Cloud Development Environment Platform  ·  View 10 of 21  ·  Data

Data Flow — From Keystroke to Snapshot

What leaves the guest, what is deliberately dropped, and what never touches the control plane.

Editable source SVG draw.io All views
Sources Developer edits in the guest Agent writes Repository clone Inside the guest Workspace filesystem 32 GB disk Process memory 8 GB nominal Ephemeral paths excluded by default Capture Copy-on-write delta block level Dedup vs prebuild median 2 GB Durable Snapshot store 11 nines · tenant key Memory image decays to disk-only Evidence Lifecycle events RPO 1 min Session records 400 d Audit records 5 y immutable Consumers Billing & showback Operations latency by cause Security review dropped lazy restore Data Flow — From Keystroke to Snapshot Person or role External / third party Data store Risk / gap Application we own Queue / topic Security / platform failure / alternate synchronous No workspace content ever reaches control-plane storage; the only thing that leaves the guest for the control plane is an event. v 1.0 · owner Platform Architecture · date 2026-10

Decisions

  • Declared ephemeral paths — /tmp, package and build caches — are excluded from the persisted delta by default, because paying to store a rebuildable cache twice a day is the easiest money the platform can waste.
  • The snapshot delta is deduplicated against the prebuild it derives from, so the stored bytes approximate the developer's work rather than the image.
  • The only thing that leaves the guest for the control plane is an event. Content goes to the snapshot store, under a tenant-scoped key.

Assumptions

  • Median persisted delta ≤ 2 GB compressed, p95 ≤ 9 GB, 180 TB of new snapshot data per day.
  • Session, command and egress records 400 days; audit 5 years, immutable; raw telemetry 14 days.

Risks

  • A tenant whose build writes into a persisted path can quietly multiply storage cost; the exclusion list is overridable, which makes it a support burden as well as a feature.
  • Command and egress records are close to a developer's keystrokes. The retention is defensible only while the purpose is stated and the access is audited.
◀  Interface Catalogue All views Storage Zones by Ownership and Rebuildability  ▶
Solution Architecture v1.0 · open source on raw/bare-metal compute · Platform Architecture · 2026-1010-workspace-data-flow