Jihwan Park / engineering notebook
Engineering notes on constrained infrastructure, service recovery, and resource-aware AI systems.
This site collects sanitized engineering notes about restricted-network platform migration, stateful service recovery, private artifact workflows, operational AI serving constraints, and resource-constrained AI projects.
Work Map
Concrete Threads
The recurring pattern is not a broad interest in infrastructure; it is a set of specific failure modes, constraints, and design decisions that kept showing up across service operation and AI workload projects.
Restricted-Network Platform Migration
Selected services moved from ad-hoc Docker operation toward a lightweight Kubernetes/K3S-based platform while dealing with image distribution, service exposure, persistent storage, and deployment standardization.
Stateful Service Recovery
PVC-backed services required thinking beyond pod rescheduling: storage detach/attach delay, rebuild behavior, service readiness, downtime, and RTO/RPO boundaries.
Private Artifact and Bootstrap Supply Chain
Restricted networks made artifact distribution part of reliability work: image mirroring, Maven/Nexus behavior, bootstrap dependency paths, and registry self-dependency risks.
Full-GPU Serving Availability
Full-GPU workloads made fine-grained sharing less straightforward, so degraded serving, active-active replicas, and traffic rerouting became more realistic availability questions.
TinyML Memory Planning
TensorFlow Lite Micro work focused on tensor lifetimes, conflict relationships, BFS-style allocation ordering, and in-place update opportunities to reduce peak memory.
Hardware-Aware CNN Acceleration
The Zynq FPGA accelerator project used bit-shift operations, INQ/STE quantization, IM2COL, systolic array dataflow, and AXI4-Lite PS/PL integration to study area/runtime tradeoffs.
Index
Technical Index
Concept study notes are separated from operational records so DevOps stays closer to troubleshooting, runbooks, deployment decisions, and recovery paths.
Notes
Concept notes for systems, cloud infrastructure, networks, databases, operating systems, algorithms, and implementation details.
60
DevOps
Operational records: troubleshooting notes, runbooks, deployment decisions, recovery paths, and infrastructure constraints.
5
Projects
Project-level records around restricted-network platform migration, stateful recovery, artifact workflows, and resource-constrained AI systems.
0
Research
Research-oriented notes around reliable systems, storage recovery, ML systems, AI serving infrastructure, and hardware-aware execution.
0
Paper Reviews
Technical paper reading notes focused on core ideas, implementation details, limitations, and references.
0
Recent
Recent Entries
Latest published notes from the public technical sections.
Nexus3 HTTPS Registry: Root CA 기반 IP SAN 인증서와 Docker 신뢰 설정
DNS 없이 IP와 포트 기반으로 Nexus3 Docker Registry와 Traefik TLSStore를 운영할 때 Root CA, 서버 인증서, Java keystore, Docker ca.crt를 구분해 설정하는 runbook.
K3s Traefik: NodePort와 MetalLB VIP 병행 노출
K3s Traefik에서 기존 NodePort 노출을 유지하면서 MetalLB VIP를 추가하고, tls: {} 기본 인증서 문제를 TLSStore로 정리한 운영 기록
K3s MetalLB: ServiceLB/Klipper와 HostPort 충돌 분석
K3s에서 MetalLB를 도입할 때 ServiceLB/Klipper가 LoadBalancer Service의 hostPort를 점유하면서 기존 HostPort/NodePort 노출과 충돌한 문제를 정리한다.
K3s kubeconfig: kubectl과 k9s 다중 클러스터 접근 설정
K3s server node의 kubeconfig를 로컬에 등록해 kubectl과 k9s에서 여러 Kubernetes cluster context를 전환하는 정석적인 설정 절차를 정리한다.