Repository navigation
Bench: migration timing across scale, browsers and pinned readers #838
Description
Completed — 2026-11-04
Implemented in #851 (merged into v3) and project-minigraf/minigraf-docs#89 (merged). Published report, frozen fixtures, raw results and reproduction protocol.
All five acceptance criteria are satisfied within the maintainer-approved bounded scope: all nine native fixtures through 200k facts; Chromium, Firefox and WebKit at 2k/20k facts; separate migration/read-only/reopen controls; supported post-migration snapshot churn. The report retains 810 checked timing samples, 117 separate untimed checks, both native sessions and their variance, exact fixture bytes/hashes, build/runtime provenance and explicit exclusions. Benchmark scratch has been removed.
The exploratory 200k-fact browser observation is untimed and excluded from measured coverage by explicit scope decision. Cold startup, broader browser lifecycle and sustained many-pin/RSS/quarantine-byte scaling remain documented exclusions coordinated with X04/X06, not outstanding acceptance work for this bounded sweep. X03 remains MANUAL; no nightly coverage is promoted.
Original scope and completed checklist
Follow-up of #776: X03 in the remaining exclusions and follow-ups table, finalized in #834.
Disposition: MANUAL migration.
Milestone: v3.next — This is a broader future migration timing sweep beyond the verified small native representative, not a new nightly release acceptance route.
Remaining scope recorded by the audit:
Native migration uses one small frozen v7 fixture, not a large dictionary/scale sweep. Browser migration and migration with pinned-reader churn have no V timing; IndexedDB open/write coverage is not migration coverage.
C07 selects open/storage_v3/{migrate_v7,read_only_v7}/multi_checkpoint. It uses one small frozen v7 multi-checkpoint corpus and validates every manifest query; copying is outside timed open. Read-only v7 is nonmigrating and must not be reported as migration. This establishes a native representative, not large dictionary/scale behavior.
Remaining work is a native v7-to-v8 sweep with larger dictionaries and fact/value distributions, browser migration timings, and migration/persistence scenarios involving pinned-reader churn where supported. Document actual supported reader lifetimes rather than implying an impossible concurrent migration contract. Browser indexeddb_write/1k and indexeddb_reopen/1k only measure write-through/reopen in the same page/context, not migration or cold startup. Coordinate browser churn timing with X04 and sustained pin scaling with X06.
Why outside the nightly shortlist: growing migration fixtures and host-specific IndexedDB/pinned-reader scenarios need a separate manual sweep; ordinary open/write representatives cannot certify these distinct costs.
Acceptance criteria:
- Define and publish a pinned native migration fixture ladder, including larger dictionary cardinalities and repeated/unique long values, with exact fact counts and expected results.
- Measure native v7-to-v8 migration separately from read-only v7 open, keeping fixture copies/setup outside timing and verifying migrated data and reopen results.
- Measure browser migration independently of IndexedDB open/write, recording browser/runtime, core/binding revisions, persistence setup and correctness checks.
- Exercise supported pinned-reader/churn migration or persistence scenarios; record explicit unsupported combinations and snapshot/pin correctness checks.
- Publish timings, scaling, fixture/build/runtime metadata and remaining exclusions without promoting unmeasured cases to nightly coverage.
Run heavy builds, enumeration, measurements and comparisons on idle hardware; retain reproducible evidence and clean temporary scratch afterward. This is benchmark/verification work aligned with PHILOSOPHY.md's reliability, portability and simplicity principles; any production change needs its own review.
Evidence: docs/bench-coverage-v3.md; validation run 41642356471. The validation run establishes only its recorded representative latency coverage; it does not complete this follow-up.
Activity
Merge pull request #851 from project-minigraf/bench/838-migration-sweep