e2e-image.outputs.image }}"
+      - name: Load prebuilt test image
+        run: docker load -i e2e-image.tar
 
       - name: Download Cached Docker Image
         uses: actions/download-artifact@v8

From f644145bfe4876b4e77e47cd1e0f696e51eddd7d Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 17:19:10 +0200
Subject: [PATCH 7/9] Rerun failing tests to avoid flakiness

---
 .github/workflows/check-sdk-tests.yml | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index a978d05..c08cb6b 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -292,6 +292,10 @@ jobs:
       # :9944) and reach it at ws://localhost:9944. All deps are baked into the image,
       # so there is no per-test apt/clone/install.
       - name: Run e2e tests
+        # --reruns retries a flaky test (e.g. when the localnet is listening but its RPC
+        # isn't ready yet, surfacing as a websocket InvalidMessage during connect). The
+        # whole test reruns, re-spinning the localnet fixture, so by the retry the node
+        # is ready. pytest-rerunfailures ships with bittensor[dev].
         run: |
           docker run --rm \
             --network host \
@@ -299,7 +303,7 @@ jobs:
             -w /opt/bittensor \
             -e SKIP_PULL=1 \
             "${{ needs.build-e2e-image.outputs.image }}" \
-            pytest "${{ matrix.test-case }}" -s
+            pytest "${{ matrix.test-case }}" -s --reruns 2 --reruns-delay 5
 
 
   run-integration-and-unit-test:

From 8e6d4ddd840372f090db591060a6127ad24e253d Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 18:58:34 +0200
Subject: [PATCH 8/9] Replaces `always()` with `!cancelled()` as per PR
 suggestions

---
 .github/workflows/check-sdk-tests.yml | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index c08cb6b..ce8caa8 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -55,7 +55,7 @@ jobs:
   check-labels:
     needs: apply-label-to-new-pr
     runs-on: ubuntu-latest
-    if: always()
+    if: ${{ !cancelled() }}
     outputs:
       run-sdk-tests: ${{ steps.check-manual.outputs.run-sdk-tests || steps.get-labels-pr.outputs.run-sdk-tests }}
     steps:
@@ -88,7 +88,7 @@ jobs:
 
   find-e2e-tests:
     needs: check-labels
-    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
     runs-on: ubuntu-latest
     outputs:
       test-cases: ${{ steps.get-tests.outputs.test-cases }}
@@ -138,7 +138,7 @@ jobs:
       - find-e2e-tests
 
     runs-on: ubuntu-latest
-    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
     steps:
       - name: Log in to GitHub Container Registry
         run: echo "${{ secrets.GITHUB_TOKEN }}" | docker login ghcr.io -u $GITHUB_ACTOR --password-stdin
@@ -161,7 +161,7 @@ jobs:
     # matrix jobs below just pull it and run pytest — no per-test setup.
     needs:
       - check-labels
-    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
     runs-on: ubuntu-latest
     outputs:
       image: ${{ steps.meta.outputs.image }}
@@ -257,7 +257,7 @@ jobs:
       - pull-docker-image
       - build-e2e-image
 
-    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
     runs-on: ubuntu-latest
 
     strategy:
@@ -309,7 +309,7 @@ jobs:
   run-integration-and-unit-test:
     needs:
       - check-labels
-    if: always() && needs.check-labels.outputs.run-sdk-tests == 'true'
+    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
     runs-on: ubuntu-latest
     steps:
       - name: Check-out repository

From 8c643869a4ef588fd21f84d2aa051067b740f866 Mon Sep 17 00:00:00 2001
From: BD Himes <b@latent.to>
Date: Mon, 1 Jun 2026 19:07:40 +0200
Subject: [PATCH 9/9] Add success check as per PR suggestion

---
 .github/workflows/check-sdk-tests.yml | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/check-sdk-tests.yml b/.github/workflows/check-sdk-tests.yml
index ce8caa8..3016b3c 100644
--- a/.github/workflows/check-sdk-tests.yml
+++ b/.github/workflows/check-sdk-tests.yml
@@ -257,7 +257,15 @@ jobs:
       - pull-docker-image
       - build-e2e-image
 
-    if: ${{ !cancelled() && needs.check-labels.outputs.run-sdk-tests == 'true' }}
+    # Only fan out the matrix if every input-producing dependency succeeded — otherwise
+    # all 139 jobs would spawn only to fail (missing image/localnet artifact, or no
+    # test list). !cancelled() alone would let them run on upstream failure.
+    if: >-
+      ${{ !cancelled()
+      && needs.check-labels.outputs.run-sdk-tests == 'true'
+      && needs.find-e2e-tests.result == 'success'
+      && needs.pull-docker-image.result == 'success'
+      && needs.build-e2e-image.result == 'success' }}
     runs-on: ubuntu-latest
 
     strategy: