fix(webui): dispatch "scenario" tasks in the real task runner, collapse duplicate target switch
The Scenario page's Run button enqueued a task with target "scenario" that never launched any load: the queue runs each task in an external bee-worker subprocess (RunPersistedTask -> executeTaskWithOptions in task_runner.go), whose target switch had no "scenario" case, so the task died with "unknown target: scenario". The case had only been added to taskQueue.runTask's switch in tasks.go — a stale copy exercised solely by unit tests, which is why the tests passed while the button did nothing. Add the "scenario" case (ReadScenario -> ParseScenarioJSON -> RunScenario) to executeTaskWithOptions, and collapse runTask into a thin delegate to it so there is a single target dispatch. The old runTask switch had already drifted into a subset (missing nvme-format, *-write, raid-*, nvidia-config, ...); removing it eliminates the divergence that hid this bug. Also drop the now-orphaned taskQueue.statusDB helper. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Claude Opus 4.8
+36
-386
@@ -45,7 +45,7 @@ var taskNames = map[string]string{
|
||||
"nvidia-stress": "NVIDIA GPU Stress",
|
||||
"memory": "Memory SAT",
|
||||
"storage": "Storage SAT",
|
||||
"nvidia-config": "GPU Config & NVLink Check",
|
||||
"nvidia-config": "GPU Config & NVLink Check",
|
||||
"cpu": "CPU SAT",
|
||||
"amd": "AMD GPU SAT",
|
||||
"amd-mem": "AMD GPU MEM Integrity",
|
||||
@@ -121,34 +121,34 @@ type Task struct {
|
||||
|
||||
// taskParams holds optional parameters parsed from the run request.
|
||||
type taskParams struct {
|
||||
Duration int `json:"duration,omitempty"`
|
||||
StressMode bool `json:"stress_mode,omitempty"`
|
||||
GPUIndices []int `json:"gpu_indices,omitempty"`
|
||||
ExcludeGPUIndices []int `json:"exclude_gpu_indices,omitempty"`
|
||||
StaggerGPUStart bool `json:"stagger_gpu_start,omitempty"`
|
||||
SizeMB int `json:"size_mb,omitempty"`
|
||||
Passes int `json:"passes,omitempty"`
|
||||
Loader string `json:"loader,omitempty"`
|
||||
BurnProfile string `json:"burn_profile,omitempty"`
|
||||
BenchmarkProfile string `json:"benchmark_profile,omitempty"`
|
||||
BenchmarkKind string `json:"benchmark_kind,omitempty"`
|
||||
RunNCCL bool `json:"run_nccl,omitempty"`
|
||||
ParallelGPUs bool `json:"parallel_gpus,omitempty"`
|
||||
RampStep int `json:"ramp_step,omitempty"`
|
||||
RampTotal int `json:"ramp_total,omitempty"`
|
||||
RampRunID string `json:"ramp_run_id,omitempty"`
|
||||
DisplayName string `json:"display_name,omitempty"`
|
||||
Device string `json:"device,omitempty"` // for install
|
||||
LBAF int `json:"lbaf,omitempty"`
|
||||
PlatformComponents []string `json:"platform_components,omitempty"`
|
||||
SAADmiChanges []saaChange `json:"saa_dmi_changes,omitempty"`
|
||||
FRUChanges []fruChange `json:"fru_changes,omitempty"`
|
||||
HuaweiElabelChanges []huaweiChange `json:"huawei_elabel_changes,omitempty"`
|
||||
RAIDController int `json:"raid_controller,omitempty"`
|
||||
RAIDDevices []string `json:"raid_devices,omitempty"`
|
||||
RAIDArrayName string `json:"raid_array_name,omitempty"`
|
||||
RAIDSlot string `json:"raid_slot,omitempty"`
|
||||
ScenarioName string `json:"scenario_name,omitempty"`
|
||||
Duration int `json:"duration,omitempty"`
|
||||
StressMode bool `json:"stress_mode,omitempty"`
|
||||
GPUIndices []int `json:"gpu_indices,omitempty"`
|
||||
ExcludeGPUIndices []int `json:"exclude_gpu_indices,omitempty"`
|
||||
StaggerGPUStart bool `json:"stagger_gpu_start,omitempty"`
|
||||
SizeMB int `json:"size_mb,omitempty"`
|
||||
Passes int `json:"passes,omitempty"`
|
||||
Loader string `json:"loader,omitempty"`
|
||||
BurnProfile string `json:"burn_profile,omitempty"`
|
||||
BenchmarkProfile string `json:"benchmark_profile,omitempty"`
|
||||
BenchmarkKind string `json:"benchmark_kind,omitempty"`
|
||||
RunNCCL bool `json:"run_nccl,omitempty"`
|
||||
ParallelGPUs bool `json:"parallel_gpus,omitempty"`
|
||||
RampStep int `json:"ramp_step,omitempty"`
|
||||
RampTotal int `json:"ramp_total,omitempty"`
|
||||
RampRunID string `json:"ramp_run_id,omitempty"`
|
||||
DisplayName string `json:"display_name,omitempty"`
|
||||
Device string `json:"device,omitempty"` // for install
|
||||
LBAF int `json:"lbaf,omitempty"`
|
||||
PlatformComponents []string `json:"platform_components,omitempty"`
|
||||
SAADmiChanges []saaChange `json:"saa_dmi_changes,omitempty"`
|
||||
FRUChanges []fruChange `json:"fru_changes,omitempty"`
|
||||
HuaweiElabelChanges []huaweiChange `json:"huawei_elabel_changes,omitempty"`
|
||||
RAIDController int `json:"raid_controller,omitempty"`
|
||||
RAIDDevices []string `json:"raid_devices,omitempty"`
|
||||
RAIDArrayName string `json:"raid_array_name,omitempty"`
|
||||
RAIDSlot string `json:"raid_slot,omitempty"`
|
||||
ScenarioName string `json:"scenario_name,omitempty"`
|
||||
}
|
||||
|
||||
type persistedTask struct {
|
||||
@@ -846,364 +846,14 @@ func setCPUGovernor(governor string) {
|
||||
}
|
||||
}
|
||||
|
||||
// runTask executes the work for a task, writing output to j.
|
||||
// runTask executes the work for a task, writing output to j. It delegates
|
||||
// to executeTaskWithOptions — the exact same target dispatch the external
|
||||
// task runner (RunPersistedTask, see task_runner.go) uses in its own
|
||||
// subprocess — so the in-process path (recovered tasks, unit tests) and the
|
||||
// subprocess path can never drift to different target coverage. Adding a new
|
||||
// task target means editing one switch, not two.
|
||||
func (q *taskQueue) runTask(t *Task, j *jobState, ctx context.Context) {
|
||||
if q.opts == nil {
|
||||
j.append("ERROR: handler options not configured")
|
||||
j.finish("handler options not configured")
|
||||
return
|
||||
}
|
||||
a := q.opts.App
|
||||
|
||||
recovered := len(j.lines) > 0
|
||||
j.append(fmt.Sprintf("Starting %s...", t.Name))
|
||||
if recovered {
|
||||
j.append(fmt.Sprintf("Recovered after bee-web restart at %s", time.Now().UTC().Format(time.RFC3339)))
|
||||
}
|
||||
|
||||
var (
|
||||
archive string
|
||||
err error
|
||||
)
|
||||
|
||||
switch t.Target {
|
||||
case "nvidia":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
diagLevel := 2
|
||||
if t.params.StressMode {
|
||||
diagLevel = 3
|
||||
}
|
||||
if len(t.params.GPUIndices) > 0 || diagLevel > 0 {
|
||||
result, e := a.RunNvidiaAcceptancePackWithOptions(
|
||||
ctx, "", diagLevel, t.params.GPUIndices, j.append,
|
||||
)
|
||||
if e != nil {
|
||||
err = e
|
||||
} else {
|
||||
archive = result.Body
|
||||
}
|
||||
} else {
|
||||
archive, err = a.RunNvidiaAcceptancePack("", j.append)
|
||||
}
|
||||
case "nvidia-targeted-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if dur <= 0 {
|
||||
dur = 300
|
||||
}
|
||||
archive, err = a.RunNvidiaTargetedStressValidatePack(ctx, "", dur, t.params.GPUIndices, j.append)
|
||||
case "nvidia-bench-perf":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = a.RunNvidiaBenchmarkCtx(ctx, "", platform.NvidiaBenchmarkOptions{
|
||||
Profile: t.params.BenchmarkProfile,
|
||||
SizeMB: t.params.SizeMB,
|
||||
GPUIndices: t.params.GPUIndices,
|
||||
ExcludeGPUIndices: t.params.ExcludeGPUIndices,
|
||||
RunNCCL: t.params.RunNCCL,
|
||||
ParallelGPUs: t.params.ParallelGPUs,
|
||||
RampStep: t.params.RampStep,
|
||||
RampTotal: t.params.RampTotal,
|
||||
RampRunID: t.params.RampRunID,
|
||||
}, j.append)
|
||||
case "nvidia-bench-power":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = a.RunNvidiaPowerBenchCtx(ctx, app.DefaultBeeBenchPowerDir, platform.NvidiaBenchmarkOptions{
|
||||
Profile: t.params.BenchmarkProfile,
|
||||
GPUIndices: t.params.GPUIndices,
|
||||
ExcludeGPUIndices: t.params.ExcludeGPUIndices,
|
||||
RampStep: t.params.RampStep,
|
||||
RampTotal: t.params.RampTotal,
|
||||
RampRunID: t.params.RampRunID,
|
||||
}, j.append)
|
||||
case "nvidia-bench-autotune":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = a.RunNvidiaPowerSourceAutotuneCtx(ctx, app.DefaultBeeBenchAutotuneDir, platform.NvidiaBenchmarkOptions{
|
||||
Profile: t.params.BenchmarkProfile,
|
||||
SizeMB: t.params.SizeMB,
|
||||
}, t.params.BenchmarkKind, j.append)
|
||||
case "nvidia-compute":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
rampPlan, planErr := resolveNvidiaRampPlan(t.params.BurnProfile, t.params.StaggerGPUStart, t.params.GPUIndices)
|
||||
if planErr != nil {
|
||||
err = planErr
|
||||
break
|
||||
}
|
||||
if t.params.BurnProfile != "" && t.params.StaggerGPUStart && dur <= 0 {
|
||||
dur = rampPlan.DurationSec
|
||||
}
|
||||
if rampPlan.StaggerSeconds > 0 {
|
||||
j.append(fmt.Sprintf("NVIDIA staggered ramp-up enabled: %ds per GPU; post-ramp hold: %ds; total runtime: %ds", rampPlan.StaggerSeconds, dur, rampPlan.TotalDurationSec))
|
||||
}
|
||||
archive, err = a.RunNvidiaOfficialComputePack(ctx, "", dur, t.params.GPUIndices, rampPlan.StaggerSeconds, j.append)
|
||||
case "nvidia-targeted-power":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
archive, err = a.RunNvidiaTargetedPowerPack(ctx, "", dur, t.params.GPUIndices, j.append)
|
||||
case "nvidia-pulse":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
archive, err = a.RunNvidiaPulseTestPack(ctx, "", dur, t.params.GPUIndices, j.append)
|
||||
case "nvidia-bandwidth":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = a.RunNvidiaBandwidthPack(ctx, "", t.params.GPUIndices, j.append)
|
||||
case "nvidia-interconnect":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = a.RunNCCLTests(ctx, "", t.params.GPUIndices, j.append)
|
||||
case "scenario":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
var data []byte
|
||||
data, err = a.ReadScenario(t.params.ScenarioName)
|
||||
if err != nil {
|
||||
break
|
||||
}
|
||||
var spec platform.ScenarioSpec
|
||||
spec, err = platform.ParseScenarioJSON(data)
|
||||
if err != nil {
|
||||
break
|
||||
}
|
||||
archive, err = a.RunScenario(ctx, "", spec, j.append)
|
||||
case "nvidia-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
rampPlan, planErr := resolveNvidiaRampPlan(t.params.BurnProfile, t.params.StaggerGPUStart, t.params.GPUIndices)
|
||||
if planErr != nil {
|
||||
err = planErr
|
||||
break
|
||||
}
|
||||
if t.params.BurnProfile != "" && t.params.StaggerGPUStart && dur <= 0 {
|
||||
dur = rampPlan.DurationSec
|
||||
}
|
||||
if rampPlan.StaggerSeconds > 0 {
|
||||
j.append(fmt.Sprintf("NVIDIA staggered ramp-up enabled: %ds per GPU; post-ramp hold: %ds; total runtime: %ds", rampPlan.StaggerSeconds, dur, rampPlan.TotalDurationSec))
|
||||
}
|
||||
archive, err = runNvidiaStressPackCtx(a, ctx, "", platform.NvidiaStressOptions{
|
||||
DurationSec: dur,
|
||||
Loader: t.params.Loader,
|
||||
GPUIndices: t.params.GPUIndices,
|
||||
ExcludeGPUIndices: t.params.ExcludeGPUIndices,
|
||||
StaggerSeconds: rampPlan.StaggerSeconds,
|
||||
}, j.append)
|
||||
case "memory":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
sizeMB, passes := resolveMemoryValidatePreset(t.params.BurnProfile, t.params.StressMode)
|
||||
j.append(fmt.Sprintf("Memory validate preset: %d MB x %d pass(es)", sizeMB, passes))
|
||||
archive, err = runMemoryAcceptancePackCtx(a, ctx, "", sizeMB, passes, j.append)
|
||||
case "storage":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = runStorageAcceptancePackCtx(a, ctx, "", t.params.StressMode, j.append)
|
||||
case "nvidia-config":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = runNvidiaConfigCheckPackCtx(a, ctx, "", j.append)
|
||||
case "cpu":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
if dur <= 0 {
|
||||
if t.params.StressMode {
|
||||
dur = 1800
|
||||
} else {
|
||||
dur = 60
|
||||
}
|
||||
}
|
||||
j.append(fmt.Sprintf("CPU stress duration: %ds", dur))
|
||||
archive, err = runCPUAcceptancePackCtx(a, ctx, "", dur, j.append)
|
||||
case "amd":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = runAMDAcceptancePackCtx(a, ctx, "", j.append)
|
||||
case "amd-mem":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = runAMDMemIntegrityPackCtx(a, ctx, "", j.append)
|
||||
case "amd-bandwidth":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
archive, err = runAMDMemBandwidthPackCtx(a, ctx, "", j.append)
|
||||
case "amd-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
archive, err = runAMDStressPackCtx(a, ctx, "", dur, j.append)
|
||||
case "memory-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
archive, err = runMemoryStressPackCtx(a, ctx, "", dur, j.append)
|
||||
case "sat-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
dur := t.params.Duration
|
||||
if t.params.BurnProfile != "" && dur <= 0 {
|
||||
dur = resolveBurnPreset(t.params.BurnProfile).DurationSec
|
||||
}
|
||||
archive, err = runSATStressPackCtx(a, ctx, "", dur, j.append)
|
||||
case "platform-stress":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
opts := resolvePlatformStressPreset(t.params.BurnProfile)
|
||||
opts.Components = t.params.PlatformComponents
|
||||
archive, err = a.RunPlatformStress(ctx, "", opts, j.append)
|
||||
case "audit":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
result, e := a.RunAuditNow(q.opts.RuntimeMode)
|
||||
if e != nil {
|
||||
err = e
|
||||
} else {
|
||||
for _, line := range splitLines(result.Body) {
|
||||
j.append(line)
|
||||
}
|
||||
}
|
||||
case "support-bundle":
|
||||
j.append("Building support bundle...")
|
||||
archive, err = buildSupportBundle(q.opts.ExportDir)
|
||||
case "install":
|
||||
if strings.TrimSpace(t.params.Device) == "" {
|
||||
err = fmt.Errorf("device is required")
|
||||
break
|
||||
}
|
||||
installLogPath := platform.InstallLogPath(t.params.Device)
|
||||
j.append("Install log: " + installLogPath)
|
||||
err = streamCmdJob(j, installCommand(ctx, t.params.Device, installLogPath))
|
||||
case "install-to-ram":
|
||||
if a == nil {
|
||||
err = fmt.Errorf("app not configured")
|
||||
break
|
||||
}
|
||||
err = a.RunInstallToRAM(ctx, j.append)
|
||||
default:
|
||||
j.append("ERROR: unknown target: " + t.Target)
|
||||
j.finish("unknown target")
|
||||
return
|
||||
}
|
||||
|
||||
// If the SAT archive was produced, check overall_status and write to component DB.
|
||||
if archive != "" {
|
||||
archivePath := app.ExtractArchivePath(archive)
|
||||
if err == nil {
|
||||
if app.ReadSATOverallStatus(archivePath) == "FAILED" {
|
||||
if reason := app.SATFailureDetail(archivePath); reason != "" {
|
||||
err = fmt.Errorf("SAT FAILED: %s", reason)
|
||||
} else {
|
||||
err = fmt.Errorf("SAT overall_status=FAILED (see summary.txt)")
|
||||
}
|
||||
}
|
||||
}
|
||||
// A user-aborted run (ctx canceled) may still have produced a partial
|
||||
// archive/summary.txt — that incomplete result must not overwrite the
|
||||
// component status DB, which is why this is skipped here rather than
|
||||
// relying on satKeyStatus's PARTIAL/UNSUPPORTED handling below.
|
||||
if db := q.statusDB(); db != nil && ctx.Err() == nil {
|
||||
app.ApplySATResultToDB(db, t.Target, archivePath)
|
||||
}
|
||||
}
|
||||
|
||||
if err != nil {
|
||||
if ctx.Err() != nil {
|
||||
j.append("Aborted.")
|
||||
j.finish("aborted")
|
||||
} else {
|
||||
j.append("ERROR: " + err.Error())
|
||||
j.finish(err.Error())
|
||||
}
|
||||
return
|
||||
}
|
||||
if archive != "" {
|
||||
j.append("Archive: " + archive)
|
||||
}
|
||||
j.finish("")
|
||||
}
|
||||
|
||||
func (q *taskQueue) statusDB() *app.ComponentStatusDB {
|
||||
if q.opts == nil || q.opts.App == nil {
|
||||
return nil
|
||||
}
|
||||
return q.opts.App.StatusDB
|
||||
executeTaskWithOptions(q.opts, t, j, ctx)
|
||||
}
|
||||
|
||||
func splitLines(s string) []string {
|
||||
|
||||
Reference in New Issue
Block a user