chore(config): strip _comment_*/_example_* clutter now that cvs man documents these

cvs/input/config_file/health/mi300_health_config.json

@@ -3,7 +3,6 @@
         "path": "/opt/amd/agfhc",
         "package_tar_ball": "/home/{user-id}/PACKAGES/agfhc-mi300x_1.22.0_ub2204.tar.bz2",
         "install_dir": "/home/{user-id}/INSTALL/agfhc/",
-        "_comments_log_dir": "log_dir has to be a NON NFS local file system",
         "log_dir": "/root/agfhc_logs",
         "hbm_test_duration": "00:01:30"
     },
@@ -12,7 +11,6 @@
         "git_install_path": "/home/{user-id}/INSTALL/",
         "git_url": "https://github.com/ROCm/TransferBench.git",
         "git_tag": "v1.67.00",
-        "_comment_rocm_path": "ROCm installation path. Set the placeholder changeme to auto-detect from /opt/rocm or /opt/rocm/core-*",
         "rocm_path": "<changeme>",
         "results": {
             "bytes_to_transfer": "268435456",
@@ -34,13 +32,10 @@
         "git_install_path": "/home/{user-id}/INSTALL/rvs",
         "git_url": "https://github.com/ROCm/ROCmValidationSuite.git -b release/rocm-rel-7.2",
         "nfs_install": "True",
-        "_comment_rocm_path": "ROCm installation path.Set to placeholder changeme auto-detect from /opt/rocm or /opt/rocm/core-*",
         "rocm_path": "<changeme>",
-        "_comment_rocm_runtime_lib_path": "Optional. Colon-separated dirs prepended to LD_LIBRARY_PATH for every rvs invocation (also used during the install ldd check). Use this when the rvs binary under /opt/rocm/extras-7 was built against a newer ROCm than /opt/rocm symlinks to and you need it to load amd_smi/rocm_smi from a side-by-side install (e.g. /home/{user-id}/install/lib:/home/{user-id}/install/lib/rocm_sysdeps:/home/{user-id}/install/lib/llvm/lib). Leave empty for default loader behavior.",
         "rocm_runtime_lib_path": "",
 	"config_path_mi300x": "/opt/rocm/share/rocm-validation-suite/conf/MI300X",
         "config_path_default": "/opt/rocm/share/rocm-validation-suite/conf",
-        "_comment_rvs_test_level": "RVS test level configuration (0-5). 0: Run individual tests (skip level test), 1-5: Run LEVEL config test if RVS >= 1.3.0, else run individual tests. Default is 4.",
         "rvs_test_level": 4,
         "tests": [
             {

cvs/input/config_file/preflight/README_preflight_config.md

@@ -99,9 +99,19 @@ preflight/
 
 ## Configuration Parameters
 
-### Complete Parameter Reference
+Every parameter -- with its type, real default, constraints and an example --
+is documented by `cvs man`:
 
-All parameters below are optional and have sensible defaults. The sample configuration file includes all available parameters with their default values and inline comments explaining their purpose.
+```bash
+cvs man preflight_checks              # every parameter
+cvs man preflight_checks gid_index    # a single parameter
+cvs man preflight_checks --json       # machine-readable
+```
+
+That reference is generated from `PreflightConfigFile` in
+`cvs/parsers/schemas.py`, so it cannot drift from the code the way a
+hand-written table does. All parameters are optional and have sensible
+defaults.
 
 ### Important Update: RDMA Connectivity Testing
 
@@ -115,93 +125,7 @@ All parameters below are optional and have sensible defaults. The sample configu
 
 **Updated parameter names**: Configuration parameters now use accurate names (`ibv_test_timeout`, `ibv_test_port_range`) that reflect the use of `ibv_rc_pingpong` for testing.
 
-### RDMA Batching (`connectivity_check.rdma`)
-
-- **`nodes_per_full_mesh_group`** (default: 128)
-  - Group size for parallel RDMA connectivity testing (2-512 nodes per group)
-  - Smaller groups use fewer resources per node but require more rounds
-  - Adjust based on cluster size and resource constraints
-
-### Debug Settings (`debug`)
-
-- **`scriptlet`** (default: false)
-  - Enable ScriptLet debug mode: preserve generated scripts/logs on remote nodes
-  - For RDMA connectivity, wraps each ibv_rc_pingpong server in strace
-  - Creates per-test traces under /tmp/preflight/strace_server_<iface>_<port>.log
-  - **Warning**: Can be expensive at scale due to strace overhead
-
-### Node Check Settings (`node_check`)
-
-- **`enabled`** (default: `true`)
-  - Enables GPU visibility, AMDGPU/KFD, kernel-health, and ROCm validation
-  - Set to `false` to skip node-local health checks
-
-- **`gpus_per_node`** (default: `4`)
-  - Exact number of AMD GPUs expected on every node
-  - GPU visibility is generation-independent and can run on older or newer AMD hardware
-
-- **`expected_rocm_version`** (default: "6.2.0")
-  - Expected ROCm version across all cluster nodes
-  - Must match the output of `amd-smi version` on all nodes
-  - Format: "major.minor.patch" (e.g., "6.2.0", "5.7.1")
-
-### Connectivity Check Settings (`connectivity_check`)
-
-#### RDMA Settings (`connectivity_check.rdma`)
-
-- **`connectivity_mode`** (default: "basic")
-  - **"basic"**: Test adjacent node pairs (fast, ~14% coverage for 8 nodes)
-  - **"full_mesh"**: Test all possible node pairs (comprehensive, 100% coverage)
-  - **"skip"**: Skip RDMA interface presence, GID validation, and pairwise connectivity
-
-- **`gid_index`** (default: "3")
-  - GID index to check on all configured RDMA interfaces
-  - Typically "3" for RoCE (RDMA over Converged Ethernet)
-  - Must be a valid GID index for your InfiniBand/RoCE setup
-
-- **`interfaces`** (default: `["rocep28s0", "rocep62s0", "rocep79s0", "rocep96s0"]`)
-  - List of RDMA device names that should be present on all cluster nodes
-  - Examples:
-    - `["rocep28s0", "rocep62s0", "rocep79s0", "rocep96s0"]` - Standard 4-interface setup
-    - `["mlx5_0", "mlx5_1"]` - Mellanox 2-interface setup
-
-- **`ibv_test_timeout`** (default: 90)
-  - Timeout in seconds for each ibv_rc_pingpong connectivity test
-  - Integer value (seconds), used directly as configured
-  - Uses `ibv_rc_pingpong` (direct InfiniBand verbs) for RCCL-compatible testing
-  - Increase for slower networks or high-latency connections
-
-- **`ibv_test_port_range`** (default: "10000-50000")
-  - Port range for ibv_rc_pingpong tests to avoid conflicts
-  - Format: "start-end" (e.g., "10000-50000", "10000-10999")
-  - Ensure ports are not blocked by firewalls
-
-- **`inter_full_mesh_group_pairs_per_wave`** (default: "auto")
-  - Max ordered group-pairs (Gi→Gj keys) per wave during inter-group RDMA testing
-  - "auto" calculates as max(1, num_groups - 1)
-  - Can be set to a specific integer to control wave size and reduce memory/CPU load
-
-- **`prune_failure_threshold`** (default: 0.5)
-  - Prune nodes whose fraction of peers with ≥1 FAIL intra test is ≥ this value
-  - Range: 0.0 to 1.0 (0.5 = 50% failure threshold)
-  - Helps remove problematic nodes before inter-group testing
-  - Lower values (0.2-0.3) are more aggressive at removing problematic nodes
-
-- **`port_retry_max`** (default: 3)
-  - Max retry attempts for pairs whose logs show PORT_LISTEN_FAILED
-  - Range: 0-10 retries with new TCP ports after each wave
-  - Helps handle port conflicts during large-scale testing
-
-- **`port_retry_gap`** (default: 1000)
-  - Port gap when remapping ports for PORT_LISTEN_FAILED retries
-  - Range: 1-65535
-  - Starts at (max port in batch) + this gap to reduce overlap with ephemeral ports
-
-- **`exclude_failed_interface_nodes`** (default: "true")
-  - Legacy hint for reporting: preflight now prunes interface/GID-failed nodes automatically
-  - Interface failures are excluded from mesh testing regardless of this flag
-
-##### Legacy RDMA paths — deprecated
+### Legacy RDMA paths — deprecated
 
 Existing RDMA users may temporarily retain `node_check.gid_index` and
 `node_check.rdma_interfaces`. CVS normalizes them to
@@ -210,15 +134,12 @@ and emits a deprecation warning. If a legacy and canonical value are both
 present, they must match. New configurations should use the canonical RDMA
 paths; the compatibility paths will be removed in a future release.
 
-#### IFoE Settings (`connectivity_check.ifoe`) — MI4XX scale-up fabric
-
-IFoE validation is organized into fabric admission, strict L2 connectivity,
-and TransferBench data-path validation. CVS owns `afmctl` discovery, privilege
-handling, BDF and port discovery, strict coverage, traffic selection, timeout
-derivation, and result parsing.
+### IFoE settings that are no longer configurable
 
-The earlier configuration shape exposed those implementation details directly.
-They now follow this fixed policy:
+IFoE validation (`connectivity_check.ifoe`) used to expose implementation
+details -- fabric discovery, privilege handling, port selection, timeouts --
+directly as config keys. Those keys are no longer read; CVS now derives all
+of this automatically and follows a fixed policy:
 
 | Previous setting | Current CVS behavior |
 |---|---|
@@ -232,117 +153,6 @@ They now follow this fixed policy:
 | `loss_threshold_pct` | Fail on any reported loss or incomplete coverage |
 | `per_ping_timeout` / `ssh_timeout` | Derive conservative timeouts from the requested workload |
 
-- **`fabric_checks`** (default: `false`)
-  - Enables MI4XX-only AIFM/AFM/vPOD, station-mask, and IFoE port admission checks
-  - Set to `true` only on MI4XX systems; it remains disabled for MI3XX systems
-  - Requires `node_check.enabled: true`
-
-##### L2 ping (`connectivity_check.ifoe.l2ping`)
-
-Runs `afmctl test ping` with strict full-mesh coverage on every admitted IFoE
-port and validates per-port and aggregate summary accounting.
-
-- **`enabled`** (default: `false`)
-  - Enables the mandatory L2 connectivity gate before TransferBench and RDMA
-- **`pings_per_port`** (default: `3`)
-  - Number of ping samples sent per selected IFoE port pair
-
-##### TransferBench (`connectivity_check.ifoe.transferbench`)
-
-- **`enabled`** (default: `false`)
-  - Enables the TransferBench IFoE data-path gate before RDMA
-- **`scope`** (default: `"node"`)
-  - `"node"` runs an independent smoketest on each node
-  - `"cluster"` runs one multi-rank test across the admitted cluster
-- **`profile`** (default: `"smoketest"`)
-  - Selects the CVS-supported test profile; `"smoketest"` is currently supported
-- **`message_sizes`** (default: `["1K", "16M"]`)
-  - Message sizes exercised by the selected profile
-- **`iterations`** (default: `2`)
-  - Validated iterations per test and message size
-- **`warmup_iterations`** (default: `0`)
-  - Warmup iterations performed before validation
-
-#### Node Smoke Settings (`node_smoke`) — opt-in (Primus Tier 1)
-
-Runs Primus `node_smoke` on each reachable node via `primus-cli direct --single -- node_smoke`
-over parallel SSH (no Slurm required). Reference: Primus `docs/node-smoke-test-instruction.md`
-on branch `dev/preflight-direct-test`.
-
-- **`connectivity_mode`** (default: `"skip"`)
-  - `"run"` — execute node_smoke on every reachable node
-  - `"skip"` — preflight records a SKIPPED result and does not invoke Primus
-- **`auto_setup`** (default: `true`)
-  - Clone/update Primus and create the venv with minimal deps (ROCm PyTorch) before node_smoke
-- **`setup_timeout`** (default: `600`)
-  - SSH timeout (seconds) for the per-node Primus auto_setup step
-- **`force_reclone`** (default: `false`)
-  - Remove `primus_dir` and clone fresh on every run (destructive)
-- **`shared_install`** (default: `true`)
-  - Leader node clones/installs on shared NFS home; other nodes wait (recommended for shared home)
-- **`pip_install_mode`** (default: `"minimal"`)
-  - `"minimal"` — ROCm PyTorch only; `"requirements"` — `pip install -r requirements.txt`; `"skip"` — venv only
-- **`torch_pip_index_url`** (default: `"https://download.pytorch.org/whl/rocm6.2"`)
-  - PyTorch wheel index for minimal install; match your ROCm version
-- **`primus_git_url`** (default: `"https://github.com/AMD-AIG-AIMA/Primus.git"`)
-- **`primus_git_branch`** (default: `"dev/preflight-direct-test"`)
-- **`primus_git_recurse_submodules`** (default: `false`)
-- **`primus_dir`** (default: `"/home/{user-id}/INSTALL/Primus"`)
-  - Required when `connectivity_mode` is `"run"`; `{user-id}` is resolved at runtime
-- **`venv_activate`** (default: `"/home/{user-id}/envs/preflight/.venv/bin/activate"`)
-  - Required when `connectivity_mode` is `"run"`
-- **`gpus_per_node`** (default: `8`)
-- **`master_port`** (default: `1234`)
-- **`dump_path`** (default: `""`)
-  - Per-node smoke JSON output; empty uses `<reporting.artifacts_root_dir>/node_smoke`
-- **`expected_rdma_nics`** (default: `null`)
-  - Defaults to `len(node_check.rdma_interfaces)` when null
-- **`ulimit_l_min_gb`** (default: `32`) — FAIL below this memlock limit; `0` disables
-- **`shm_min_gb`** (default: `8`) — FAIL below this `/dev/shm` size; `0` disables
-- **`skip_dmesg`** (default: `false`)
-- **`allow_foreign_procs`** (default: `false`)
-- **`allowed_procs`** (default: `"gpuagent,rocm-smi-daemon,amd-smi,dcgm-exporter"`)
-- **`require_tools`** (default: `""`) — empty = warn only
-- **`nccl_socket_ifname`** / **`gloo_socket_ifname`** (default: `""`)
-- **`nccl_ib_hca`** (default: `""`) — defaults to comma-joined `node_check.rdma_interfaces`
-- **`nccl_ib_gid_index`** (default: `null`) — defaults to `node_check.gid_index`
-- **`ssh_timeout`** (default: `300`)
-- **`extra_args`** (default: `[]`) — additional flags forwarded to primus-cli
-
-#### Tier 2 perf sanity (`node_smoke.tier2_perf`) — optional
-
-When `tier2_perf` is `true`, preflight forwards `--tier2-perf` to Primus `node_smoke`, enabling all three Tier 2 checks on each node (same as `launch_nodesmoke_ssh.sh -- --tier2-perf`):
-
-1. **Large GEMM TFLOPS floor** — 8192³ bf16 `torch.matmul`; FAIL below `gemm_tflops_min` (default 600)
-2. **HBM D2D bandwidth** — 512 MB device-to-device copy; FAIL below `hbm_gbs_min` (default 2000 GB/s)
-3. **Local multi-GPU RCCL all-reduce** — node-local only; FAIL below `rccl_gbs_min` (default 100 GB/s)
-
-Set `NCCL_IB_HCA`, `NCCL_SOCKET_IFNAME`, and `NCCL_IB_GID_INDEX` (via `node_smoke` config or cluster `env_vars`) before enabling Tier 2 — RCCL init enumerates every transport even though the all-reduce is local-only.
-
-- **`tier2_perf`** (default: `false`) — master switch; maps to `--tier2-perf`
-- **`gemm_tflops_min`** (default: `600`) — `--gemm-tflops-min`
-- **`hbm_gbs_min`** (default: `2000`) — `--hbm-gbs-min`
-- **`rccl_gbs_min`** (default: `100`) — `--rccl-gbs-min`
-- **`rccl_size_mb`** (default: `64`) — `--rccl-size-mb`
-- **`rccl_timeout_sec`** (default: `120`) — `--rccl-timeout-sec`
-
-Tier 2 runs need a longer SSH budget; when `tier2_perf` is enabled the effective timeout is at least 600 seconds even if `ssh_timeout` is lower.
-
-### Reporting Settings (`reporting`)
-
-- **`generate_html_report`** (default: `true`)
-  - Whether to generate detailed HTML report
-  - Set to `false` to disable HTML report generation
-
-- **`artifacts_root_dir`** (default: `"/tmp/{user-id}/preflight"`)
-  - Root directory where preflight artifacts are saved
-  - Includes HTML reports and RDMA full_mesh workspace logs under `rdma_connectivity_workspace/`
-  - Must be writable by the user running the tests
-
-- **`generate_rdma_pairs_csv`** (default: `true`)
-  - Whether to generate CSV file with failed RDMA pairs alongside HTML report
-  - Set to `false` to disable CSV generation
-
 ## Usage Examples
 
 ### Basic 8-Node Cluster Check