packages/workerd-spike runs the full opencode core + server stack inside a
real Durable Object via @cloudflare/vitest-pool-workers: boot with all 42
migrations journaled on real DO SQLite, session create over the HTTP API, a
complete prompt turn against a fetchMock'd OpenAI-compatible provider read
back through the durable session log cursor route, a turn that completes with
no request in flight, and recovery of a session evicted mid-turn.
Eviction is simulated with DurableObjectState.abort() between prompt-accepted
and turn-complete; a fresh instance boots over the same storage, resumes the
claimed execution, and replays the drain. The durable log stays gapless
across the death, so a consumer resuming from a pre-eviction cursor sees no
gaps and no duplicates.
The profile persists durable events because that history is what recovery
replays. Harness notes: pins pool 0.12.6 (newer pool/workerd pairings segfault
on macOS) with a patch extending the stale workerd builtin allowlist and
fixing fallback-service handling of case-insensitive filesystems, /@fs ids,
JSON requires, and unanchored module-rule globs; missing node builtins resolve
to unenv polyfills via vite aliases.
Kit Langton
22ebbe279c