Skip to content

Fix Xid healthcheck for replicated GPUs #1992

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
abraaoteixeira wants to merge 1 commit into
base: main
Choose a base branch
from
+615 −31
Open

Fix Xid healthcheck for replicated GPUs #1992

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
67 changes: 36 additions & 31 deletions internal/rm/health.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ func (r *nvmlResourceManager) checkHealth(stop <-chan interface{}, devices Devic
_ = eventSet.Free()
}()

parentToDeviceMap := make(map[string]*Device)
parentToDeviceMap := make(map[string][]*Device)
deviceIDToGiMap := make(map[string]uint32)
deviceIDToCiMap := make(map[string]uint32)

Expand All @@ -85,30 +85,33 @@ func (r *nvmlResourceManager) checkHealth(stop <-chan interface{}, devices Devic
}
deviceIDToGiMap[d.ID] = gi
deviceIDToCiMap[d.ID] = ci
parentToDeviceMap[uuid] = d

gpu, ret := r.nvml.DeviceGetHandleByUUID(uuid)
if ret != nvml.SUCCESS {
klog.Infof("unable to get device handle from UUID: %v; marking it as unhealthy", ret)
unhealthy <- d
continue
}
if _, exists := parentToDeviceMap[uuid]; !exists {
gpu, ret := r.nvml.DeviceGetHandleByUUID(uuid)
if ret != nvml.SUCCESS {
klog.Infof("unable to get device handle from UUID: %v; marking it as unhealthy", ret)
unhealthy <- d
continue
}

supportedEvents, ret := gpu.GetSupportedEventTypes()
if ret != nvml.SUCCESS {
klog.Infof("unable to determine the supported events for %v: %v; marking it as unhealthy", d.ID, ret)
unhealthy <- d
continue
}
supportedEvents, ret := gpu.GetSupportedEventTypes()
if ret != nvml.SUCCESS {
klog.Infof("unable to determine the supported events for %v: %v; marking it as unhealthy", d.ID, ret)
unhealthy <- d
continue
}

ret = gpu.RegisterEvents(eventMask&supportedEvents, eventSet)
switch {
case ret == nvml.ERROR_NOT_SUPPORTED:
klog.Warningf("Device %v is too old to support healthchecking.", d.ID)
case ret != nvml.SUCCESS:
klog.Infof("Marking device %v as unhealthy: %v", d.ID, ret)
unhealthy <- d
ret = gpu.RegisterEvents(eventMask&supportedEvents, eventSet)
switch {
case ret == nvml.ERROR_NOT_SUPPORTED:
klog.Warningf("Device %v is too old to support healthchecking.", d.ID)
case ret != nvml.SUCCESS:
klog.Infof("Marking device %v as unhealthy: %v", d.ID, ret)
unhealthy <- d
}
}

parentToDeviceMap[uuid] = append(parentToDeviceMap[uuid], d)
}

for {
Expand Down Expand Up @@ -151,23 +154,25 @@ func (r *nvmlResourceManager) checkHealth(stop <-chan interface{}, devices Devic
continue
}

d, exists := parentToDeviceMap[eventUUID]
devs, exists := parentToDeviceMap[eventUUID]
if !exists {
klog.Infof("Ignoring event for unexpected device: %v", eventUUID)
continue
}

if d.IsMigDevice() && e.GpuInstanceId != 0xFFFFFFFF && e.ComputeInstanceId != 0xFFFFFFFF {
gi := deviceIDToGiMap[d.ID]
ci := deviceIDToCiMap[d.ID]
if gi != e.GpuInstanceId || ci != e.ComputeInstanceId {
continue
for _, d := range devs {
if d.IsMigDevice() && e.GpuInstanceId != 0xFFFFFFFF && e.ComputeInstanceId != 0xFFFFFFFF {
gi := deviceIDToGiMap[d.ID]
ci := deviceIDToCiMap[d.ID]
if gi != e.GpuInstanceId || ci != e.ComputeInstanceId {
continue
}
klog.Infof("Event for mig device %v (gi=%v, ci=%v)", d.ID, gi, ci)
}
klog.Infof("Event for mig device %v (gi=%v, ci=%v)", d.ID, gi, ci)
}

klog.Infof("XidCriticalError: Xid=%d on Device=%s; marking device as unhealthy.", e.EventData, d.ID)
unhealthy <- d
klog.Infof("XidCriticalError: Xid=%d on Device=%s; marking device as unhealthy.", e.EventData, d.ID)
unhealthy <- d
}
}
}

Expand Down
Loading