From 330a60a51c0bc12f2b60196404d2425710d3d5be Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 30 Mar 2026 16:08:29 -0700
Subject: [PATCH 01/14] Use 1ES.Pool=diskann-github

---
 .github/workflows/ci.yml | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 375c3307e..788b6dd26 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -342,14 +342,7 @@ jobs:
   test-workspace:
     needs: basics
     name: test workspace
-    runs-on: ${{ matrix.os }}
-    strategy:
-      matrix:
-        os:
-          - windows-latest
-          - ubuntu-latest
-          - ubuntu-24.04-arm
-          - macos-latest
+    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
 
     steps:
       - uses: actions/checkout@v4

From d42f85a5ed180aaf15605834136ac1ec1e36294a Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 12:40:58 -0700
Subject: [PATCH 02/14] Revert "Use 1ES.Pool=diskann-github"

This reverts commit 330a60a51c0bc12f2b60196404d2425710d3d5be.
---
 .github/workflows/ci.yml | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 788b6dd26..375c3307e 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -342,7 +342,14 @@ jobs:
   test-workspace:
     needs: basics
     name: test workspace
-    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        os:
+          - windows-latest
+          - ubuntu-latest
+          - ubuntu-24.04-arm
+          - macos-latest
 
     steps:
       - uses: actions/checkout@v4

From e2d7c789fccd440a7f376af4be431478c45d637d Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 12:45:44 -0700
Subject: [PATCH 03/14] Update benchmark workflows to use self-hosted runners
 for improved performance

---
 .github/workflows/disk-benchmarks-aa.yml | 2 +-
 .github/workflows/disk-benchmarks.yml    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index 800acbdfb..d15ad629d 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -37,7 +37,7 @@ jobs:
   # A/A benchmark: run main vs main to detect environment noise.
   aa-benchmark:
     name: A/A - ${{ matrix.dataset }}
-    runs-on: ubuntu-latest
+    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
     timeout-minutes: 120
     strategy:
       fail-fast: false
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 919e74c15..63e3f7060 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -54,7 +54,7 @@ jobs:
   # Macro benchmark: compare current branch against baseline
   macro-benchmark:
     name: Macro Benchmark - ${{ matrix.dataset }}
-    runs-on: ubuntu-latest
+    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
     # TODO: For production benchmarks, consider using a self-hosted runner with:
     # - NVMe storage for consistent I/O performance
     # - CPU pinning (taskset) for reduced variance

From 637067c8b58e29d620bd5085924e34561414c60d Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 13:10:43 -0700
Subject: [PATCH 04/14] Install Rustup

---
 .github/actions/setup-disk-benchmark/action.yml | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index 19c6fcbaa..e7c63da44 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -23,7 +23,10 @@ runs:
   steps:
     - name: Install Rust
       shell: bash
-      run: rustup show
+      run: |
+        sudo apt-get update
+        sudo apt-get install -y rustup
+        rustup show
 
     - name: Install system dependencies
       shell: bash

From db6ade499c95acb68bb4dffce3b5bac494deff1e Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 15:44:08 -0700
Subject: [PATCH 05/14] Install Rustup

---
 .github/actions/setup-disk-benchmark/action.yml | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index e7c63da44..b98c1c1a3 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -24,8 +24,10 @@ runs:
     - name: Install Rust
       shell: bash
       run: |
-        sudo apt-get update
-        sudo apt-get install -y rustup
+        if ! command -v rustup &>/dev/null; then
+          curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+          echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
+        fi
         rustup show
 
     - name: Install system dependencies

From e7acfa004845cad988dd17c44fb3d07c86e53b07 Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 17:39:48 -0700
Subject: [PATCH 06/14] Move to a separate step: rustup show

---
 .github/actions/setup-disk-benchmark/action.yml | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index b98c1c1a3..f0db8d66b 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -28,7 +28,10 @@ runs:
           curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
         fi
-        rustup show
+
+    - name: Show Rust toolchain
+      shell: bash
+      run: rustup show
 
     - name: Install system dependencies
       shell: bash

From 699e4be25919547f40d4b8f232ba6e52b684773b Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 18:37:34 -0700
Subject: [PATCH 07/14] Mount SSD

---
 .../actions/setup-disk-benchmark/action.yml   | 12 +++++++
 .github/workflows/disk-benchmarks-aa.yml      | 31 ++++++++++---------
 .github/workflows/disk-benchmarks.yml         | 30 ++++++++++--------
 3 files changed, 46 insertions(+), 27 deletions(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index f0db8d66b..3621630a0 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -17,6 +17,10 @@ inputs:
   extract-to:
     description: 'Directory to extract the dataset into'
     required: true
+  nvme-mount-path:
+    description: 'Mount point for the high-speed NVMe SSD'
+    required: false
+    default: '/mnt/nvme'
 
 runs:
   using: composite
@@ -33,6 +37,14 @@ runs:
       shell: bash
       run: rustup show
 
+    - name: Mount high-speed NVMe SSD
+      shell: bash
+      run: |
+        sudo mkdir -p -m 777 ${{ inputs.nvme-mount-path }}
+        sudo lsblk
+        sudo mkfs.ext4 /dev/nvme0n1
+        sudo mount /dev/nvme0n1 ${{ inputs.nvme-mount-path }}
+
     - name: Install system dependencies
       shell: bash
       run: |
diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index d15ad629d..e2689d609 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -24,6 +24,8 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
+  NVME_MOUNT_PATH: /mnt/nvme
+  DISKANN_RUST: /mnt/nvme/diskann_rust
 
 defaults:
   run:
@@ -51,41 +53,42 @@ jobs:
             archive: openai-100K.tar.gz
 
     steps:
+      - name: Setup benchmark environment
+        uses: ${{ env.DISKANN_RUST }}/.github/actions/setup-disk-benchmark
+        with:
+          dataset: ${{ matrix.dataset }}
+          archive: ${{ matrix.archive }}
+          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
+          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
+
       - name: Checkout main
         uses: actions/checkout@v4
         with:
           ref: main
-          path: diskann_rust
+          path: ${{ env.DISKANN_RUST }}
           lfs: true
 
-      - name: Setup benchmark environment
-        uses: ./diskann_rust/.github/actions/setup-disk-benchmark
-        with:
-          dataset: ${{ matrix.dataset }}
-          archive: ${{ matrix.archive }}
-          extract-to: diskann_rust/target/tmp
-
       # A/A: build once, run twice (identical code — only detecting environment noise)
       - name: Build benchmark binary
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: cargo build -p diskann-benchmark --features disk-index --release
 
       - name: Run baseline benchmark
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run target benchmark
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Validate benchmark results
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             check run \
@@ -100,8 +103,8 @@ jobs:
         with:
           name: aa-results-${{ matrix.dataset }}
           path: |
-            diskann_rust/target/tmp/${{ matrix.dataset }}_target.json
-            diskann_rust/target/tmp/${{ matrix.dataset }}_baseline.json
+            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_target.json
+            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_baseline.json
           retention-days: 30
 
   # Notify diskann-disk-maintainers on A/A failure — but only when the failure
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 63e3f7060..098e302fe 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -42,6 +42,9 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
+  NVME_MOUNT_PATH: /mnt/nvme
+  DISKANN_RUST: /mnt/nvme/diskann_rust
+  BASELINE: /mnt/nvme/baseline
 
 defaults:
   run:
@@ -77,50 +80,51 @@ jobs:
       - name: Checkout current branch
         uses: actions/checkout@v4
         with:
-          path: diskann_rust
+          path: ${{ env.DISKANN_RUST }}
           lfs: true
 
       - name: Checkout baseline (${{ inputs.baseline_ref || 'main' }})
         uses: actions/checkout@v4
         with:
           ref: ${{ inputs.baseline_ref || 'main' }}
-          path: baseline
+          path: ${{ env.BASELINE }}
           lfs: true
 
       - name: Setup benchmark environment
-        uses: ./diskann_rust/.github/actions/setup-disk-benchmark
+        uses: ${{ env.DISKANN_RUST }}/.github/actions/setup-disk-benchmark
         with:
           dataset: ${{ matrix.dataset }}
           archive: ${{ matrix.archive }}
-          extract-to: diskann_rust/target/tmp
+          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
+          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
 
       - name: Copy dataset to baseline
         run: |
-          mkdir -p baseline/target/tmp
-          cp -r diskann_rust/target/tmp/${{ matrix.data_dir }} baseline/target/tmp/
+          mkdir -p ${{ env.BASELINE }}/target/tmp
+          cp -r ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.data_dir }} ${{ env.BASELINE }}/target/tmp/
 
       - name: Run baseline benchmark
-        working-directory: baseline
+        working-directory: ${{ env.BASELINE }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
-            run --input-file ../diskann_rust/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
+            run --input-file ../${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run current branch benchmark
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Validate benchmark results
-        working-directory: diskann_rust
+        working-directory: ${{ env.DISKANN_RUST }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             check run \
             --tolerances ${{ env.PERF_INPUTS }}/disk-index-tolerances.json \
             --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
-            --before ../baseline/target/tmp/${{ matrix.dataset }}_baseline.json \
+            --before ${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json \
             --after target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Upload benchmark results
@@ -129,6 +133,6 @@ jobs:
         with:
           name: benchmark-results-${{ matrix.dataset }}
           path: |
-            diskann_rust/target/tmp/${{ matrix.dataset }}_target.json
-            baseline/target/tmp/${{ matrix.dataset }}_baseline.json
+            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_target.json
+            ${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json
           retention-days: 30
\ No newline at end of file

From bcb5e4127db55fb0c0af8ccde450cd05bb5fddc1 Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Mon, 11 May 2026 19:04:35 -0700
Subject: [PATCH 08/14] Checkout action definitions

---
 .github/workflows/disk-benchmarks-aa.yml |  9 ++++++++-
 .github/workflows/disk-benchmarks.yml    | 25 +++++++++++++++---------
 2 files changed, 24 insertions(+), 10 deletions(-)

diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index e2689d609..9ded86fa5 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -53,8 +53,15 @@ jobs:
             archive: openai-100K.tar.gz
 
     steps:
+      # Shallow checkout to $GITHUB_WORKSPACE so the composite action is accessible via uses:.
+      - name: Checkout action definitions
+        uses: actions/checkout@v4
+        with:
+          sparse-checkout: .github
+          path: github_actions
+
       - name: Setup benchmark environment
-        uses: ${{ env.DISKANN_RUST }}/.github/actions/setup-disk-benchmark
+        uses: ./github_actions/.github/actions/setup-disk-benchmark
         with:
           dataset: ${{ matrix.dataset }}
           archive: ${{ matrix.archive }}
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 098e302fe..f2c5124a0 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -77,6 +77,21 @@ jobs:
             data_dir: OpenAIArXiv
 
     steps:
+      # Shallow checkout to $GITHUB_WORKSPACE so the composite action is accessible via uses:.
+      - name: Checkout action definitions
+        uses: actions/checkout@v4
+        with:
+          sparse-checkout: .github
+          path: github_actions
+
+      - name: Setup benchmark environment
+        uses: ./github_actions/.github/actions/setup-disk-benchmark
+        with:
+          dataset: ${{ matrix.dataset }}
+          archive: ${{ matrix.archive }}
+          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
+          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
+
       - name: Checkout current branch
         uses: actions/checkout@v4
         with:
@@ -90,14 +105,6 @@ jobs:
           path: ${{ env.BASELINE }}
           lfs: true
 
-      - name: Setup benchmark environment
-        uses: ${{ env.DISKANN_RUST }}/.github/actions/setup-disk-benchmark
-        with:
-          dataset: ${{ matrix.dataset }}
-          archive: ${{ matrix.archive }}
-          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
-          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
-
       - name: Copy dataset to baseline
         run: |
           mkdir -p ${{ env.BASELINE }}/target/tmp
@@ -107,7 +114,7 @@ jobs:
         working-directory: ${{ env.BASELINE }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
-            run --input-file ../${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
+            run --input-file ${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run current branch benchmark

From e7e8b7806e36c1e226e418c60ac645996cfdcf8e Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 17:16:01 -0700
Subject: [PATCH 09/14] sudo chmod 777 ${{ inputs.nvme-mount-path }}

---
 .github/actions/setup-disk-benchmark/action.yml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index 3621630a0..7ad7d4174 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -40,10 +40,11 @@ runs:
     - name: Mount high-speed NVMe SSD
       shell: bash
       run: |
-        sudo mkdir -p -m 777 ${{ inputs.nvme-mount-path }}
+        sudo mkdir -p ${{ inputs.nvme-mount-path }}
         sudo lsblk
         sudo mkfs.ext4 /dev/nvme0n1
         sudo mount /dev/nvme0n1 ${{ inputs.nvme-mount-path }}
+        sudo chmod 777 ${{ inputs.nvme-mount-path }}
 
     - name: Install system dependencies
       shell: bash

From 8e698125964bc03b991a3a34140c77d17a64b5fc Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 17:53:11 -0700
Subject: [PATCH 10/14] Paths are relative to $GITHUB_WORKSPACE

---
 .github/workflows/disk-benchmarks-aa.yml |  4 ++--
 .github/workflows/disk-benchmarks.yml    | 10 +++++-----
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index 9ded86fa5..d5669ae11 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -24,8 +24,8 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
-  NVME_MOUNT_PATH: /mnt/nvme
-  DISKANN_RUST: /mnt/nvme/diskann_rust
+  NVME_MOUNT_PATH: nvme
+  DISKANN_RUST: nvme/diskann_rust
 
 defaults:
   run:
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index f2c5124a0..e93e62b75 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -42,9 +42,9 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
-  NVME_MOUNT_PATH: /mnt/nvme
-  DISKANN_RUST: /mnt/nvme/diskann_rust
-  BASELINE: /mnt/nvme/baseline
+  NVME_MOUNT_PATH: nvme
+  DISKANN_RUST: nvme/diskann_rust
+  BASELINE: nvme/baseline
 
 defaults:
   run:
@@ -114,7 +114,7 @@ jobs:
         working-directory: ${{ env.BASELINE }}
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
-            run --input-file ${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
+            run --input-file ../../${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run current branch benchmark
@@ -131,7 +131,7 @@ jobs:
             check run \
             --tolerances ${{ env.PERF_INPUTS }}/disk-index-tolerances.json \
             --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
-            --before ${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json \
+            --before ../../${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json \
             --after target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Upload benchmark results

From e8380d4ea3a7a08f435ee8470b2237525c9e6e79 Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 21:45:38 -0700
Subject: [PATCH 11/14] Revert "Update benchmark workflows to use self-hosted
 runners for improved performance"

This reverts commit e2d7c789fccd440a7f376af4be431478c45d637d.
---
 .../actions/setup-disk-benchmark/action.yml   | 21 --------
 .github/workflows/disk-benchmarks-aa.yml      | 36 +++++---------
 .github/workflows/disk-benchmarks.yml         | 49 +++++++------------
 3 files changed, 32 insertions(+), 74 deletions(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index 7ad7d4174..19c6fcbaa 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -17,35 +17,14 @@ inputs:
   extract-to:
     description: 'Directory to extract the dataset into'
     required: true
-  nvme-mount-path:
-    description: 'Mount point for the high-speed NVMe SSD'
-    required: false
-    default: '/mnt/nvme'
 
 runs:
   using: composite
   steps:
     - name: Install Rust
-      shell: bash
-      run: |
-        if ! command -v rustup &>/dev/null; then
-          curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
-          echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
-        fi
-
-    - name: Show Rust toolchain
       shell: bash
       run: rustup show
 
-    - name: Mount high-speed NVMe SSD
-      shell: bash
-      run: |
-        sudo mkdir -p ${{ inputs.nvme-mount-path }}
-        sudo lsblk
-        sudo mkfs.ext4 /dev/nvme0n1
-        sudo mount /dev/nvme0n1 ${{ inputs.nvme-mount-path }}
-        sudo chmod 777 ${{ inputs.nvme-mount-path }}
-
     - name: Install system dependencies
       shell: bash
       run: |
diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index d5669ae11..800acbdfb 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -24,8 +24,6 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
-  NVME_MOUNT_PATH: nvme
-  DISKANN_RUST: nvme/diskann_rust
 
 defaults:
   run:
@@ -39,7 +37,7 @@ jobs:
   # A/A benchmark: run main vs main to detect environment noise.
   aa-benchmark:
     name: A/A - ${{ matrix.dataset }}
-    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
+    runs-on: ubuntu-latest
     timeout-minutes: 120
     strategy:
       fail-fast: false
@@ -53,49 +51,41 @@ jobs:
             archive: openai-100K.tar.gz
 
     steps:
-      # Shallow checkout to $GITHUB_WORKSPACE so the composite action is accessible via uses:.
-      - name: Checkout action definitions
+      - name: Checkout main
         uses: actions/checkout@v4
         with:
-          sparse-checkout: .github
-          path: github_actions
+          ref: main
+          path: diskann_rust
+          lfs: true
 
       - name: Setup benchmark environment
-        uses: ./github_actions/.github/actions/setup-disk-benchmark
+        uses: ./diskann_rust/.github/actions/setup-disk-benchmark
         with:
           dataset: ${{ matrix.dataset }}
           archive: ${{ matrix.archive }}
-          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
-          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
-
-      - name: Checkout main
-        uses: actions/checkout@v4
-        with:
-          ref: main
-          path: ${{ env.DISKANN_RUST }}
-          lfs: true
+          extract-to: diskann_rust/target/tmp
 
       # A/A: build once, run twice (identical code — only detecting environment noise)
       - name: Build benchmark binary
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: cargo build -p diskann-benchmark --features disk-index --release
 
       - name: Run baseline benchmark
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run target benchmark
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Validate benchmark results
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             check run \
@@ -110,8 +100,8 @@ jobs:
         with:
           name: aa-results-${{ matrix.dataset }}
           path: |
-            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_target.json
-            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_baseline.json
+            diskann_rust/target/tmp/${{ matrix.dataset }}_target.json
+            diskann_rust/target/tmp/${{ matrix.dataset }}_baseline.json
           retention-days: 30
 
   # Notify diskann-disk-maintainers on A/A failure — but only when the failure
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index e93e62b75..919e74c15 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -42,9 +42,6 @@ concurrency:
 env:
   RUST_BACKTRACE: 1
   PERF_INPUTS: diskann-benchmark/perf_test_inputs
-  NVME_MOUNT_PATH: nvme
-  DISKANN_RUST: nvme/diskann_rust
-  BASELINE: nvme/baseline
 
 defaults:
   run:
@@ -57,7 +54,7 @@ jobs:
   # Macro benchmark: compare current branch against baseline
   macro-benchmark:
     name: Macro Benchmark - ${{ matrix.dataset }}
-    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
+    runs-on: ubuntu-latest
     # TODO: For production benchmarks, consider using a self-hosted runner with:
     # - NVMe storage for consistent I/O performance
     # - CPU pinning (taskset) for reduced variance
@@ -77,61 +74,53 @@ jobs:
             data_dir: OpenAIArXiv
 
     steps:
-      # Shallow checkout to $GITHUB_WORKSPACE so the composite action is accessible via uses:.
-      - name: Checkout action definitions
-        uses: actions/checkout@v4
-        with:
-          sparse-checkout: .github
-          path: github_actions
-
-      - name: Setup benchmark environment
-        uses: ./github_actions/.github/actions/setup-disk-benchmark
-        with:
-          dataset: ${{ matrix.dataset }}
-          archive: ${{ matrix.archive }}
-          extract-to: ${{ env.DISKANN_RUST }}/target/tmp
-          nvme-mount-path: ${{ env.NVME_MOUNT_PATH }}
-
       - name: Checkout current branch
         uses: actions/checkout@v4
         with:
-          path: ${{ env.DISKANN_RUST }}
+          path: diskann_rust
           lfs: true
 
       - name: Checkout baseline (${{ inputs.baseline_ref || 'main' }})
         uses: actions/checkout@v4
         with:
           ref: ${{ inputs.baseline_ref || 'main' }}
-          path: ${{ env.BASELINE }}
+          path: baseline
           lfs: true
 
+      - name: Setup benchmark environment
+        uses: ./diskann_rust/.github/actions/setup-disk-benchmark
+        with:
+          dataset: ${{ matrix.dataset }}
+          archive: ${{ matrix.archive }}
+          extract-to: diskann_rust/target/tmp
+
       - name: Copy dataset to baseline
         run: |
-          mkdir -p ${{ env.BASELINE }}/target/tmp
-          cp -r ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.data_dir }} ${{ env.BASELINE }}/target/tmp/
+          mkdir -p baseline/target/tmp
+          cp -r diskann_rust/target/tmp/${{ matrix.data_dir }} baseline/target/tmp/
 
       - name: Run baseline benchmark
-        working-directory: ${{ env.BASELINE }}
+        working-directory: baseline
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
-            run --input-file ../../${{ env.DISKANN_RUST }}/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
+            run --input-file ../diskann_rust/${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_baseline.json
 
       - name: Run current branch benchmark
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             run --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
             --output-file target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Validate benchmark results
-        working-directory: ${{ env.DISKANN_RUST }}
+        working-directory: diskann_rust
         run: |
           cargo run -p diskann-benchmark --features disk-index --release -- \
             check run \
             --tolerances ${{ env.PERF_INPUTS }}/disk-index-tolerances.json \
             --input-file ${{ env.PERF_INPUTS }}/${{ matrix.config }} \
-            --before ../../${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json \
+            --before ../baseline/target/tmp/${{ matrix.dataset }}_baseline.json \
             --after target/tmp/${{ matrix.dataset }}_target.json
 
       - name: Upload benchmark results
@@ -140,6 +129,6 @@ jobs:
         with:
           name: benchmark-results-${{ matrix.dataset }}
           path: |
-            ${{ env.DISKANN_RUST }}/target/tmp/${{ matrix.dataset }}_target.json
-            ${{ env.BASELINE }}/target/tmp/${{ matrix.dataset }}_baseline.json
+            diskann_rust/target/tmp/${{ matrix.dataset }}_target.json
+            baseline/target/tmp/${{ matrix.dataset }}_baseline.json
           retention-days: 30
\ No newline at end of file

From ac9760e39604c7bc0d621c7d2418542deb1f7b37 Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 21:53:53 -0700
Subject: [PATCH 12/14] Mount to local folders

---
 .github/actions/setup-disk-benchmark/action.yml | 10 +++++++++-
 .github/workflows/disk-benchmarks-aa.yml        | 12 ++++++++++++
 .github/workflows/disk-benchmarks.yml           | 12 ++++++++++++
 3 files changed, 33 insertions(+), 1 deletion(-)

diff --git a/.github/actions/setup-disk-benchmark/action.yml b/.github/actions/setup-disk-benchmark/action.yml
index 19c6fcbaa..69d7cef74 100644
--- a/.github/actions/setup-disk-benchmark/action.yml
+++ b/.github/actions/setup-disk-benchmark/action.yml
@@ -23,8 +23,16 @@ runs:
   steps:
     - name: Install Rust
       shell: bash
-      run: rustup show
+      run: |
+        if ! command -v rustup &>/dev/null; then
+          curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
+          echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
+        fi
 
+    - name: Show Rust toolchain
+      shell: bash
+      run: rustup show
+      
     - name: Install system dependencies
       shell: bash
       run: |
diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index 800acbdfb..164a7ae62 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -51,6 +51,18 @@ jobs:
             archive: openai-100K.tar.gz
 
     steps:
+      - name: Mount high-speed NVMe SSD
+        shell: bash
+        run: |
+          sudo mkdir -p nvme
+          sudo lsblk
+          sudo mkfs.ext4 /dev/nvme0n1
+          sudo mount /dev/nvme0n1 nvme
+          sudo chmod 777 nvme
+          mkdir -p nvme/diskann_rust nvme/baseline
+          ln -s nvme/diskann_rust diskann_rust
+          ln -s nvme/baseline baseline
+
       - name: Checkout main
         uses: actions/checkout@v4
         with:
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 919e74c15..68c13d062 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -74,6 +74,18 @@ jobs:
             data_dir: OpenAIArXiv
 
     steps:
+      - name: Mount high-speed NVMe SSD
+        shell: bash
+        run: |
+          sudo mkdir -p nvme
+          sudo lsblk
+          sudo mkfs.ext4 /dev/nvme0n1
+          sudo mount /dev/nvme0n1 nvme
+          sudo chmod 777 nvme
+          mkdir -p nvme/diskann_rust nvme/baseline
+          ln -s nvme/diskann_rust diskann_rust
+          ln -s nvme/baseline baseline
+
       - name: Checkout current branch
         uses: actions/checkout@v4
         with:

From 504b9dc8474b8ce5102ff90e172906af71d4aaff Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 22:38:01 -0700
Subject: [PATCH 13/14] Fixing SSD mount

---
 .github/workflows/disk-benchmarks-aa.yml | 12 ++++++------
 .github/workflows/disk-benchmarks.yml    | 12 ++++++------
 2 files changed, 12 insertions(+), 12 deletions(-)

diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index 164a7ae62..5d47de794 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -54,14 +54,14 @@ jobs:
       - name: Mount high-speed NVMe SSD
         shell: bash
         run: |
-          sudo mkdir -p nvme
+          sudo mkdir -p /mnt/nvme
           sudo lsblk
           sudo mkfs.ext4 /dev/nvme0n1
-          sudo mount /dev/nvme0n1 nvme
-          sudo chmod 777 nvme
-          mkdir -p nvme/diskann_rust nvme/baseline
-          ln -s nvme/diskann_rust diskann_rust
-          ln -s nvme/baseline baseline
+          sudo mount /dev/nvme0n1 /mnt/nvme
+          sudo chmod 777 /mnt/nvme
+          mkdir -p /mnt/nvme/diskann_rust /mnt/nvme/baseline
+          ln -s /mnt/nvme/diskann_rust diskann_rust
+          ln -s /mnt/nvme/baseline baseline
 
       - name: Checkout main
         uses: actions/checkout@v4
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 68c13d062..5cb1073a7 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -77,14 +77,14 @@ jobs:
       - name: Mount high-speed NVMe SSD
         shell: bash
         run: |
-          sudo mkdir -p nvme
+          sudo mkdir -p /mnt/nvme
           sudo lsblk
           sudo mkfs.ext4 /dev/nvme0n1
-          sudo mount /dev/nvme0n1 nvme
-          sudo chmod 777 nvme
-          mkdir -p nvme/diskann_rust nvme/baseline
-          ln -s nvme/diskann_rust diskann_rust
-          ln -s nvme/baseline baseline
+          sudo mount /dev/nvme0n1 /mnt/nvme
+          sudo chmod 777 /mnt/nvme
+          mkdir -p /mnt/nvme/diskann_rust /mnt/nvme/baseline
+          ln -s /mnt/nvme/diskann_rust diskann_rust
+          ln -s /mnt/nvme/baseline baseline
 
       - name: Checkout current branch
         uses: actions/checkout@v4

From db2965511830f41cf956d51bfe918c0f574e8af0 Mon Sep 17 00:00:00 2001
From: "Alex Razumov (from Dev Box)" <alrazu@microsoft.com>
Date: Tue, 12 May 2026 22:49:27 -0700
Subject: [PATCH 14/14] Mount SSD

---
 .github/workflows/disk-benchmarks-aa.yml | 2 +-
 .github/workflows/disk-benchmarks.yml    | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/disk-benchmarks-aa.yml b/.github/workflows/disk-benchmarks-aa.yml
index 5d47de794..60e38aa08 100644
--- a/.github/workflows/disk-benchmarks-aa.yml
+++ b/.github/workflows/disk-benchmarks-aa.yml
@@ -37,7 +37,7 @@ jobs:
   # A/A benchmark: run main vs main to detect environment noise.
   aa-benchmark:
     name: A/A - ${{ matrix.dataset }}
-    runs-on: ubuntu-latest
+    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
     timeout-minutes: 120
     strategy:
       fail-fast: false
diff --git a/.github/workflows/disk-benchmarks.yml b/.github/workflows/disk-benchmarks.yml
index 5cb1073a7..9ef3b4075 100644
--- a/.github/workflows/disk-benchmarks.yml
+++ b/.github/workflows/disk-benchmarks.yml
@@ -54,7 +54,7 @@ jobs:
   # Macro benchmark: compare current branch against baseline
   macro-benchmark:
     name: Macro Benchmark - ${{ matrix.dataset }}
-    runs-on: ubuntu-latest
+    runs-on: [ self-hosted, 1ES.Pool=diskann-github, ubuntu-latest ]
     # TODO: For production benchmarks, consider using a self-hosted runner with:
     # - NVMe storage for consistent I/O performance
     # - CPU pinning (taskset) for reduced variance