Normalize task queue priorities by workflow
This commit is contained in:
@@ -36,6 +36,16 @@ var apiListNvidiaGPUStatuses = func(a *app.App) ([]platform.NvidiaGPUStatus, err
|
||||
return a.ListNvidiaGPUStatuses()
|
||||
}
|
||||
|
||||
const (
|
||||
taskPriorityBenchmark = 10
|
||||
taskPriorityBurn = 20
|
||||
taskPriorityValidateStress = 30
|
||||
taskPriorityValidate = 40
|
||||
taskPriorityAudit = 50
|
||||
taskPriorityInstallToRAM = 60
|
||||
taskPriorityInstall = 70
|
||||
)
|
||||
|
||||
// ── Job ID counter ────────────────────────────────────────────────────────────
|
||||
|
||||
var jobCounter atomic.Uint64
|
||||
@@ -109,6 +119,30 @@ func shouldSplitHomogeneousNvidiaTarget(target string) bool {
|
||||
}
|
||||
}
|
||||
|
||||
func defaultTaskPriority(target string, params taskParams) int {
|
||||
switch strings.TrimSpace(target) {
|
||||
case "install":
|
||||
return taskPriorityInstall
|
||||
case "install-to-ram":
|
||||
return taskPriorityInstallToRAM
|
||||
case "audit":
|
||||
return taskPriorityAudit
|
||||
case "nvidia-benchmark":
|
||||
return taskPriorityBenchmark
|
||||
case "nvidia-stress", "amd-stress", "memory-stress", "sat-stress", "platform-stress", "nvidia-compute":
|
||||
return taskPriorityBurn
|
||||
case "nvidia", "nvidia-targeted-stress", "nvidia-targeted-power", "nvidia-pulse",
|
||||
"nvidia-interconnect", "nvidia-bandwidth", "memory", "storage", "cpu",
|
||||
"amd", "amd-mem", "amd-bandwidth":
|
||||
if params.StressMode {
|
||||
return taskPriorityValidateStress
|
||||
}
|
||||
return taskPriorityValidate
|
||||
default:
|
||||
return 0
|
||||
}
|
||||
}
|
||||
|
||||
func expandHomogeneousNvidiaSelections(gpus []platform.NvidiaGPU, include, exclude []int) ([]nvidiaTaskSelection, error) {
|
||||
if len(gpus) == 0 {
|
||||
return nil, fmt.Errorf("no NVIDIA GPUs detected")
|
||||
@@ -458,6 +492,7 @@ func (h *handler) handleAPIAuditRun(w http.ResponseWriter, _ *http.Request) {
|
||||
ID: newJobID("audit"),
|
||||
Name: "Audit",
|
||||
Target: "audit",
|
||||
Priority: defaultTaskPriority("audit", taskParams{}),
|
||||
Status: TaskPending,
|
||||
CreatedAt: time.Now(),
|
||||
}
|
||||
@@ -526,7 +561,7 @@ func (h *handler) handleAPISATRun(target string) http.HandlerFunc {
|
||||
DisplayName: body.DisplayName,
|
||||
PlatformComponents: body.PlatformComponents,
|
||||
}
|
||||
tasks, err := buildNvidiaTaskSet(target, 0, time.Now(), params, name, h.opts.App, "sat-"+target)
|
||||
tasks, err := buildNvidiaTaskSet(target, defaultTaskPriority(target, params), time.Now(), params, name, h.opts.App, "sat-"+target)
|
||||
if err != nil {
|
||||
writeError(w, http.StatusBadRequest, err.Error())
|
||||
return
|
||||
@@ -613,7 +648,7 @@ func (h *handler) handleAPIBenchmarkNvidiaRun(w http.ResponseWriter, r *http.Req
|
||||
ID: newJobID("benchmark-nvidia"),
|
||||
Name: stepName,
|
||||
Target: "nvidia-benchmark",
|
||||
Priority: 15,
|
||||
Priority: defaultTaskPriority("nvidia-benchmark", taskParams{}),
|
||||
Status: TaskPending,
|
||||
CreatedAt: now,
|
||||
params: taskParams{
|
||||
@@ -645,7 +680,7 @@ func (h *handler) handleAPIBenchmarkNvidiaRun(w http.ResponseWriter, r *http.Req
|
||||
name = fmt.Sprintf("%s · sequential", name)
|
||||
}
|
||||
|
||||
tasks, err := buildNvidiaTaskSet("nvidia-benchmark", 15, time.Now(), taskParams{
|
||||
params := taskParams{
|
||||
GPUIndices: body.GPUIndices,
|
||||
ExcludeGPUIndices: body.ExcludeGPUIndices,
|
||||
SizeMB: body.SizeMB,
|
||||
@@ -653,7 +688,8 @@ func (h *handler) handleAPIBenchmarkNvidiaRun(w http.ResponseWriter, r *http.Req
|
||||
RunNCCL: runNCCL,
|
||||
ParallelGPUs: parallelGPUs,
|
||||
DisplayName: body.DisplayName,
|
||||
}, name, h.opts.App, "benchmark-nvidia")
|
||||
}
|
||||
tasks, err := buildNvidiaTaskSet("nvidia-benchmark", defaultTaskPriority("nvidia-benchmark", params), time.Now(), params, name, h.opts.App, "benchmark-nvidia")
|
||||
if err != nil {
|
||||
writeError(w, http.StatusBadRequest, err.Error())
|
||||
return
|
||||
@@ -1054,7 +1090,7 @@ func (h *handler) handleAPIInstallToRAM(w http.ResponseWriter, r *http.Request)
|
||||
ID: newJobID("install-to-ram"),
|
||||
Name: "Install to RAM",
|
||||
Target: "install-to-ram",
|
||||
Priority: 10,
|
||||
Priority: defaultTaskPriority("install-to-ram", taskParams{}),
|
||||
Status: TaskPending,
|
||||
CreatedAt: time.Now(),
|
||||
}
|
||||
@@ -1169,7 +1205,7 @@ func (h *handler) handleAPIInstallRun(w http.ResponseWriter, r *http.Request) {
|
||||
ID: newJobID("install"),
|
||||
Name: "Install to Disk",
|
||||
Target: "install",
|
||||
Priority: 20,
|
||||
Priority: defaultTaskPriority("install", taskParams{}),
|
||||
Status: TaskPending,
|
||||
CreatedAt: time.Now(),
|
||||
params: taskParams{
|
||||
@@ -1461,4 +1497,3 @@ func (h *handler) rollbackPendingNetworkChange() error {
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user