[Bug] pods_allocation.csv coming out empty

[Mapalmeira]

When running emulations that use tracer_cluster_autoscaler.py / tracer_karpenter.py / tracer_kwok.py, the method process_and_save_pods_allocation() can produce an empty pods_allocation.csv, leading to pods_and_nodes_map.csv also becoming empty, which causes all initial pods to not enter the emulation.

This happens in the following section of code

        df_pods_allocation = self.df_final[self.df_final['timestamp'] == self.df_final['timestamp'].min()]

        df_pods_allocation = df_pods_allocation[~df_pods_allocation["node"].isin(["unallocated"]) & ~df_pods_allocation["instance_type"].isin(["unallocated"])]

        # Remove nodes that occur in just one timestamp
        node_counts = df_pods_allocation.groupby('node')['timestamp'].nunique()
        nodes_with_single_timestamp = node_counts[node_counts == 1].index.tolist()
        df_pods_allocation = df_pods_allocation[~df_pods_allocation['node'].isin(nodes_with_single_timestamp)]

First, df_final is being filtered down to a single snapshot (the first timestamp). Later, all nodes which appear in only one timestamp are deleted. As every node in df_pods_allocation is within one timestamp (because only the first timestamp was selected), every node gets deleted and df_pods_allocation becomes empty.

Question: why is it necessary to remove the nodes that appear in just one timestamp?