- gpuBandwidthSocketGroups: a single GPU whose NUMA node fails to resolve
no longer collapses the whole per-socket nvbandwidth split into one
fallback pass — it now folds into the last resolved group instead,
preserving isolation for the sockets that did resolve.
- blackbox discoverMarkedTargets: skip mounting/unmounting devices that
already have a running worker on every 2s discovery tick. This was
observed hammering the same USB target continuously (mount+unmount
every ~2s for the whole session) and contending with the worker's own
sync cycle, plausibly explaining multi-minute sync cycles seen on a
real crash bundle.
- syncFilesystem now calls syscall.Sync() directly instead of spawning
/bin/sync per copied file; blackbox mounts removable targets with
-o sync so writes are durable without relying on the app-level sync as
the primary mechanism.
- New platform.SetSyncBracketHook / satJob.syncBracket: blocks (with a
bounded timeout) on blackbox actually reaching removable media right
before and right after a diagnostic's real load step (nvbandwidth,
memtester, stress-ng, dcgmi diag, nccl, smartctl/nvme self-test...),
instead of only firing a fire-and-forget kick after the job's own log
file is written. A crash mid-load now has durable evidence the load
started, not just whatever streamed to the RAM-backed export dir before
blackbox's next scheduled cycle.
Found investigating a real support bundle where blackbox's last
successful sync (19:55:25) predated both the previous job finishing and
the crashing nvbandwidth job starting (19:56:57) — none of the crash
window ever reached durable media.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
A crash mid-command (e.g. the nvbandwidth reboot) previously lost that
job's entire output: streamExecOutput only buffered stdout/stderr in
memory and the job's log file was written once, after the process
exited. It now also streams each line straight to that file as it
arrives, so whatever printed before a crash survives.
Root filesystem here is a tmpfs overlay (toram boot), so the only real
persistence boundary is blackbox's mirror to removable media, not the
local write itself. platform.SetJobBoundaryHook lets app wire a touch
of a small kick-file after each job's output is written; blackboxWorker
now polls that file's mtime alongside its normal adaptive timer and
syncs immediately on a kick instead of waiting out the current flush
period (up to 30s).
On multi-socket systems, run the NVIDIA bandwidth diagnostic once per
CPU socket before the all-GPU pass, so a crash confined to the all-GPU
run (with clean per-socket passes preceding it) isolates a cross-socket
peer-to-peer fault instead of leaving it conflated with a general
GPU/PCIe issue. Single-socket systems keep the original one-pass shape.
Also expand the support-bundle README with reference notes distilled
from a real analysis pass (BMC clock drift, "0/empty" tool output
meaning absent hardware rather than a fault, timestamp-matching before
assigning causality, and a normal-power-cycle SEL signature), plus a
step-by-step recipe for diagnosing an unexpected reboot/crash during a
specific test.
Storage acceptance runs already write disk-<prefix>-report.txt (per-drive
health + pseudographic resource bars) into the timestamped bee-sat run dir.
Also mirror it into techdump, a sibling of baseDir under the same export
dir, so a support bundle surfaces these reports alongside the rest of the
diagnostics without anyone having to dig into a specific bee-sat run dir.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
dcgmi discovery -l is a preflight/metadata step ahead of the real DCGM
diag jobs; a transient failure racing nv-hostengine startup shouldn't
flip the whole pack's status, so it's now marked informational with a
couple of retries. Separately, bound the fabricmanager/nvidia-dcgm
systemctl restart/start calls in bee-nvidia-load with a timeout so a
wedged unit (e.g. fabric training stuck on a bad NVSwitch fabric)
can't hang bee-nvidia.service forever and block dcgm from ever
starting.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
dcgmi diag's failures only surface a terse wrapper message ("Detected
driver major version 0 is not between the required versions 346 and
2000"), which doesn't say why. Reading DCGM's own nvvs source
(NvidiaValidationSuite.cpp::CheckDriverVersion) shows this fires when
dcgmSystem.GetDeviceAttributes(0, ...) comes back empty — the version
string is never actually malformed. Diagnosing further requires info
bee never collected: whether dcgmi discovery even sees GPU 0/1
correctly, and nvvs's own internal debug log (normally
/var/log/nvidia-dcgm/nvvs.log, which bee never captured).
Add a "dcgmi discovery -l" job before every DCGM diag invocation
(check-gpu-dcgm-l2, nvbandwidth, targeted_power, pulse_test,
targeted_stress), and route dcgmi diag through
"-v -d DEBUG --debugLogFile {{run_dir}}/..." so the full nvvs debug
log lands in the SAT run dir and gets picked up by the support bundle
automatically, instead of needing a live SSH session on hardware we
usually can't get a second look at.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Disk report now ends with a Conclusion section judging a drive NEW/USED
against loose thresholds (<110% capacity written, <210% read, <7d
uptime, <30 power cycles), listing which ones tripped. Data
Written/Read in the Usage section now scale to TB/PB via
formatBytesHuman instead of always printing raw GB. storageSATCommands
now runs smartctl with -i so SATA/SAS reports get Model/Serial/
Firmware/Capacity, which the Conclusion needs to evaluate the
write/read criteria (previously only -H -A was collected).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
smartctl -t short only launches the self-test and returns immediately
("Testing has begun"); unlike nvme device-self-test --wait, it has no
blocking mode. Validate/Load runs closed the task and produced reports
before the drive actually finished the test. Now poll smartctl -a until
the test completes (or times out) and report the real result.
Also add a per-disk "Resource" section with pseudographic progress bars
for uptime (vs 5y design life), bytes written (vs 1 DWPD x 5y budget),
and bytes read (percent from SMART attribute 242), all rendered in
human-scaled units (days/years, TB/PB) instead of raw hour/byte counts.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Check mode: read-only SMART/NVMe data collection, no self-test.
Load mode: same collection + short self-test (nvme device-self-test -s 1,
smartctl -t short). Card descriptions updated accordingly.
After each storage SAT run, a disk-N-devname-report.txt is written
per device into the runDir (auto-included in support bundles).
Web UI task page renders one card per disk directly below Task Report.
Also fixes pre-existing TestDashboardRendersRuntimeHealthTable failure:
test fixture used "inactive" status but code now treats inactive as OK
for completed oneshot services; updated to "failed" to match intent.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- sat.go: DetectGPUVendor lspci fallback now checks GPU device classes
([0300]/[0302]/[0380]) per line instead of scanning the whole output for
vendor name; AMD EPYC servers have dozens of AMD-branded PCIe entries
(Root Complex, IOMMU, Host Bridge) that were triggering the old check
- blackbox.go: fix deadlock in finishCycle — it held w.mu while calling
persistState(), which acquires rt.mu then re-acquires w.mu inside
persistStateLocked(); now w.mu is released before persistState()
- build.sh: remove NVIDIA-specific overlay files (bee-gpu-burn,
bee-john-gpu-stress, bee-nccl-gpu-stress, bee-nvidia-recover,
bee-dcgmproftester-staggered, bee-check-nvswitch,
nvidia-fabricmanager.service.d/) for non-nvidia build variants
- bee-selfheal: gate NVIDIA recovery on BEE_GPU_VENDOR=nvidia so the
script does not attempt to restart bee-nvidia.service on NOGPU builds
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
targeted_stress, targeted_power, and the Level 2/3 diag were dispatched
one GPU at a time from the UI, turning a single dcgmi command into 8
sequential ~350–450 s runs. DCGM supports -i with a comma-separated list
of GPU indices and runs the diagnostic on all of them in parallel.
Move nvidia, nvidia-targeted-stress, nvidia-targeted-power into
nvidiaAllGPUTargets so expandSATTarget passes all selected indices in one
API call. Simplify runNvidiaValidateSet to match runNvidiaFabricValidate.
Update sat.go constants and page_validate.go estimates to reflect all-GPU
simultaneous execution (remove n× multiplier from total time estimates).
Stress test on 8-GPU system: ~5.3 h → ~2.5 h.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
A single dcgmproftester process without -i only loads GPU 0 regardless of
CUDA_VISIBLE_DEVICES. Now always routes multi-GPU runs through
bee-dcgmproftester-staggered (--stagger-seconds 0 for parallel mode),
which spawns one process per GPU so all GPUs are loaded simultaneously.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add SATEstimated* constants to sat.go derived from _v8 production logs,
with a rule to recalculate them whenever the script changes
- Extend validateInventory with NvidiaGPUCount to make estimates GPU-aware
- Update all validate card duration strings: CPU, memory, storage, NVIDIA GPU,
targeted stress/power, pulse test, NCCL, nvbandwidth
- Fix nvbandwidth description ("intended to stay short" → actual ~45 min)
- Top-level profile labels show computed total including GPU count
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Benchmark page now shows two result sections: Performance (scores) and
Power / Thermal Fit (slot table). After any benchmark task completes
the results section auto-refreshes via GET /api/benchmark/results
without a full page reload.
- Power results table shows each GPU slot with nominal TDP, achieved
stable power limit, and P95 observed power. Rows with derated cards
are highlighted amber so under-performing slots stand out at a glance.
Older runs are collapsed in a <details> summary.
- memtester is now wrapped with timeout(1) so a stuck memory controller
cannot cause Validate Memory to hang indefinitely. Wall-clock limit is
~2.5 min per 100 MB per pass plus a 2-minute buffer.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- nvidia.go: add Name field to nvidiaGPUInfo, include model name in
nvidia-smi query, set dev.Model in enrichPCIeWithNVIDIAData
- pages.go: fix duplicate GPU count in validate card summary (4 GPU: 4 x …
→ 4 x … GPU); fix PSU UNKNOWN fallback from hw.PowerSupplies; treat
activating/deactivating/reloading service states as OK in Runtime Health
- support_bundle.go: use "150405" time format (no colons) for exFAT compat
- sat.go / benchmark.go / platform_stress.go / sat_fan_stress.go: remove
.tar.gz archive creation from export dirs — export packs everything itself
- charts_svg.go: add min-max downsampling (1400 pt cap) for SVG chart perf
- benchmark_report.go / sat.go: normalize GPU fallback to "Unknown GPU"
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Replace diag level 1-4 dropdown with Validate/Stress radio buttons
- Validate: dcgmi L2, 60s CPU, 256MB/1p memtester, SMART short
- Stress: dcgmi L3 + targeted_stress in Run All, 30min CPU, 1GB/3p memtester, SMART long/NVMe extended
- Parallel GPU mode: spawn single task for all GPUs instead of splitting per model
- Benchmark table: per-GPU columns for sequential runs, server-wide column for parallel
- Benchmark report converted to Markdown with server model, GPU model, version in header; only steady-state charts
- Fix IPMI power parsing in benchmark (was looking for 'Current Power', correct field is 'Instantaneous power reading')
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
nvvs (DCGM validation suite) survives when dcgmi is killed mid-run,
leaving the GPU occupied. The next dcgmi diag invocation then fails
with "affected resource is in use".
Two-part fix:
- Add nvvs and dcgmi to KillTestWorkers patterns so they are cleaned
up by the global cancel handler
- Call KillTestWorkers at the start of RunNvidiaTargetedStressValidatePack
to clear any stale processes before dcgmi diag runs
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
exec.CommandContext only kills the direct child (the shell script), leaving
grandchildren (john, gpu-burn, etc.) as orphans. Set Setpgid so each SAT
job runs in its own process group, then send SIGKILL to the whole group
(-pgid) in the Cancel hook.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
rvs was not in PATH so the stress job exited immediately (UNSUPPORTED).
Now resolveRVSCommand searches /opt/rocm-*/bin/rvs before failing.
Also add a Copy button overlay on all .terminal elements and set
user-select:text so logs can be copied from the web UI.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add rocm-validation-suite, rocblas, rocrand, hip-runtime-amd,
hipblaslt, comgr to ISO (~700MB, needed for HIP compute)
- RunAMDStressPack: run RVS GST (SGEMM ~31 TFLOPS/GPU) + bandwidth test
- Add rvs symlink in chroot setup hook
- Pin all new package versions in VERSIONS
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Task queue: all SAT/audit jobs enqueue and run one-at-a-time;
tasks persist past page navigation; new Tasks page with cancel/priority/log stream
- UI: consolidate nav (Validate, Burn, Tasks, Tools); Audit becomes modal;
Dashboard hardware summary badges + split metrics charts (load/temp/power);
Tools page consolidates network, services, install, support bundle
- AMD GPU: acceptance test and stress burn cards; GPU presence API greys
out irrelevant SAT cards automatically
- Burn tests: Memory Stress (stress-ng --vm), SAT Stress (stressapptest)
- Install to RAM: copies squashfs to /dev/shm, re-associates loop devices
via LOOP_CHANGE_FD ioctl so live media can be ejected
- Charts: relative time axis (0 = now, negative left)
- memtester: LimitMEMLOCK=infinity in bee-web.service; empty output → UNSUPPORTED
- SAT overlay applied dynamically on every /audit.json serve
- MIME panic guard for LiveCD ramdisk I/O errors
- ISO: add memtest86+, stressapptest packages; memtest86+ GRUB entry;
disable screensaver/DPMS in bee-openbox-session
- Unknown SAT status severity = 1 (does not override OK)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- New tui/sat_progress.go: polls {DefaultSATBaseDir}/{prefix}-*/verbose.log every 300ms and parses completed/in-progress steps
- Busy screen now shows each step as PASS lscpu (234ms) / FAIL stress-ng (60.0s) / ... sensors-after instead of just "Working..."
2. Test results shown on screen (instead of just "Archive written to /path")
- RunCPUAcceptancePackResult, RunMemoryAcceptancePackResult, RunStorageAcceptancePackResult, RunAMDAcceptancePackResult now read summary.txt from the run directory and return a formatted per-step result:
Run: 2025-03-25T10:00:00Z
PASS lscpu
PASS sensors-before
FAIL stress-ng
PASS sensors-after
Overall: FAILED (ok=3 failed=1)
3. AMD GPU SAT with auto-detection
- platform.System.DetectGPUVendor(): checks /dev/nvidia0 → "nvidia", /dev/kfd → "amd"
- platform.System.RunAMDAcceptancePack(): runs rocm-smi, rocm-smi --showallinfo, dmidecode
- GPU SAT (G key / GPU row enter) automatically routes to AMD or NVIDIA based on detected vendor
- "Run All" also auto-detects vendor
4. Panel detail view
- GPU detail now shows the most recent (NVIDIA or AMD) SAT result, whichever is newer
- All SAT detail views use the same human-readable formatSATDetail format
--start is not a valid nvme-cli flag; correct syntax is -s 1 (short test).
Add --wait so the command blocks until the test completes.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
BMC:
- collector/board.go: collectBMCFirmware() via ipmitool mc info, graceful skip if /dev/ipmi0 absent
- collector/collector.go: append BMC firmware record to snap.Firmware
- app/panel.go: show BMC version in TUI right-panel header alongside BIOS
CPU SAT:
- platform/sat.go: RunCPUAcceptancePack(baseDir, durationSec) — lscpu + sensors before/after + stress-ng
- app/app.go: RunCPUAcceptancePack + RunCPUAcceptancePackResult methods, satRunner interface updated
- app/panel.go: CPU row now reads real PASS/FAIL from cpu-*/summary.txt via satStatuses(); cpuDetailResult shows last SAT summary + audit data
- tui/types.go: actionRunCPUSAT, confirmBody for CPU test with mode label
- tui/screen_health_check.go: hcCPUDurations [60,300,900]s; hcRunSingle(CPU)→confirm screen; executeRunAll uses RunCPUAcceptancePackResult
- tui/forms.go: actionRunCPUSAT → RunCPUAcceptancePackResult with mode duration
- cmd/bee/main.go: bee sat cpu [--duration N] subcommand
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- TUI: duration presets (10m/1h/8h/24h), GPU multi-select checkboxes
- nvtop launched concurrently with SAT via tea.ExecProcess; can reopen or abort
- GPU metrics collected per-second during bee-gpu-stress (temp/usage/power/clock)
- Outputs: gpu-metrics.csv, gpu-metrics.html (offline SVG), gpu-metrics-term.txt
- Terminal chart: asciigraph-style line chart with box-drawing chars and ANSI colours
- AUDIT_VERSION bumped 0.1.1 → 1.0.0; nvtop added to ISO package list
- runtime-flows.md updated with full NVIDIA SAT TUI flow documentation
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
mchusandClaude Sonnet 5