Uploaded July 2026 | Updated September 2026, 3 weeks ago
Modern frontend development ecosystems, particularly Node.js, are notorious for generating enormous metadata footprints. A single 'node_modules' directory can easily contain over 100,000 files and deeply nested directory trees. Now multiply that by dozens or hundreds of concurrent developer workspaces, all backed by CephFS, and you have a perfect storm hitting your MDS layer.
In this talk, we share our real-world experience delivering on-demand developer workspaces to engineering teams using the DevWorkspace Operator on Kubernetes, with Rook-Ceph providing persistent storage via CephFS. Our primary challenge was workspace startup time: when a developer launches their environment, dependency resolution and filesystem population generate a massive burst of metadata operations: 'open', 'stat', 'readdir', 'create', that can overwhelm a single MDS daemon and turn a quick "start my workspace" into minutes of waiting.
We will walk through the journey from identifying the MDS as the bottleneck to implementing a multi-layered tuning strategy on Ceph Squid (v19) that dramatically improved the developer experience
Slides: ceph.io/assets/pdfs/events/2026/ceph-days-london/scaling-cephfs-mds.pdf
Modern frontend development ecosystems, particularly Node.js, are notorious for generating enormous metadata footprints. A single 'node_modules' directory can easily contain over 100,000 files and deeply nested directory trees. Now multiply that by dozens or hundreds of concurrent developer workspaces, all backed by CephFS, and you have a perfect storm hitting your MDS layer.
In this talk, we share our real-world experience delivering on-demand developer workspaces to engineering teams using the DevWorkspace Operator on Kubernetes, with Rook-Ceph providing persistent storage via CephFS. Our primary challenge was workspace startup time: when a developer launches their environment, dependency resolution and filesystem population generate a massive burst of metadata operations: 'open', 'stat', 'readdir', 'create', that can overwhelm a single MDS daemon and turn a quick "start my workspace" into minutes of waiting.
We will walk through the journey from identifying the MDS as the bottleneck to implementing a multi-layered tuning strategy on Ceph Squid (v19) that dramatically improved the developer experience
Slides: ceph.io/assets/pdfs/events/2026/ceph-days-london/scaling-cephfs-mds.pdf
