all_test.yml

name: ci_2_virt_node

on:
  push:
  pull_request:
    types:
      - opened
      - synchronize
      - reopened

permissions:
  contents: read

jobs:
  ci-2-virt-node:
    runs-on: ubuntu-latest
    steps:
      - name: Reclaim runner disk before CI
        run: |
          set -euo pipefail
          df -h / /mnt

          # ubuntu-latest includes large SDKs that this Rust/Python workflow never uses.
          sudo rm -rf \
            /opt/ghc \
            /opt/hostedtoolcache/CodeQL \
            /usr/local/.ghcup \
            /usr/local/lib/android \
            /usr/share/dotnet \
            /usr/share/swift

          sudo apt-get clean
          sudo rm -rf /var/lib/apt/lists/*

          if command -v docker >/dev/null 2>&1; then
            sudo systemctl start docker
            docker system prune --all --force --volumes
            docker system df
          fi

          df -h / /mnt

      - name: Check out repository
        uses: actions/checkout@v6
        with:
          fetch-depth: 0

      - name: Set up Python
        uses: actions/setup-python@v5
        with:
          python-version: "3.10"

      - name: Install host dependencies
        run: |
          sudo apt-get update
          sudo apt-get install -y --no-install-recommends \
            git \
            pigz \
            protobuf-compiler \
            python3-venv \
            rsync
          sudo apt-get clean
          sudo rm -rf /var/lib/apt/lists/*

      - name: Verify Docker availability
        run: |
          command -v docker >/dev/null
          if ! docker info >/dev/null 2>&1; then
            sudo systemctl start docker
            docker info >/dev/null
          fi
          docker version

      - name: Install Python dependencies
        run: python3 -m pip install PyYAML

      - name: Sync rather_no_git_submodule workspace inputs
        run: python3 fluxon_rs/scripts/rather_no_git_submodule.py

      - name: Write ci_2_virt_node suite
        run: |
          python3 - <<'PY'
          import copy
          from pathlib import Path
          import yaml

          repo_root = Path.cwd()
          src_path = repo_root / "fluxon_test_stack" / "ci_test_list.yaml"
          out_path = Path("${{ runner.temp }}") / "ci_test_list.ci.yaml"
          suite = yaml.safe_load(src_path.read_text(encoding="utf-8"))

          top_attention_scene_prefix = "ci_top_attention_"

          def top_attention_command(scene_id, *, case_config, timeout_seconds=21600):
              if not scene_id.startswith(top_attention_scene_prefix):
                  raise ValueError(f"not a top-attention CI scene id: {scene_id}")
              suffix = scene_id[len(top_attention_scene_prefix):]
              command = (
                  "__RUN_DIR__/venv/bin/python3 -u "
                  f"__RUN_DIR__/src/fluxon_test_stack/top_attention_test_index/_{suffix}.py"
              )
              if case_config:
                  command += " --case-config __RUN_DIR__/configs/ci_scene_config.yaml"
              return {
                  "id": f"top_attention_{suffix}",
                  "command": command,
                  "timeout_seconds": timeout_seconds,
              }

          top_attention_ci_scenes = {
              "ci_top_attention_doc_page_build": {
                  "subject": "doc_page",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_3gib",
                  "case_config": True,
                  "timeout_seconds": 10800,
                  "scene_config": {
                      "doc_site_base_url": "${{ github.repository_owner }}.github.io/${{ github.event.repository.name }}",
                  },
              },
              "ci_top_attention_bin_kvtest": {
                  "subject": "rust",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {
                      "kv_test_rounds": "p2p_only,p2p_only_ssd",
                  },
              },
              "ci_top_attention_cargo_kv_unit": {
                  "subject": "rust",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {},
              },
              "ci_top_attention_log_mgmt": {
                  "subject": "rust",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {
                      "enabled": True,
                  },
              },
              "ci_top_attention_ctrl_c_kv": {
                  "subject": "rust",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_3gib",
                  "case_config": False,
                  "scene_config": {},
              },
              "ci_top_attention_ctrl_c_mq": {
                  "subject": "mq",
                  "runtime_contract": "rust_self_managed",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": False,
                  "scene_config": {},
              },
              "ci_top_attention_mq_core": {
                  "subject": "mq",
                  "runtime_contract": "cluster_kv_owner",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {},
              },
              "ci_top_attention_mq_mpsc": {
                  "subject": "mq",
                  "runtime_contract": "cluster_kv_owner",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {},
              },
              "ci_top_attention_mq_mpmc": {
                  "subject": "mq",
                  "runtime_contract": "cluster_kv_owner",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {},
              },
              "ci_top_attention_mq_mpmc_bench": {
                  "subject": "mq",
                  "runtime_contract": "cluster_kv_owner",
                  "scale": "n1_kvowner_dram_20gib",
                  "case_config": True,
                  "scene_config": {},
              },
          }

          for scene_id, scene_def in top_attention_ci_scenes.items():
              commands = [
                  top_attention_command(
                      scene_id,
                      case_config=scene_def["case_config"],
                      timeout_seconds=scene_def.get("timeout_seconds", 21600),
                  )
              ]
              existing_scene = suite["scenes"].get(scene_id)
              if existing_scene is None:
                  suite["scenes"][scene_id] = {
                      "ci": {
                          "subject": scene_def["subject"],
                          "runtime_contract": scene_def["runtime_contract"],
                          "commands": commands,
                      },
                      "select": {
                          "scales": [scene_def["scale"]],
                          "profiles": ["fluxon_tcp"],
                      },
                  }
                  continue
              existing_scene["ci"]["subject"] = scene_def["subject"]
              existing_scene["ci"]["runtime_contract"] = scene_def["runtime_contract"]
              existing_scene["ci"]["commands"] = commands

          # Scene selection:
          # - ci_top_attention_doc_page_build validates doc build through the prebuilt Docker image.
          # - ci_top_attention_bin_kvtest covers the memory and native SSD P2P rounds.
          # - ci_top_attention_cargo_kv_unit runs the Rust KV crate unit tests.
          # - ci_top_attention_log_mgmt keeps log rolling/sharding coverage under the same CI testbed contract.
          # - ci_top_attention_ctrl_c_kv keeps runtime child-retirement Ctrl-C coverage in this workflow.
          # - ci_top_attention_ctrl_c_mq keeps MQ Ctrl-C exit coverage in this workflow.
          # - ci_top_attention_mq_core keeps MQ correctness coverage inside the same CI testbed contract.
          # - ci_top_attention_mq_mpsc keeps MPSC API channel coverage inside the same CI testbed contract.
          # - ci_top_attention_mq_mpmc keeps MPMC API channel coverage inside the same CI testbed contract.
          # - ci_top_attention_mq_mpmc_bench keeps heavier MPMC benchmark-style coverage in this workflow.
          suite["scenes"] = {
              key: value
              for key, value in suite["scenes"].items()
              if key in top_attention_ci_scenes
          }

          # Profile selection:
          # - Keep one Fluxon transport profile in this workflow to constrain matrix size.
          # - ci_2_virt_node will still rewrite the selected public profile into the local same-host CI profile.
          suite["profiles"] = {
              "fluxon_tcp": copy.deepcopy(suite["profiles"]["fluxon_tcp"])
          }
          suite["run"]["selectors"]["profile_ids"] = ["fluxon_tcp"]

          # Scene-local workload config:
          # - Restrict kv_test to one round in CI for runtime control.
          # - Keep the doc-site base URL in scene_configs instead of workflow env.
          scene_configs = suite["profiles"]["fluxon_tcp"]["runtime"]["ci"].setdefault("scene_configs", {})
          for scene_id, scene_def in top_attention_ci_scenes.items():
              scene_configs[scene_id] = copy.deepcopy(scene_def["scene_config"])

          # Scale selection:
          # - Keep the original per-scene scales from ci_test_list.yaml.
          # - ci_top_attention_doc_page_build stays on n1_kvowner_dram_3gib.
          # - ci_top_attention_bin_kvtest stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_cargo_kv_unit stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_log_mgmt stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_ctrl_c_kv stays on n1_kvowner_dram_3gib.
          # - ci_top_attention_ctrl_c_mq stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_mq_core stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_mq_mpsc stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_mq_mpmc stays on n1_kvowner_dram_20gib.
          # - ci_top_attention_mq_mpmc_bench stays on n1_kvowner_dram_20gib.

          out_path.write_text(
              yaml.safe_dump(suite, sort_keys=False, allow_unicode=False),
              encoding="utf-8",
          )
          print(out_path)
          PY

      - name: Run ci_2_virt_node full flow
        run: |
          python3 fluxon_test_stack/ci_2_virt_node.py \
            --suite-path "$RUNNER_TEMP/ci_test_list.ci.yaml" \
            --print-generated \
            --testbed-hostworkdir "$RUNNER_TEMP/fluxon_deploy" \
            --cleanup-pack-runtime-after-success \
            --cleanup-successful-case-artifacts

      - name: Print ci_2_virt_node failure summary
        if: ${{ failure() }}
        run: |
          python3 - <<'PY'
          from pathlib import Path
          import yaml

          workdir = Path(".dever/ci_2_virt_node/runner_run")
          case_runs_path = workdir / "case_runs.yaml"
          if not case_runs_path.exists():
              print(f"missing {case_runs_path}")
              raise SystemExit(0)

          case_runs = yaml.safe_load(case_runs_path.read_text(encoding="utf-8"))
          print("=== case_runs.yaml ===")
          print(yaml.safe_dump(case_runs, sort_keys=False, allow_unicode=False))

          for case in case_runs.get("cases", []):
              last_run = case.get("last_run", {})
              if last_run.get("outcome") != "SUCCESS":
                  case_id = case.get("case_id")
                  run_index = last_run.get("run_index")
                  print(f"=== failed case: {case_id} run_{run_index} ===")
                  run_dir = workdir / "results" / str(case_id) / f"run_{run_index}"
                  for rel in (
                      "summary.yaml",
                      "exception.txt",
                      "logs/ci_runner/exit_code.txt",
                  ):
                      path = run_dir / rel
                      print(f"=== {path} ===")
                      if path.exists():
                          print(path.read_text(encoding="utf-8", errors="replace"))
                      else:
                          print(f"missing {path}")

                  stdout_path = run_dir / "logs" / "ci_runner" / "stdout.log"
                  print(f"=== {stdout_path} tail ===")
                  if stdout_path.exists():
                      lines = stdout_path.read_text(encoding="utf-8", errors="replace").splitlines()
                      print("\n".join(lines[-240:]))
                  else:
                      print(f"missing {stdout_path}")
          PY

      - name: Normalize ci_2_virt_node debug artifact permissions
        if: ${{ failure() }}
        run: |
          chmod -R a+rX .dever || true
          chmod -R a+rX fluxon_release || true
          chmod -R a+rX setup_and_pack/nix/runs || true
          find .dever -type f -name '*.yaml' -exec chmod a+r {} + || true
          find .dever -type f -name '*.log' -exec chmod a+r {} + || true
          find .dever -type f -name '*.json' -exec chmod a+r {} + || true
          find .dever -type f -name '*.html' -exec chmod a+r {} + || true
          find .dever -type f -name '*.txt' -exec chmod a+r {} + || true
          find .dever -type f -name '*.sha256' -exec chmod a+r {} + || true
          find .dever -type d -path '*/pack_release_runtime/*' -exec chmod a+rx {} + || true
          find .dever -type d -path '*/pack_release_runtime/project-data/*' -exec chmod a+rx {} + || true
          find .dever -path '*/pack_release_runtime/project-data/*' \
            \( -path '*/instances/*/logs' -o -path '*/instances/*/release' -o -path '*/assemblies/*/profile' \) \
            -exec chmod -R a+rX {} + || true

      - name: Upload ci_2_virt_node debug artifacts
        if: ${{ failure() }}
        uses: actions/upload-artifact@v4
        with:
          name: ci-2-virt-node-debug-${{ github.sha }}
          if-no-files-found: warn
          compression-level: 1
          path: |
            .dever/**
            fluxon_release/**
            setup_and_pack/nix/runs/**
            .dever/**/pack_release_runtime/project-data/**/instances/**/logs/**
            .dever/**/pack_release_runtime/project-data/**/instances/**/release/**
            .dever/**/pack_release_runtime/project-data/**/assemblies/**/profile/**

      - name: Report final runner disk usage
        if: ${{ always() }}
        run: |
          df -h / /mnt
          docker system df || true