Yahoo! Japan’s owner consolidating 164 OpenStack clusters into one

LY Corporation is ripping up a heavily customised OpenStack estate and stitching a much leaner cloud back together. It has been reported that the group — formed when Yahoo! Japan merged with LINE in 2023 — is replacing its bespoke stacks with a unified platform called “Flava,” cutting hundreds of clusters down to a single upstream-aligned OpenStack cluster.

The scale (and the mess)

It has been reported that LINE’s internal cloud, Verda, ran about 130,000 VMs across 11,000 hosts and four OpenStack clusters, while Yahoo! Japan’s YNW cloud spanned roughly 27,000 servers and more than 160 OpenStack clusters. Why so drastic a consolidation? According to Ryuutarou Inoue, head of LY’s Cloud Infrastructure Unit, too many custom patches made upgrades painful. “Flava adopts an architecture that stays aligned with upstream OpenStack,” he said, adding that LY will minimise local patches and contribute changes back upstream.

Design choices and operational focus

Flava is being built for resilience by design: pursue statelessness, shift availability responsibility toward application architecture, and prioritise fast recovery over restoring the exact prior state. It has been reported that the target footprint is roughly 500 hosts and 9,000+ VMs in a single cluster, with core plumbing using Envoy, eBPF/XDP, FRRouting and Ceph, and observability via Prometheus and Grafana. The message is clear — embrace upstream, automate failures, and stop paying for bespoke complexity.

Why this matters

This is a classic tech‑debt reckoning. Custom forks can feel liberating — until they become a trap that blocks upgrades and security fixes. LY’s move echoes a broader industry trend: large operators returning to upstream to reduce maintenance overhead and accelerate feature delivery. The gamble? That a leaner, upstream-first Flava will keep 300 million monthly users humming without being held hostage by bespoke code.

Sources: The Register