<!-- markdownlint-disable MD013 -->
# Data Platform Systems Engineering — Operating History
AI Summary
Purpose:
- Keep the verified systems-engineering narrative used for the Toss Payments application, with explicit role boundaries and before/after metrics.
Key points:
- Domain identity comes first: Hyunwook collects software-supply-chain security data and validates whether vulnerability records actually describe the affected product/version. The data includes package/version metadata, repositories/source/patches, binaries, licenses, CVE/OSV and Linux-distribution advisories.
- The systems role grew from that domain work: collection gaps, resource contention, DB failures, storage and customer-delivery delays directly reduce security-data freshness and trustworthiness. Infrastructure ownership is therefore a continuation of the data/vulnerability role, not an unrelated second career.
- Common platform scope: more than 80 crawlers; Hyunwook owns shared Airflow scheduling, Kubernetes resources/networking, DB access boundaries, storage, backup and monitoring. Individual crawler logic remains with each crawler owner.
- Kubernetes capacity/topology: pre-existing IDC cluster 2 nodes → added 4 in-house workers through WireGuard/NAT → 6 nodes total. Pod allocation failures fell from roughly 70–100/week to fewer than 1–5/week.
- DB host stabilization and migration: a general-purpose M.2 host froze under DB heat/load. Syslog, network conditions and schedule concentration were checked; schedule spreading reduced incidents from 10–20/month to fewer than five. After the 2024 IDC migration the confirmed DB operational incident count is 1–2 (user-corrected in
ai/worklog/2026/2026-W29.md; an earlier version of this line said "zero"). - DB access boundary: more than 80 parallel crawlers moved from direct DB access to a DML Broker with a bounded worker queue and connection pool. User-confirmed lock/connection errors fell from 3–4/day to zero.
- Backup automation: monthly operator-driven backup/wait/NAS-copy work became scheduled generation, split/checksum, NAS storage and messenger notification. Internal backup history includes a 767GB MySQL archive split into 200GB pieces.
- Observability: Prometheus/Grafana cover approximately 10–20 team servers, DB read/write and replica state, collection gaps/volume, and customer data delivery; PostgreSQL time-series data supports data-flow views.
- Access automation: the team Linux account process covered 12 IDC/in-house servers and used a user-management script, personal IDs/SSH keys and shared directory-permission rules.
- Storage architecture: open-source metadata/binaries and Git repositories are separated by mutation pattern across SeaweedFS/XFS and ZFS/Forgejo.
Role boundary:
- Do not say Hyunwook designed the original Kubernetes cluster.
- Do not say he fixed every crawler's business logic.
- Safe wording: project-lead/technical ownership for common infrastructure, capacity planning, topology changes, incident response, automation and monitoring.
Sources:
- [[../../sources/career/2026-07-16-resume-introduction-feedback.md]]
- [[../../sources/career/2026-07-14-toss-systems-engineer-user-history.md]]
- Confluence pages 4026335502, 4079321125, 4142465026, 4160618647 and 4160389371.
- Public portfolio pages under
human/portfolio/items/.