testops: add shared runner platform

.github/workflows/testops.yml

@@ -0,0 +1,53 @@
+name: TestOps
+
+on:
+  pull_request:
+    paths:
+      - "testops/**"
+      - ".github/workflows/testops.yml"
+  push:
+    branches:
+      - main
+    paths:
+      - "testops/**"
+      - ".github/workflows/testops.yml"
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  go:
+    name: Go test + docs
+    runs-on: ubuntu-24.04
+    timeout-minutes: 15
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+
+      - name: Setup Go
+        uses: actions/setup-go@v6
+        with:
+          go-version-file: testops/go.mod
+          cache-dependency-path: testops/go.sum
+
+      - name: Test
+        working-directory: testops
+        run: go test ./...
+
+      - name: Vet
+        working-directory: testops
+        run: go vet ./...
+
+      - name: Validate RDMA gate scenario
+        working-directory: testops
+        run: |
+          go run ./cmd/sweedrdma validate scenarios/rdma-unified-lab-gate.yaml
+          go run ./cmd/sweedrdma validate scenarios/nixl-provider-cpu-gate.yaml
+
+      - name: Build docs
+        working-directory: testops
+        run: |
+          python -m pip install -r requirements-docs.txt
+          python -m mkdocs build --strict

rdma-lab-ci/README.md

@@ -24,6 +24,18 @@ object/loader path:
 - SHA/byte correctness;
 - benchmark lines for the active rows.
 
+The second supported profile is `lmcache-nixl`. It runs the mono
+`enterprise/python/seaweedkv-lmcache/tests/lab/lmcache-connector-gate`
+scripts and checks the SeaweedKV/LMCache connector path:
+
+- `libseaweedfs_sw_rdma_nixl.so` built with `--features real-rdma`;
+- canonical RDMA-enabled `enterprise/seaweed-volume` on M02;
+- LMCache source staged on M01;
+- `nixl==1.3.0` installed in the M01 LMCache venv if missing;
+- 1 MiB and 20 MiB connector smokes through `loader-rdma`;
+- real NIXL CPU runtime transfer from retained Seaweed CPU descriptor to
+  caller-owned CPU DRAM descriptor, with SHA match.
+
 ## Install on M01
 
 1. Install a Buildkite agent on M01.
@@ -43,6 +55,17 @@ bash rdma-lab-ci/run-mono-rdma-lab.sh \
   --profile unified
 ```
 
+For the LMCache/NIXL CPU runtime gate:
+
+```bash
+bash rdma-lab-ci/run-mono-rdma-lab.sh \
+  --repo https://github.com/seaweedfs/seaweed-mono.git \
+  --ref main \
+  --profile lmcache-nixl \
+  --lmcache-repo https://github.com/LMCache/LMCache.git \
+  --lmcache-ref dev
+```
+
 For a PR branch:
 
 ```bash
@@ -83,6 +106,20 @@ The run must print:
 - `UNIFIED_VFS_READ_MATRIX_DONE`
 - `UNIFIED_RDMA_GATE_PASS`
 
+For `lmcache-nixl`, the run must print:
+
+- `SEAWEEDKV_LMCACHE_CONNECTOR_GATE_PASS` for both configured sizes;
+- `SEAWEEDKV_LMCACHE_WORKLOAD_PASS` with nonzero `batch_saves` and
+  `batch_loads` when `LMCACHE_WORKLOAD_SECONDS` is enabled;
+- `UNIFIED_COMMIT_WITNESS_OK`;
+- `LMCACHE_NIXL_RUNTIME_GATE_PASS`.
+
+When `LMCACHE_REQUIRE_NIXL_RUNTIME=1`, the run must also print:
+
+- `SEAWEEDKV_LMCACHE_NIXL_CPU_CALLER`;
+- `SEAWEEDKV_LMCACHE_NIXL_RUNTIME_CPU`;
+- `LMCACHE_CONNECTOR_NIXL_RUNTIME_SIZE_PASS`.
+
 For the current object hot path, the PR evidence should also record:
 
 - RC push throughput for 20MiB/c32 and 128MiB/c32;
@@ -102,6 +139,11 @@ Each run writes to `rdma-lab-runs/<timestamp>-<ref>-<profile>/`:
 - `run.log`
 - `summary.env`
 
+For `lmcache-nixl`, `summary.env` additionally records:
+
+- `RDMA_CI_LMCACHE_ROWS`;
+- `RDMA_CI_NIXL_RUNTIME_ROWS`.
+
 Buildkite uploads that directory as an artifact. For manual runs, serve the
 directory with any static file server if a simple web view is needed.
 

rdma-lab-ci/config.example.env

@@ -12,3 +12,18 @@ ARTIFACT_DIR=$PWD/rdma-lab-runs
 RDMA_PIPES=8
 ENABLE_DC=0
 DC_INITIATORS=4
+
+# Used by RDMA_CI_PROFILE=lmcache-nixl.
+LMCACHE_REPO=https://github.com/LMCache/LMCache.git
+LMCACHE_REF=dev
+LMCACHE_VENV=/opt/work/lmcache-connector-venv
+LMCACHE_SIZES="1048576 20971520"
+LMCACHE_WORKLOAD_SECONDS=0
+LMCACHE_WORKLOAD_SAVE_BATCH_SIZE=4
+LMCACHE_WORKLOAD_LOAD_BATCH_SIZE=4
+LMCACHE_WORKLOAD_SEED_SETTLE_SECS=0.5
+LMCACHE_RUN_NIXL_RUNTIME=1
+LMCACHE_REQUIRE_NIXL_RUNTIME=1
+LMCACHE_NIXL_RUNTIME_BACKEND=UCX
+LMCACHE_NIXL_RUNTIME_SIZES=1048576
+NIXL_PIP_SPEC=nixl==1.3.0