Free tools Windows power users keep installed
One-click scans. No signup required.
A task stopping while Amazon ECS repairs a GPU host can be expected, but a replacement task that remains pending, repeatedly fails, or exits immediately needs separate diagnosis. Start with three records: the stopped task’s stoppedReason and stopCode, recent service events, and the affected container instance’s health details. Then determine whether the task was stopped during draining, could not be placed on replacement capacity, or started and failed at runtime. That sequence separates normal repair behavior from a bad replacement, an unavailable GPU, an ECS agent problem, an incompatible image, or a service deployment constraint.
Why a GPU repair can stop an otherwise healthy ECS task
Amazon ECS Managed Instances with GPU hardware use NVIDIA Data Center GPU Manager (DCGM) to monitor GPU health. When a critical GPU check fails and GPU auto-repair is enabled, ECS can mark the container instance impaired, set it to DRAINING, provision replacement capacity, allow existing tasks to stop within their configured stop timeout, and terminate the impaired host. This is a start-before-stop process rather than an indication that every task on the host is defective.
As an Amazon Associate I earn from qualifying purchases.
During draining, ECS does not place new tasks on the impaired instance. Service tasks are stopped and the service scheduler attempts to launch replacements if eligible capacity and deployment settings permit it. GPU auto-repair is rate-limited to no more than 20% of a capacity provider’s instances at one time. If the capacity provider has fewer than nine instances, ECS drains at most one instance at a time. Consequently, a service task stopping at the same time as a documented drain event is normally an expected lifecycle event; it does not by itself prove that the replacement GPU has failed.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →This behavior applies to ECS Managed Instances and their documented GPU auto-repair workflow. An ECS cluster using self-managed EC2 instances does not automatically acquire every managed-instance repair behavior. For self-managed capacity, inspect the EC2, ECS agent, operating-system, and instance-management events that actually apply to your cluster.
#1 Best Overall
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
Capture the three evidence sets before changing anything
Preserve the task ARN, cluster, service, container-instance ARN, EC2 instance ID, AWS Region, task-definition revision, AMI version, ECS agent version, event timestamps, and health status in one incident timeline. Avoid deleting stopped tasks or replacing a host before collecting these fields; the sequence of events often identifies the failure class.
1. Read the stopped task
Use DescribeTasks for the exact task ARN or ID. Keep the task-level stoppedReason, stopCode, each container’s exit code and reason, the startedAt and stoppedAt timestamps, and the container-instance ARN.
aws ecs describe-tasks --cluster CLUSTER_NAME --tasks TASK_ARN_OR_ID --query 'tasks[0].{taskArn:taskArn,lastStatus:lastStatus,desiredStatus:desiredStatus,stoppedReason:stoppedReason,stopCode:stopCode,startedAt:startedAt,stoppedAt:stoppedAt,containerInstanceArn:containerInstanceArn,containers:containers[*].{name:name,lastStatus:lastStatus,exitCode:exitCode,reason:reason}}' --output jsonA task stopped during a host drain may have a reason associated with termination or service replacement, whereas a task that never reached RUNNING is more likely to expose a placement or image-start problem. Do not infer the cause from an exit code alone: correlate it with the instance and service timestamps.
Recommended Free Tools
2. Read recent service events
For a service task, inspect the newest events first. Events can show that a task was stopped because an instance entered draining, that no container instance met the requirements, or that task launches repeatedly failed.
aws ecs describe-services --cluster CLUSTER_NAME --services SERVICE_NAME --query 'services[0].events[0:20]' --output jsonA repeated-start warning is a pointer to inspect the individual failed tasks, not a root cause. Record the event message and its timestamp, then match it to the task ARN and replacement instance. Standalone tasks launched with RunTask do not have a service event stream that will replace them automatically.
3. Read container-instance health and status
Resolve the container-instance ARN from the task, then request health details and status.
aws ecs describe-container-instances --cluster CLUSTER_NAME --container-instances CONTAINER_INSTANCE_ARN --include TAGS --output jsonInspect status, agentConnected, runningTasksCount, pendingTasksCount, the overall health status, individual health checks, and statusReason. Also record whether the instance was DRAINING, whether it later disappeared, and whether a replacement registered as ACTIVE.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchRank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
Classify the failure by where it occurred
The lifecycle stage determines which evidence matters most. Use the following classification before applying a fix.
| Lifecycle stage | Typical evidence | Most useful next checks |
|---|---|---|
| Stopped while the original host drained | Instance changed to DRAINING; service event mentions replacement or termination; task has a stop timestamp during the repair window |
Confirm replacement capacity, deployment limits, and the new task’s placement |
| Never placed on a replacement | Task remains PENDING or service reports no suitable container instance |
Check RESOURCE:GPU, CPU, memory, attributes, constraints, networking, and agent connectivity |
| Placed but never starts | Task stops shortly after placement; container reason refers to image, runtime, mount, or resource initialization | Inspect container reason, image architecture, NVIDIA runtime configuration, and task-definition revision |
| Started and then exits | Task reached RUNNING; container has an exit code or application log; GPU restart metrics may be present |
Review application logs, exit details, XID records, health checks, and workload behavior |
| Replacement is registered but cannot receive the task | Replacement is ACTIVE yet placement events continue |
Compare requested GPUs and required attributes with the replacement’s registered resources and instance family |
Interpret ECS container-instance health correctly
For ECS Managed Instances, the overall health status reflects the most severe individual check. Documented values include OK, IMPAIRED, INSUFFICIENT_DATA, and INITIALIZING. The ACCELERATED_COMPUTE check covers GPU health. When a managed GPU is impaired, its statusReason contains an NVIDIA XID value in the form XID_<number>.
INITIALIZING and INSUFFICIENT_DATA are not equivalent to a confirmed GPU failure. They mean that health information is not yet complete or sufficient. Allow the replacement to register and populate health data, then compare its status with task-placement events before attributing a task failure to hardware.
Automated container-instance health monitoring requires ECS agent version 1.57.0 or later. The AWS CLI workflow for inspecting these health details requires AWS CLI version 1.22.3 or later, or version 2.3.6 or later. If your agent or CLI is older, a missing or incomplete health response may reflect tooling capability rather than a GPU condition.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteDistinguish a drained host from an impaired replacement
- Original host is draining, replacement is healthy: the stopped task is consistent with the repair sequence. Focus on service replacement and capacity.
- Replacement is
IMPAIREDwith an accelerated-compute reason: treat it as a new host-health incident and preserve its XID and timestamps. - Replacement is
INITIALIZINGorINSUFFICIENT_DATA: wait for health data while checking whether the task can be placed; do not label the GPU defective solely from that state. - Agent is disconnected: placement can fail even when the instance has physical GPUs. Investigate agent connectivity and instance status before changing the task definition.
Resolve placement failures before debugging the application
If a task never reaches RUNNING, inspect the service event and ECS placement failure message. A resource or attribute mismatch is more likely than an application crash at this stage.
RESOURCE:GPU
RESOURCE:GPU means the number of GPUs requested by the task is unavailable on eligible container instances. Compare the task definition’s GPU count with the replacement’s registered GPU resources, not just the number of hosts in the cluster. A host can be present but unusable because all of its GPUs are allocated, the required GPU count is larger than any free device set, or the host is excluded by another requirement. The remedy may be additional GPU-enabled capacity or a task definition that requests the amount the workload actually needs.
AGENT and disconnected instances
An AGENT placement failure indicates that the target container instance has a disconnected ECS agent. Verify agentConnected, instance status, network reachability to ECS endpoints, and the agent version. Do not treat an agent error as proof of a failed GPU. A healthy GPU cannot accept a task through an unavailable control-plane agent.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
Attributes, constraints, and non-GPU resources
After GPU capacity, check CPU, memory, port mappings, operating-system or architecture attributes, availability-zone constraints, task-group constraints, and any custom attributes required by the service. Placement succeeds only when one instance satisfies all requirements simultaneously. A replacement may have the right GPU model but fail because it lacks memory, a required attribute, a compatible architecture, or network capacity.
Nominal instance count is therefore an unreliable capacity measure. Evaluate usable capacity: free GPUs, free CPU and memory, required attributes, placement constraints, networking, and the health of the instance and its agent.
Investigate tasks that start and then exit
Once a task has reached RUNNING, inspect the container’s exit code, reason, application logs, and the exact time relative to the repair. Also compare NVIDIA XID records and ECS GPU restart metrics.
ECS Managed Instance GPU workloads expose TaskGPURestartAppXidCount and ContainerGPURestartAppXidCount. These count XID errors in NVIDIA’s RESTART_APP resolution category. AWS describes these metrics as a way to detect transient GPU faults affecting a task or container without implying underlying hardware failure. A nonzero application-restart count can therefore indicate a recoverable application-level fault; it is not, on its own, evidence that the host GPU must be replaced.
Evidence that points toward the application
- The replacement instance is healthy and remains
ACTIVE. - The task reaches
RUNNINGbefore the container exits. - Application logs show an out-of-memory condition, an internal assertion, a failed model load, or a process-level error.
- Only one task or container records a restart-app XID while other workloads on the host remain healthy.
Evidence that points toward the host or driver
- The managed instance reports an
ACCELERATED_COMPUTEimpairment with an XID status reason. - Multiple tasks on the same host fail around the same time with GPU initialization or device-access errors.
- The replacement also becomes impaired after registration.
- The task cannot access a GPU device even though the task definition requests one and the host reports free GPU resources.
Use the combined evidence. A single XID, a single application exit, or a task stop during draining is insufficient to establish a hardware failure.
Verify the GPU task definition and ECS agent configuration
Confirm ECS GPU support is enabled
Check that the ECS agent configuration includes ECS_ENABLE_GPU_SUPPORT=true. Without GPU support enabled in the agent, a task definition that requests GPUs may not be scheduled or initialized as intended. Confirm the setting on the replacement host, not only on the repaired instance.
Confirm the task’s GPU request
Inspect the task-definition revision actually deployed by the service. Verify each GPU-requiring container’s resource requirement and the total number of GPUs requested by the task. A revision change during the repair window can make a previously placeable task unschedulable. Compare the requested count with the replacement’s registered resources and with other tasks already using its GPUs.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
Understand device assignment
For a container with a GPU resource requirement, ECS assigns the NVIDIA container runtime and sets NVIDIA_VISIBLE_DEVICES to the GPU device IDs allocated to that container. Inside a running replacement task, verify that the variable is present and that the application can enumerate the assigned devices. An empty or unexpected device list points to runtime, image, agent, or task-definition configuration rather than a service deployment percentage.
Check AMI, instance-family, and driver compatibility
Verify that the replacement uses a GPU-optimized AMI and an instance family supported by the ECS GPU documentation. The older g2 family is deprecated in that guidance. AWS also warns that in-place NVIDIA or CUDA driver updates on specified older families can cause GPU workload failures. Compare the replacement AMI and driver versions with the versions used by the task’s image and with the versions supported for that instance family.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
For a custom image that does not use an NVIDIA or CUDA base image, review its NVIDIA_DRIVER_CAPABILITIES requirements and ensure the image includes the libraries and capabilities the application needs. A container can be scheduled successfully yet fail during initialization if its user-space CUDA stack is incompatible with the host driver.
Understand draining, deployment percentages, and task type
What happens to a service during drain
When an instance enters DRAINING, ECS places no new tasks there. Service tasks are stopped according to the drain and task-stop behavior, and the service scheduler attempts replacements on eligible instances. Replacement success depends on available GPU, CPU, memory, attributes, networking, and health checks, plus the service deployment configuration.
Review minimumHealthyPercent and maximumPercent for the service. The minimum controls how far below the desired count ECS may go while replacing tasks. The maximum caps how many tasks can run during replacement. If both are set to 100%, ECS cannot stop old tasks or start additional replacements, so a deployment or drain can stall when no spare capacity exists. Change these values only when the observed service behavior and your availability requirements justify it; do not alter them merely because a task was stopped during an expected repair.
Standalone tasks are different
A standalone task launched with RunTask is not managed by the service scheduler. A pending or running standalone task is not automatically replaced simply because its container instance drains. It may finish, remain pending, or need to be stopped and launched again by your own automation. Do not apply a service replacement expectation to a standalone batch, inference, or maintenance task.
Repair concurrency and temporary capacity
The 20% repair limit, and the one-instance limit for capacity providers with fewer than nine instances, reduce simultaneous disruption but do not guarantee immediate placement of every replacement task. A cluster with little spare GPU capacity can still leave tasks pending after one host drains. Check whether the capacity provider has enough eligible free GPUs to absorb the replacement before assuming the repair process failed.
Validate the replacement host before returning traffic
- Confirm the replacement container instance is registered in the correct cluster and has
ACTIVEstatus. - Confirm the ECS agent is connected and its version supports the health checks and GPU features you rely on.
- Confirm the overall health is
OK, or wait forINITIALIZINGorINSUFFICIENT_DATAto resolve while monitoring events. - Confirm the accelerated-compute health check is not impaired and that no critical XID status reason is reported.
- Compare registered GPU resources and instance attributes with the task definition’s GPU count and placement constraints.
- Check free CPU, memory, ports, and any custom attributes required by the service.
- Launch or allow the service to launch the replacement task, then verify that it reaches
RUNNING. - Inspect container logs, exit details, and GPU device visibility inside the running task.
- If the service uses a load balancer, wait for the target to pass its configured health checks before treating the replacement as ready.
- Keep the original and replacement task timelines together so a later recurrence can be compared against the repair event.
A practical decision tree for common symptoms
The task stopped exactly when the host entered DRAINING
Classify this as expected repair behavior first. Verify that a service replacement was created and that an eligible, healthy host exists. If the task is standalone, decide whether your automation should launch a new task. Escalate to host diagnosis only if the replacement is impaired or the task fails independently after it starts.
The replacement task is stuck in PENDING with RESOURCE:GPU
Compare requested and available GPUs on eligible instances. Remove an incorrect resource request only if the workload does not need it; otherwise add or restore GPU-enabled capacity. Then recheck CPU, memory, attributes, and constraints, because fixing the GPU count may expose a second placement limit.
The replacement task reports AGENT
Investigate the replacement’s agent connection, instance status, network path, and agent version. A disconnected agent cannot receive a task even when the GPU is healthy. Resolve the control-plane condition and re-evaluate placement.
The task reaches RUNNING and exits on the repaired host
Read the container exit details and application logs, verify GPU device visibility, and compare XID and restart-app metrics. If the host is healthy and only the application fails, debug the image, model, memory use, and process. If the managed host reports accelerated-compute impairment or several tasks fail together, preserve the hardware and driver evidence for host-level remediation.
The replacement is healthy but the service never reaches the desired count
Inspect deployment percentages, available spare capacity, placement constraints, load-balancer health checks, and task-start events. A 100% minimum and 100% maximum can prevent replacement when no extra capacity exists. A task that starts but fails its load-balancer check is not equivalent to a GPU placement failure.
Build a repair-aware incident runbook
Keep a standard capture template containing the Region, cluster, capacity provider, service or standalone-task identifier, task ARN, task-definition revision, original and replacement container-instance ARNs, EC2 instance IDs, AMI and ECS agent versions, requested GPU count, registered GPU resources, health-check results, XID values, service events, task stop reasons, container exit details, and load-balancer health. Include timestamps in a single time zone so drain, replacement registration, placement, and exit can be ordered accurately.
Use alerts for service events indicating repeated task starts, pending tasks that exceed your normal startup window, loss of agent connectivity, and an impaired accelerated-compute check. Pair those alerts with the task and instance identifiers needed for immediate DescribeTasks and DescribeContainerInstances calls.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Maintain enough eligible GPU capacity for the service’s deployment policy. The exact amount depends on desired count, GPU-per-task, placement constraints, and the maximum number of hosts that may drain concurrently; the repair percentage is not a promise of spare capacity. Test the replacement path with the same task definition, image, AMI family, and networking rules used in production.
Finally, document whether each workload is a service or a standalone task. That one distinction determines whether ECS should replace a stopped task automatically and prevents operators from waiting for a scheduler action that will never occur.
What not to conclude from a single symptom
- A task stopped during draining does not prove the replacement GPU is bad.
- An XID value does not, without its health context, prove that hardware failed.
- A nonzero restart-app GPU metric does not necessarily indicate an underlying hardware fault.
- A healthy instance count does not prove that usable GPU capacity is available.
INITIALIZINGorINSUFFICIENT_DATAdoes not equalIMPAIRED.- A service placement warning is not a root cause until matched to the failed task and its instance evidence.
- A standalone task does not inherit service replacement behavior.
The reliable diagnosis is the one supported by the task stop record, service event, instance health state, placement resources, and runtime evidence taken together.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
