• Software engineer with a strong focus on Cloud Native, Kubernetes, Metrics, Spark, and AI Infrastructure.
• Experienced in building scalable, reliable systems and passionate about open source and community collaboration.
• Active contributor to the Kubernetes/open source ecosystem, especially in cloud-native observability and AI platforms.

• Languages: Go, Python
• Expertise: Cloud-native infrastructure, Containers & orchestration, Spark, AI Infra, High-performance computing & scheduling

Some of my representative merged PRs:

• feat(scheduler): make node lock timeout configurable (Project-HAMi/HAMi#1117)
  Added a config option for the scheduler's node lock timeout, making GPU scheduling more flexible across clusters of different scale.

• feat: introduce feature gate mechanism (kubeflow/spark-operator#2794)
  Designed and implemented a Kubernetes-style feature gate framework for the Spark Operator (+169/-2 across 4 files), so new features can ship behind flags with backward compatibility and controlled rollout risk.

• feat: skip reconcile for webhook-patched executor fields (kubeflow/spark-operator#2786)
  Skips reconciliation when only webhook-patched executor fields change (PriorityClassName, NodeSelector, Tolerations, Affinity, SchedulerName), eliminating unnecessary application restarts when users update executor scheduling configuration (+448/-4 across 8 files).

• feat: support driver and executor pod use different priority (kubeflow/spark-operator#2146)
  Enabled SparkOperator to set different priorities for driver and executor pods, improving resource allocation and scheduling fairness.

• feat: add option to control deserialization when watching events (kubernetes-client/python#2406)
  Added an option to disable automatic deserialization in Watch.stream(), cutting CPU and memory overhead when watching high-volume resources in large clusters. Merged into the official Kubernetes Python client.

• feat: Add LimitMEMLOCK=infinity to containerd systemd service (kubesphere/kubekey#2609)
  Enabled unlimited memory lock for containerd to better support GPU, eBPF, and HPC workloads.

• Parallel node addition, NVIDIA Runtime & registry mirror support (kubesphere/kubekey#2575)
  Improved cluster management by supporting parallel node addition, NVIDIA GPU runtime, and enhanced registry mirror config.

• feat: optimize variable use and remove duplicate code (kubesphere/kubeeye#181)
  Refactored the cluster inspection path, removing duplicated logic and simplifying variable usage.

• refactor: remove unused code in ReadyNodes (kubernetes-sigs/descheduler#471)
  Refactored and cleaned up unused parameters in the descheduler.

• perf: optimize RB to Work throughput for large-scale Pod distribution (karmada-io/karmada#7063) — open
  Multi-part performance rework of the ResourceBinding → Work synchronization path for 10,000+ Pod multi-cluster distribution (+2859/-139 across 19 files). Diagnosed a ~6,000-Work backlog under a 12,000-Pod dual-cluster stress test and introduced an async Work creator plus related throughput optimizations in the binding controller.

Always open to collaboration and new opportunities!