ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed

The operating system or container memory limit killed the Raylet, which made the cluster unavailable. This is node memory exhaustion, not an object-store eviction or a recoverable worker-only failure. This entry explains how to confirm the cause, apply the fix, and separate it from adjacent ray failures.

Symptom

ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed

Root cause

The operating system or container memory limit killed the Raylet, which made the cluster unavailable. This is node memory exhaustion, not an object-store eviction or a recoverable worker-only failure. The decisive evidence is the first log line that precedes "ray.

Recommended fix

inspect the node kernel or container events for the Raylet OOM kill and preserve its peak resident memory before restarting the node service.

How Denpex helps

Denpex investigates ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed using the evidence you provide or your connected workload collects. Earlier rank, host or application evidence is needed to distinguish an initiating failure from a downstream report.

The literal signature is "ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed". It is a infrastructure failure associated with Ray jobs, clusters, and runtime environments. The line identifies the failing operation or subsystem, while the surrounding evidence decides whether it is the initiating fault or a downstream symptom.

The operating system or container memory limit killed the Raylet, which made the cluster unavailable. This is node memory exhaustion, not an object-store eviction or a recoverable worker-only failure. The failure becomes visible at this call site because the operation first requires the missing resource, valid state, healthy peer, or correct result. Earlier log lines and a known-good control carry more causal value than the final wrapper exception.

• The workload stops or loses forward progress after emitting "ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed".
• A retry on the same configuration reproduces the failure because the causal state has not changed.
• The outer framework exception can hide the rank, node, allocation, or dependency that failed first.
• Increasing timeouts or reducing workload size can suppress the symptom without correcting the cause.

• Ray jobs, clusters, and runtime environments
• production-shaped multi-accelerator workloads
• containerized and bare-metal deployments of the same stack

Use this checklist to test the hypothesis against a small reproduction. No single line proves the root cause, so preserve the preceding events and compare one variable at a time.

• ✓Find the first occurrence of "ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed" and preserve at least 100 lines before it.
• ✓Identify which rank, node, device, or process emitted the first related warning.
• ✓separate Raylet heap, object store, and worker memory. Check cgroup memory.current and memory.events against the configured node limit.
• ✓Repeat the same input after the targeted change and require the signature to disappear.
• ✓Resume from job retry after the Raylet is healthy and node memory is bounded only after the control passes.

• The operating system or container memory limit killed the Raylet, which made the cluster unavailable. This is node memory exhaustion, not an object-store eviction or a recoverable worker-only failure.
• The decisive evidence is the first log line that precedes "ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed" and differs from a healthy run.
• A nearby failure remains a competing hypothesis until a control separates configuration, capacity, transport, and hardware causes.

ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed

Use this text as a lookup key in logs and upstream issue trackers. It is not presented as a captured customer log. Confirm the cause from your own preceding events, versions, configuration and the cited references.

inspect the node kernel or container events for the Raylet OOM kill and preserve its peak resident memory before restarting the node service. This changes the condition in the causal diagnosis instead of hiding the outer exception. The repeated control proves ownership before recovery from job retry after the Raylet is healthy and node memory is bounded.

# Preserve evidence before restarting
rg -n -i 'error|exception|timeout|failed' <log-file>
nvidia-smi
python -m torch.utils.collect_env

# Find the exact signature in the complete log
rg -n -F -- "ray.exceptions.ClusterUnavailableError Raylet process crashed OOM killed" <log-file>

Adapt the snippet to your framework. The same pattern holds for PyTorch Lightning, Hugging Face Trainer, DeepSpeed, Megatron-LM, and vLLM training wrappers. Where the wrapper exposes a config flag (for examplelr_scheduler_type in Trainer), prefer the flag over the imperative API to keep the schedule declarative and reproducible.

Select the traceback or open the failed terminal, then run Denpex locally to see the initiating rank, collateral failures, exact fix, and verification command without uploading the log.