droboticflow/docs/drosync/testing-acceptance.md
2026-07-29 02:50:00 +08:00

66 lines
4.1 KiB
Markdown

# Test and acceptance matrix
Completion is reported independently for source, automated tests, platform artifacts, browser, POC API, real BOS, signatures, and capacity.
## Repeatable runners
Functional POC acceptance creates an isolated Space/prefix and scoped client Key/Secret, installs no secret in its evidence, drives the installed macOS client twice, verifies excluded local entries, downloads every accepted fixture to recompute SHA-256, confirms remote no-delete, and links sessions/items/audit:
```bash
npm run acceptance:drosync-poc -- \
--base-url http://poc.example.com \
--connection-id osc_example \
--bucket acceptance-bucket
```
Both runners accept HTTP and HTTPS by default. An HTTP run records
`encrypted=false` plus an accepted-risk warning; it is not by itself a functional
acceptance blocker. Add `--require-https` or set
`DROSYNC_ACCEPTANCE_REQUIRE_HTTPS=true` when validating a deployment whose policy
requires TLS.
The formal capacity runner refuses a local volume with less than 100 GiB free, a group/world-readable Secret file, missing BOS price/quota records, or an absent explicit real-cost confirmation. Its defaults are the formal targets; reduced values produce `DIAGNOSTIC_PASS`, never `PASS`.
```bash
npm run acceptance:drosync-capacity -- \
--base-url http://poc.example.com \
--connection-id osc_example \
--bucket acceptance-bucket \
--space-id oss_capacity \
--access-key-id dsk_example \
--secret-file /run/secrets/drosync-capacity \
--volume /mnt/drosync-capacity \
--bos-cost-record evidence/bos-price.json \
--bos-quota-record evidence/bos-quota.json \
--confirmation I_ACCEPT_REAL_20_TIB_BOS_COST
```
Both runners write timestamped reports below `.codex-artifacts/` by default. Reports redact passwords, Secret values, bearer tokens, and hashes used to validate them. The functional runner always attempts cleanup in `finally`: it revokes the scoped client credential, disables the isolated Space, and removes its local mapping. Cleanup receipts are part of the report, and a failed cleanup changes the run result to `FAIL`; final objects remain available for the configured review window.
## Automated
- Go: API/store/coordinator unit tests, Mongo integration, race detector, status transitions, idempotency replay/mismatch, lease fencing, cleanup, reconcile, quotas, permission gates, and audit chain.
- Rust: format, Clippy with warnings denied, workspace/doc tests, SQLite migrations, deterministic path properties and manifest fuzz corpus, crash/reopen resume, multipart limits, and IPC permission tests. Long-running `cargo-fuzz` campaigns remain a release-certification gate rather than a claim made by routine CI.
- React: role visibility, capability gate, scoped Key/Secret management, automatic device registration, conflict, release, audit export, and rollout policy.
- Contract: parse OpenAPI 3.1, regenerate Go/Rust status models, and compare protocol fixtures.
## Real BOS
Record request IDs and object evidence for:
- conditional create/update plus HTTP 412;
- direct and multipart staging;
- multipart server-side copy above 5 GB;
- two independent devices creating/updating one key;
- response loss, disconnection, process kill, URL/lease/session expiry, and token revocation;
- external write drift that disables the space;
- traceability among final HEAD/version, session, item, audit, release manifest, and daily seal.
Verify uploads with the drosync receipt and BOS HEAD size/ETag/versionId. A generic object-list entry or UI success badge is not acceptance evidence.
A keep-both scenario is complete only when both devices and sessions are linked, the second reservation proves monotonic fencing, the original and conflict objects are independently downloaded and SHA-256 checked, and the corresponding audit-chain events are present.
## Capacity
Routine CI uses streaming/simulated million-entry manifests and sparse/fixture payloads. Formal certification must preserve the real BOS bill, elapsed time, peak client/server resources, failure/retry log, and final checksum records for 1 million files, one 1 TiB object, and a 20 TiB session.