Use build cluster in devcontainers

.devcontainer/Dockerfile

@@ -21,6 +21,8 @@ ENV DEFAULT_CONDA_ENV=rapids
 
 FROM ${PYTHON_PACKAGE_MANAGER}-base
 
+ARG TARGETARCH
+
 ARG CUDA
 ENV CUDAARCHS="RAPIDS"
 ENV CUDA_VERSION="${CUDA_VERSION:-${CUDA}}"
@@ -32,7 +34,35 @@ ENV PYTHONSAFEPATH="1"
 ENV PYTHONUNBUFFERED="1"
 ENV PYTHONDONTWRITEBYTECODE="1"
 
+ENV HISTFILE="/home/coder/.cache/._bash_history"
+
+###
+# sccache configuration
+###
+ENV AWS_ROLE_ARN="arn:aws:iam::279114543810:role/nv-gha-token-sccache-devs"
 ENV SCCACHE_REGION="us-east-2"
 ENV SCCACHE_BUCKET="rapids-sccache-devs"
-ENV VAULT_HOST="https://vault.ops.k8s.rapids.ai"
-ENV HISTFILE="/home/coder/.cache/._bash_history"
+# 2hr (1 minute longer than sccache-dist request timeout)
+ENV SCCACHE_IDLE_TIMEOUT=7200
+
+###
+# sccache-dist configuration
+###
+# Enable sccache-dist by default
+ENV DEVCONTAINER_UTILS_ENABLE_SCCACHE_DIST=1
+# Compile locally if max retries exceeded
+ENV SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=true
+# Retry transient errors 4 times (for a total of 5 attempts)
+ENV SCCACHE_DIST_MAX_RETRIES=4
+ENV SCCACHE_DIST_CONNECT_TIMEOUT=30
+ENV SCCACHE_DIST_CONNECTION_POOL=false
+# 1hr 59min (to accommodate debug builds)
+ENV SCCACHE_DIST_REQUEST_TIMEOUT=7140
+ENV SCCACHE_DIST_KEEPALIVE_ENABLED=true
+ENV SCCACHE_DIST_KEEPALIVE_INTERVAL=20
+ENV SCCACHE_DIST_KEEPALIVE_TIMEOUT=600
+ENV SCCACHE_DIST_URL="https://${TARGETARCH}.linux.sccache.rapids.nvidia.com"
+
+# Build as much in parallel as possible
+ENV INFER_NUM_DEVICE_ARCHITECTURES=1
+ENV MAX_DEVICE_OBJ_TO_COMPILE_IN_PARALLEL=20

.devcontainer/cuda12.9-conda/devcontainer.json

@@ -11,7 +11,9 @@
   "runArgs": [
     "--rm",
     "--name",
-    "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-25.10-cuda12.9-conda"
+    "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-25.10-cuda12.9-conda",
+    "--ulimit",
+    "nofile=500000"
   ],
   "hostRequirements": {"gpu": "optional"},
   "features": {

.devcontainer/cuda12.9-pip/devcontainer.json

@@ -5,13 +5,15 @@
     "args": {
       "CUDA": "12.9",
       "PYTHON_PACKAGE_MANAGER": "pip",
-      "BASE": "rapidsai/devcontainers:25.10-cpp-cuda12.9-ucx1.18.0-openmpi5.0.7"
+      "BASE": "rapidsai/devcontainers:25.10-cpp-cuda12.9-ucx1.19.0-openmpi5.0.7"
     }
   },
   "runArgs": [
     "--rm",
     "--name",
-    "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-25.10-cuda12.9-pip"
+    "${localEnv:USER:anon}-rapids-${localWorkspaceFolderBasename}-25.10-cuda12.9-pip",
+    "--ulimit",
+    "nofile=500000"
   ],
   "hostRequirements": {"gpu": "optional"},
   "features": {

.github/workflows/pr.yaml

@@ -223,12 +223,19 @@ jobs:
     needs: telemetry-setup
     uses: rapidsai/shared-workflows/.github/workflows/build-in-devcontainer.yaml@branch-25.10
     with:
-      arch: '["amd64"]'
+      arch: '["amd64", "arm64"]'
       cuda: '["12.9"]'
+      node_type: "cpu8"
+      rapids-aux-secret-1: GIST_REPO_READ_ORG_GITHUB_TOKEN
+      env: |
+        SCCACHE_DIST_MAX_RETRIES=inf
+        SCCACHE_SERVER_LOG=sccache=debug
+        SCCACHE_DIST_FALLBACK_TO_LOCAL_COMPILE=false
+        SCCACHE_DIST_AUTH_TOKEN_VAR=RAPIDS_AUX_SECRET_1
       build_command: |
-        sccache -z;
-        build-all --verbose;
-        sccache -s;
+        sccache --zero-stats;
+        build-all -j0 --verbose 2>&1 | tee telemetry-artifacts/build.log;
+        sccache --show-adv-stats | tee telemetry-artifacts/sccache-stats.txt;
   telemetry-summarize:
     # This job must use a self-hosted runner to record telemetry traces.
     runs-on: linux-amd64-cpu4