A simulated GPU cluster is writing logs right now: a training job, two vLLM inference servers, the GPU Operator, the kubelet and the Linux kernel. Run a LogQL query, or trigger a failure on the right and find the line that explains it. Companion to GPU Cluster Watch, which shows the same failures as metrics.
Labels in this demo: node, namespace, pod, container, source (kernel, container or kubelet). Kernel lines have no namespace or pod: they belong to the node. Demo clock runs in real time.
Newest first, up to 200 lines. Click a line to see its labels and what to check next.
Click any log line to inspect it.