NVIDIA / gpu-operator / 39473252966

45%

main: 37%

push

github

Log one line per reconcile and record GPU node count on both paths

The operator exposes reconcile duration as a histogram, which carries
no timestamps. Scale benchmarks had to turn on controller-runtime
V(5) logs to find when a reconcile started, and those logs are too
noisy at thousands of nodes. The gpu_operator_gpu_nodes_total gauge
was also set only on the ClusterPolicy path.

Wrap every controller's reconciler so it logs one "Reconcile finished"
line at the default info level with the start time (RFC 3339), the
duration in seconds, and the outcome (success, requeue, or error). The
existing "Reconciling ..." start line stays, so an unmatched start
shows a reconcile still in progress. The GPUCluster controller now
counts GPU nodes and sets gpu_operator_gpu_nodes_total the same way the
ClusterPolicy path does.

Signed-off-by: Rajath Agasthya <ragasthya@nvidia.com>

28 of 41 new or added lines in 7 files covered. (68.29%)

5892 of 13062 relevant lines covered (45.11%)

0.45 hits per line