This was reported in a two-node cluster with stonith-enabled=true but no stonith devices configured, but none of that may be relevant.
The reproducer is to configure a cluster, add a dummy resource, then delete the dummy resource. The resulting scheduler input can be run with CIB_file=$INPUT crm_resource --wait -T 1 to see pending actions scheduled. This likely has no effect on the actual cluster, but confuses --wait.
See: