Redeploying bazel-remote results in all gcs cached artifacts being invalidated

[GonzaloSaez]

I'm using k8s to deploy bazel-remote backed up by local SSDs and a GCS bucket. When redeploying the bazel-remote pods/deployment/statefulset (i.e. when losing the local disk cache), it seems that all the GCS cache entries are invalidated. Why is this? I haven't enabled enable_ac_key_instance_mangling so this sounds strange to me. Is there a way with --incompatible_strict_action_env to avoid this cache invalidation? Thank you