Tags

Tags are a cross-cutting view across the cookbook. Each recipe carries one or more tags from a restricted vocabulary; the pages below collect every recipe by tag.

Use the group-based navigation when you know what kind of program you are building. Use the tag pages below when you know what concern you are working on — revocation, audit-chain, cost-aware, etc. — across program shapes.

All tags

• audit-chain (4) — Recipes that emit to or consume from the tamper-evident audit chain — admission, lease lifecycle, inference completion, per-tenant rollups.
• cost-aware (8) — Recipes whose primary value is cost behavior — resources that return when idle, capacity that scales with load, attribution for billing.
• durable-state (10) — Recipes whose primary concern is durable data on fabric-leased storage — caches, KV stores, session stores, content-addressed object stores, distributed structures.
• encryption (4) — Recipes whose primary concern is encryption with lease-scoped key lifecycle — at-rest encryption, time-bounded secret vaults, clean rooms.
• fencing (3) — Recipes that use fenced leases for leader election, stale-write rejection, and conflict resolution. Fencing is the coordination primitive.
• gpu (20) — Recipes that run on a leased GPU. Includes single-tenant CUDA basics, shared inference servers, and the revocation/recovery story for mid-decode lease loss.
• inference (12) — LLM inference recipes — single-tenant, multi-tenant, batched, speculative, and the audit / attribution pattern that closes the per-request loop.
• multi-cloud (7) — Recipes that span more than one provider — pop-up bursts, mirrored pipelines, cross-cloud workflows, provider migration.
• multi-tenant (11) — Recipes where more than one tenant shares the same fabric resource (GPU, ingress, scheduler quota) with isolation, fair-share, and per-tenant accounting.
• observability (3) — Recipes about observing fabric programs — structured events, audit-chain consumers, operator dashboards.
• placement (8) — Recipes that constrain WHERE a workload runs — data affinity, anti-affinity, GPU generation targeting, drain handling.
• requires-cuda (4) — Recipes that depend on a real CUDA device (silicon-validated).
• revocation (6) — Recipes that handle a lease being reclaimed — either at the coordination layer (stale epoch), the operations layer (node drain), or the GPU layer (intra-kernel revoke).
• rpc (2) — Recipes that build typed RPC interfaces between services — zero-copy shared-memory transports and conventional request/response.
• streaming (6) — Recipes that build stream-processing or event-pipeline shapes — stages, queues, backpressure, durable handoff.
• testing (3) — Recipes about testing fabric programs — the deterministic harness, output equivalence pins, and the patterns for testing distributed code without a distributed system.
• web (2) — Recipes that expose HTTP / web ingress on leased fabric resources — public-facing servers, async-with-receipts APIs.
• workflow (5) — Recipes that build durable workflow engines, batch jobs, and orchestrated multi-step processes.