Refactor health checks by ArangoGutierrez · Pull Request #1601 · NVIDIA/k8s-device-plugin

ArangoGutierrez · 2026-01-26T12:02:09Z

No description provided.

Copilot

Pull request overview

Refactors the device health-check lifecycle to use context.Context cancellation instead of a stop channel, and adds NVML mock infrastructure/tests to validate health-check behavior.

Changes:

Switch ResourceManager.CheckHealth to CheckHealth(ctx context.Context, unhealthy chan<- *Device) and propagate through plugin + managers.
Refactor NVML health monitoring into a provider (nvmlHealthProvider) and add tests covering XID handling and ERROR_NOT_SUPPORTED behavior.
Vendor in NVML mock packages (including a dgxa100 mock server) to support deterministic health-check testing.

Reviewed changes

Copilot reviewed 7 out of 20 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
`internal/rm/rm.go`	Updates `ResourceManager` interface to use `context.Context` for health checks.
`internal/rm/nvml_manager.go`	Adapts NVML RM `CheckHealth` to the new context-based signature.
`internal/rm/tegra_manager.go`	Updates Tegra RM `CheckHealth` signature (still effectively disabled).
`internal/rm/health.go`	Refactors health-check implementation into `nvmlHealthProvider` and context-driven loop.
`internal/rm/health_test.go`	Adds tests using NVML mocks to validate event-driven health behavior.
`internal/rm/rm_mock.go`	Regenerates moq mock to match the new `CheckHealth` signature.
`internal/plugin/server.go`	Replaces stop channel with cancelable context + waitgroup for health goroutine lifecycle.
`vendor/modules.txt`	Adds vendored paths for newly included NVML mock packages.
`vendor/github.com/NVIDIA/go-nvml/pkg/nvml/mock/*.go`	Adds generated moq mocks for NVML interfaces used in tests.
`vendor/github.com/NVIDIA/go-nvml/pkg/nvml/mock/dgxa100/mig-profile.go`	Adds MIG profile/placement data for the dgxa100 mock server.
`vendor/github.com/NVIDIA/go-nvml/pkg/nvml/mock/dgxa100/dgxa100.go`	Adds dgxa100 NVML mock server used by health-check tests.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

internal/plugin/server.go

elezar

Thanks @ArangoGutierrez. I have a minor question regarding the cleaning up of the context.

elezar · 2026-01-28T07:39:51Z

internal/plugin/server.go

+	// Capture references at start to avoid race with cleanup() which may nil these fields.
+	healthCtx := plugin.healthCtx
+	health := plugin.health


Do we need to add synchronisation to ensure that these are actually initialised by the time that ListAndWatch is called, or is it guaranteed that this will be called only after the context has been initialized. Looking at the cleanup code, it is not clear why we would set plugin.healthCtx to nil there.

Note that if we were to pass the context in to the server construction, then we would not have to do this since the lifetime would be controlled at the caller and we would not set the context to nil in plugin cleanup.

Addressed, now healthCtx and healthCancel now created in devicePluginForResource()

elezar · 2026-01-28T07:41:02Z

internal/plugin/server.go

+	plugin.healthCtx = nil
+	plugin.healthCancel = nil


Why are we setting these to nil now?

removed.
Fields preserved (not niled) for restart support

elezar · 2026-01-28T07:50:21Z

internal/plugin/server.go

+		if err != nil && !errors.Is(err, context.Canceled) {
 			klog.Errorf("Failed to start health check: %v; continuing with health checks disabled", err)
 		}


What about logging the happy path case? (Also in chich cases will we return contex.Cancelled?

Suggested change

if err != nil && !errors.Is(err, context.Canceled) {

klog.Errorf("Failed to start health check: %v; continuing with health checks disabled", err)

}

switch {

case err == nil:

fallthrough

case errors.Is(err, context.Cancelled):

klog.Infof("Stopping health check")

case err != nil:

klog.Errorf("Failed to start health check: %v; continuing with health checks disabled", err)

}

}

Initialize healthCtx and healthCancel in devicePluginForResource() instead of in initialize() to eliminate race condition where ListAndWatch() might access healthCtx before initialization completes. This addresses Elezar's concern about synchronization guarantees and follows Go best practices for immutable initialization. Refs: NVIDIA#1601 Task: 1/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Remove healthCtx and healthCancel initialization from initialize() since they are now initialized at construction time in devicePluginForResource(). This eliminates redundant initialization and ensures context lifetime is tied to the plugin instance rather than the start/stop cycle. Refs: NVIDIA#1601 Task: 2/6

Recreate healthCtx and healthCancel in cleanup() after cancellation to support plugin restart. Remove nil assignments for these fields as they need to persist across restart cycles. This addresses Elezar's concern NVIDIA#2 about why we nil these fields - we no longer do. The context is recreated fresh for each restart, ensuring health checks work correctly when the plugin is restarted. Fixes plugin restart blocker identified in architecture review. Refs: NVIDIA#1601 Task: 3/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Close the health channel in cleanup() before niling to prevent panics in ListAndWatch() if cleanup happens during channel read. The WaitGroup ensures the health goroutine has completed before we close the channel, making this operation safe. Fixes critical blocker: health channel was never closed, leading to potential panics and resource leaks. Refs: NVIDIA#1601 Task: 4/6

Add ok check when receiving from health channel to gracefully handle channel closure during cleanup. This prevents potential panics and ensures clean shutdown when the plugin stops. Reading from a closed channel returns zero value and ok=false, which we now check and return gracefully. Refs: NVIDIA#1601 Task: 5/6

Initialize healthCtx and healthCancel in devicePluginForResource() instead of in initialize() to eliminate race condition where ListAndWatch() might access healthCtx before initialization completes. This addresses Elezar's concern about synchronization guarantees and follows Go best practices for immutable initialization. Refs: NVIDIA#1601 Task: 1/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Remove healthCtx and healthCancel initialization from initialize() since they are now initialized at construction time in devicePluginForResource(). This eliminates redundant initialization and ensures context lifetime is tied to the plugin instance rather than the start/stop cycle. Refs: NVIDIA#1601 Task: 2/6 Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Recreate healthCtx and healthCancel in cleanup() after cancellation to support plugin restart. Remove nil assignments for these fields as they need to persist across restart cycles. This addresses Elezar's concern NVIDIA#2 about why we nil these fields - we no longer do. The context is recreated fresh for each restart, ensuring health checks work correctly when the plugin is restarted. Fixes plugin restart blocker identified in architecture review. Refs: NVIDIA#1601 Task: 3/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Close the health channel in cleanup() before niling to prevent panics in ListAndWatch() if cleanup happens during channel read. The WaitGroup ensures the health goroutine has completed before we close the channel, making this operation safe. Fixes critical blocker: health channel was never closed, leading to potential panics and resource leaks. Refs: NVIDIA#1601 Task: 4/6 Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Add ok check when receiving from health channel to gracefully handle channel closure during cleanup. This prevents potential panics and ensures clean shutdown when the plugin stops. Reading from a closed channel returns zero value and ok=false, which we now check and return gracefully. Refs: NVIDIA#1601 Task: 5/6 Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

elezar · 2026-01-29T10:28:14Z

internal/plugin/server.go

+	// healthCtx and healthCancel control the health check goroutine lifecycle.
+	// healthWg is used to wait for the health check goroutine to complete during cleanup.
+	healthCtx    context.Context
+	healthCancel context.CancelFunc
+	healthWg     sync.WaitGroup


Suggested change

// healthCtx and healthCancel control the health check goroutine lifecycle.

// healthWg is used to wait for the health check goroutine to complete during cleanup.

healthCtx context.Context

healthCancel context.CancelFunc

healthWg sync.WaitGroup

// healthCtx and healthCancel control the health check goroutine lifecycle.

healthCtx context.Context

healthCancel context.CancelFunc

// healthWg is used to wait for the health check goroutine to complete during cleanup.

healthWg sync.WaitGroup

Thanks will implement

elezar · 2026-01-29T10:29:40Z

internal/plugin/server.go

 // nvidiaDevicePlugin implements the Kubernetes device plugin API
 type nvidiaDevicePlugin struct {
 	pluginapi.UnimplementedDevicePluginServer
 	ctx                  context.Context


Question: Why do we need a second context?

So we can cancel the healthchecks independently.

elezar · 2026-01-29T10:31:23Z

internal/plugin/server.go

+		// Recreate context for potential plugin restart. The same plugin instance
+		// may be restarted via Start() after Stop(), so we need a fresh context.
+		plugin.healthCtx, plugin.healthCancel = context.WithCancel(plugin.ctx)


When are we ever setting these to nil now? This should ONLY be done in the constructor.

Addressed by da507ef

Initialize healthCtx and healthCancel in devicePluginForResource() instead of in initialize() to eliminate race condition where ListAndWatch() might access healthCtx before initialization completes. This addresses Elezar's concern about synchronization guarantees and follows Go best practices for immutable initialization. Refs: NVIDIA#1601 Task: 1/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Remove healthCtx and healthCancel initialization from initialize() since they are now initialized at construction time in devicePluginForResource(). This eliminates redundant initialization and ensures context lifetime is tied to the plugin instance rather than the start/stop cycle. Refs: NVIDIA#1601 Task: 2/6 Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Recreate healthCtx and healthCancel in cleanup() after cancellation to support plugin restart. Remove nil assignments for these fields as they need to persist across restart cycles. This addresses Elezar's concern NVIDIA#2 about why we nil these fields - we no longer do. The context is recreated fresh for each restart, ensuring health checks work correctly when the plugin is restarted. Fixes plugin restart blocker identified in architecture review. Refs: NVIDIA#1601 Task: 3/6 Co-authored-by: Evan Lezar <elezar@nvidia.com> Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Add ok check when receiving from health channel to gracefully handle channel closure during cleanup. This prevents potential panics and ensures clean shutdown when the plugin stops. Reading from a closed channel returns zero value and ok=false, which we now check and return gracefully. Refs: NVIDIA#1601 Task: 5/6 Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

cdesiniotis · 2026-02-11T17:47:41Z

internal/rm/health.go

-	eventMask := uint64(nvml.EventTypeXidCriticalError | nvml.EventTypeDoubleBitEccError | nvml.EventTypeSingleBitEccError)
 	for _, d := range devices {
-		uuid, gi, ci, err := r.getDevicePlacement(d)
+		uuid, gi, ci, err := (&withDevicePlacements{r.nvml}).getDevicePlacement(d)


Could you provide some context as to why we landed on this change / refactor?

The goal of this refactor is to take the DevicePlugin to a similar state as our DRA after @guptaNswati Health check work.
This in order to prepare both DRA and DevicePlugin for NVsentinel integration via the Device-API

Lets add a detailed comment on how it will help. I also had to look into it. Because rn when we are only using NVML, its not making sense to decouple DevicePlacement logic from NVMLResourceManager. We need to note how this placement code can be reused with multiple providers and not just NVML.

type nvmlResourceManager struct { resourceManager nvml nvml.Interface

to having a new wrapper

type withDevicePlacements struct { nvml.Interface }

cdesiniotis · 2026-02-11T17:59:38Z

internal/rm/health.go

@@ -107,14 +137,16 @@ func (r *nvmlResourceManager) checkHealth(stop <-chan interface{}, devices Devic
 			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)


nit: could we update the log message here to indicate why we are marking the device unhealthy? (we failed during NVML event registration)

+1. on dra, its unable to register events for %s: %v; marking it as unhealthy", d.ID, ret

cdesiniotis · 2026-02-11T18:16:20Z

AGENTS.md

Could you explain how you are using this file / how it was generated? If you used a coding agent, could you acknowledge that in the PR description and describe how you used it?

This file needs to be removed before we can merge this PR.

guptaNswati · 2026-02-11T20:26:52Z

internal/plugin/server.go

 		return nil
 	}
+	// We stop the health checks
+	plugin.healthCancel()


nit: add a nil check
if plugin.healthCancel != nil { plugin.healthCancel() }

guptaNswati · 2026-02-11T20:33:18Z

internal/plugin/server.go

 	}
 	klog.Infof("Registered device plugin for '%s' with Kubelet", plugin.rm.Resource())

+	plugin.health = make(chan *rm.Device)


should these be initialized before registering the plugin?

guptaNswati · 2026-02-11T20:40:00Z

internal/plugin/server.go

+	plugin.healthCancel()
+	plugin.healthWg.Wait()
+	// Close the health channel after waiting for the health check goroutine to terminate.
+	close(plugin.health)


nil check here also.
if plugin.health != nil { close(plugin.health); plugin.health = nil }

internal/plugin/server.go

guptaNswati · 2026-02-11T20:48:21Z

internal/plugin/server.go

-		case d := <-plugin.health:
+		case d, ok := <-plugin.health:
+			if !ok {
+				// Health channel closed, health checks stopped


i would log here also. helps in debugging

guptaNswati · 2026-02-11T20:51:08Z

internal/rm/health.go

 		_ = eventSet.Free()
 	}()

+	// Construct the device maps.


oh you already called it out. Yes, we should refractor it into helpers

guptaNswati · 2026-02-11T20:57:57Z

internal/rm/health.go

 	envEnableHealthChecks = "DP_ENABLE_HEALTHCHECKS"
 )

+type nvmlHealthProvider struct {


why not nvmlDeviceHealthProvider

guptaNswati · 2026-02-11T21:13:59Z

internal/rm/health.go


 // getMigDeviceParts returns the parent GI and CI ids of the MIG device.
-func (r *nvmlResourceManager) getMigDeviceParts(d *Device) (string, uint32, uint32, error) {
+func (r *withDevicePlacements) getMigDeviceParts(d *Device) (string, uint32, uint32, error) {


nit: can we call it dp or d instead of r

guptaNswati · 2026-02-11T21:29:54Z

internal/rm/health.go

+type nvmlHealthProvider struct {
+	nvmllib           nvml.Interface
+	devices           Devices
+	parentToDeviceMap map[string]*Device


can we make it a type similar to type devicePlacementMap map[string]map[uint32]map[uint32]*Device
https://github.com/NVIDIA/k8s-dra-driver-gpu/blob/main/cmd/gpu-kubelet-plugin/device_health.go#L36

guptaNswati · 2026-02-11T22:01:06Z

internal/rm/health.go

+}

-		gpu, ret := r.nvml.DeviceGetHandleByUUID(uuid)
+func (r *nvmlHealthProvider) registerDeviceEvents(eventSet nvml.EventSet) {


nit: r -> h or hp

guptaNswati · 2026-02-11T22:01:52Z

internal/rm/health.go

+	}
+
+	p := nvmlHealthProvider{
+		nvmllib:           r.nvml,


is this needed for tests?

@ArangoGutierrez why do we need to have the lib here? nvmlResourceManager.nvml is used everywhere

guptaNswati · 2026-02-11T22:06:44Z

sorry for the delayed review. @ArangoGutierrez thank you for your patience.

I looked at it from the perspective of what we did for DRA health checks as the gaol is to align with DRA and integrate with the device-api.

I have minor nits but overall the refractor LGTM.

I still want to spend some more time on reading and trying the mocks tests.

Add vendored NVML mock packages from go-nvml including the dgxa100 mock server. These mocks enable deterministic unit testing of NVML health check behavior without requiring actual GPU hardware. Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Refactor health check implementation: - Extract health monitoring logic into nvmlHealthProvider struct with registerDeviceEvents() and runEventMonitor() methods, making health checks testable in isolation - Migrate CheckHealth interface from stop channel to context.Context for idiomatic cancellation - Extract device placement logic into withDevicePlacements struct to decouple from resource manager - Clarify log message for event registration failure to explain why device is marked unhealthy (aligns with DRA driver convention) - Add unit tests for XID event handling and ERROR_NOT_SUPPORTED regression using NVML mock infrastructure Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com>

Replace the stop channel with context.Context cancellation and sync.WaitGroup for health check goroutine lifecycle management: - Initialize health channel and healthCtx before Serve() to eliminate race where kubelet could call ListAndWatch before fields are set - Use sync.WaitGroup to ensure health goroutine completes before channel close in Stop() - Add nil guards in Stop() to handle partial Start() failure safely - Add structured error handling for health check completion (success, canceled, error cases) - Add debug logging in ListAndWatch for context cancellation and channel closure - Fix updateResponseForMPS receiver from value to pointer Signed-off-by: Carlos Eduardo Arango Gutierrez <eduardoa@nvidia.com> Co-authored-by: Evan Lezar <elezar@nvidia.com>

ArangoGutierrez · 2026-02-12T14:32:32Z

I have addressed all your comments @guptaNswati / @cdesiniotis PTAL

guptaNswati · 2026-02-12T18:40:13Z

Thank you. I dint get a chance to test it. Can you please paste the test logs

ArangoGutierrez · 2026-02-12T19:29:25Z

Thank you. I dint get a chance to test it. Can you please paste the test logs

Unfortunately, not, didn't save the logs, and the cluster I used is down

Copilot

Pull request overview

Copilot reviewed 7 out of 20 changed files in this pull request and generated 4 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-12T19:35:13Z

internal/plugin/server.go

 		return fmt.Errorf("error waiting for MPS daemon: %w", err)
 	}

+	plugin.health = make(chan *rm.Device)


plugin.health is created as an unbuffered channel and the health-check goroutine can send to it even when no ListAndWatch stream is active. Since Stop() now cancels the context and then waits for the health goroutine to exit, a blocked send on plugin.health can deadlock Stop() (context cancellation won’t unblock a channel send). Consider making the channel buffered (e.g., sized to number of devices) and/or ensuring all sends in health-check code are non-blocking/select on ctx.Done().

Suggested change

plugin.health = make(chan *rm.Device)

// Use a buffered channel for health notifications to avoid blocking when

// no ListAndWatch stream is currently consuming from plugin.health.

healthBufSize := len(plugin.rm.Devices())

if healthBufSize < 1 {

healthBufSize = 1

}

plugin.health = make(chan *rm.Device, healthBufSize)

Copilot · 2026-02-12T19:35:14Z

internal/plugin/server.go

 	err := plugin.Serve()
 	if err != nil {
 		klog.Errorf("Could not start device plugin for '%s': %s", plugin.rm.Resource(), err)
-		plugin.cleanup()
 		return err
 	}


On Serve() failure, Start() returns immediately after creating plugin.health and plugin.healthCtx but does not cancel/close them. If the same plugin instance is retried or kept around, this leaks resources and leaves fields in a partially-initialized state. Consider deferring a cleanup (cancel + close channel) on the Serve() error path.

Copilot · 2026-02-12T19:35:14Z

internal/rm/health.go

 		deviceIDToGiMap[d.ID] = gi
 		deviceIDToCiMap[d.ID] = ci
 		parentToDeviceMap[uuid] = d


parentToDeviceMap is keyed by uuid returned from getDevicePlacement(), which is the parent UUID for MIG devices. If multiple MIG devices share the same parent GPU, this assignment overwrites earlier entries, so events for that parent can only ever map to one MIG device. To correctly handle multiple MIG instances per parent, use a mapping that can represent multiple devices per parent (e.g., parent UUID -> list, or parent UUID -> (GI,CI)->device).

Copilot · 2026-02-12T19:35:14Z

internal/rm/health.go

 		if ret != nvml.SUCCESS {
 			klog.Infof("Error waiting for event: %v; Marking all devices as unhealthy", ret)
-			for _, d := range devices {
-				unhealthy <- d
+			for _, d := range h.devices {
+				h.unhealthy <- d
 			}
 			continue


When eventSet.Wait() returns a non-timeout error, the loop marks all devices unhealthy and continues. If the error is persistent, this can spin forever and repeatedly attempt blocking sends to h.unhealthy, preventing shutdown if the receiver isn’t draining fast enough. Consider returning the error (or backing off) and make unhealthy notifications context-aware (select on ctx.Done()) to avoid blocking indefinitely.

guptaNswati · 2026-02-14T00:23:19Z

Thank you. I dint get a chance to test it. Can you please paste the test logs

Unfortunately, not, didn't save the logs, and the cluster I used is down

i know these are straightforward changes and you must have tested them very well. I feel more comfortable when i see the test logs so i will try to set up a cluster next week and see if we can test them.

ArangoGutierrez requested review from Copilot and elezar January 26, 2026 12:02

ArangoGutierrez self-assigned this Jan 26, 2026

Copilot started reviewing on behalf of ArangoGutierrez January 26, 2026 12:02 View session

Copilot AI reviewed Jan 26, 2026

View reviewed changes

internal/plugin/server.go Outdated Show resolved Hide resolved

ArangoGutierrez force-pushed the healthcheck branch 2 times, most recently from ad1263e to c352a5a Compare January 26, 2026 12:23

ArangoGutierrez requested review from guptaNswati, rahulait and rajathagasthya January 27, 2026 19:27

elezar reviewed Jan 28, 2026

View reviewed changes

ArangoGutierrez force-pushed the healthcheck branch from c352a5a to 7a0f2aa Compare January 28, 2026 11:26

ArangoGutierrez force-pushed the healthcheck branch from 7a0f2aa to 14c5914 Compare January 28, 2026 11:27

ArangoGutierrez requested a review from elezar January 28, 2026 12:02

elezar reviewed Jan 29, 2026

View reviewed changes

ArangoGutierrez force-pushed the healthcheck branch from 5a400f5 to f5db035 Compare January 29, 2026 11:40

ArangoGutierrez requested a review from elezar January 29, 2026 11:40

cdesiniotis reviewed Feb 11, 2026

View reviewed changes

guptaNswati reviewed Feb 11, 2026

View reviewed changes

internal/plugin/server.go Show resolved Hide resolved

guptaNswati reviewed Feb 11, 2026

View reviewed changes

ArangoGutierrez force-pushed the healthcheck branch from f5db035 to 4a84232 Compare February 12, 2026 14:22

ArangoGutierrez and others added 3 commits February 12, 2026 15:24

ArangoGutierrez force-pushed the healthcheck branch from 4a84232 to d6c787c Compare February 12, 2026 14:28

ArangoGutierrez requested review from cdesiniotis, Copilot and guptaNswati February 12, 2026 19:29

Copilot started reviewing on behalf of ArangoGutierrez February 12, 2026 19:30 View session

Copilot AI reviewed Feb 12, 2026

View reviewed changes

-		if err != nil && !errors.Is(err, context.Canceled) {
-			klog.Errorf("Failed to start health check: %v; continuing with health checks disabled", err)
-		}
+		switch {
+		case err == nil:
+				fallthrough
+		case errors.Is(err, context.Cancelled):
+				klog.Infof("Stopping health check")
+		case err != nil:
+				klog.Errorf("Failed to start health check: %v; continuing with health checks disabled", err)
+		}
+}

-	plugin.health = make(chan *rm.Device)
+	// Use a buffered channel for health notifications to avoid blocking when
+	// no ListAndWatch stream is currently consuming from plugin.health.
+	healthBufSize := len(plugin.rm.Devices())
+	if healthBufSize < 1 {
+		healthBufSize = 1
+	}
+	plugin.health = make(chan *rm.Device, healthBufSize)

Conversation

ArangoGutierrez commented Jan 26, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

elezar left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ArangoGutierrez Jan 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

guptaNswati commented Feb 11, 2026

Uh oh!

ArangoGutierrez commented Feb 12, 2026

Uh oh!

guptaNswati commented Feb 12, 2026

ArangoGutierrez Jan 28, 2026 •

edited

Loading