Evaluating Checkpoint/Restore-Based Pod Migration in Kubernetes: An Empirical Performance and Bottleneck Study

Stateful singleton containers lacking distributed replication lose volatile heap state during Kubernetes node maintenance and cloud Spot instance preemptions. We evaluate an automated, snapshot-consistent pod migration pipeline coordinating CRIU (v4.3), the Kubelet Checkpoint API (KEP-2008), Buildah OCI packaging, CRI-O (v1.31), and Cilium eBPF routing across AWS EC2 nodes. Across 36 migrations of a stateful service compared against 20 cold reschedules, decoupled migration achieves a median cutover blackout of 1064.5 ms versus 1019.0 ms for cold restart (+4.5% overhead) while preserving heap state. A Mann-Whitney U test (p = 0.0660) detects no statistically significant cutover difference, with variance governed by Kubelet's 1 s PLEG cycle. Decoupled migration incurs bounded state drift of 669.0 ms, whereas stop-and-copy yields zero drift at the cost of extending blackout to 1.88 s. Scaling heaps to 1024 MB (1 GB) confirms 100% bit-exact SHA-256 integrity within 55.09 s worst-case source evacuation (45.9% of the 120 s Spot deadline), where uncompressed Buildah packaging slashes build time from 32.06 s to 13.74 s (reducing evacuation to 36.78 s). While 1 GB migration cutover is comparable to estimated cold state reconstruction from cloud storage (11-30 s), checkpointing eliminates application-level persistence tiers, with cutover becoming advantageous over cold rebuild from ~256 MB (2.1 s vs. 3.5-7.5 s). Finally, single-container KEP-2008 checkpointing servers in-pod loopback sockets, suggesting node-level routing as a less fragile alternative.