# Test and acceptance matrix — 1.0.7 Report source, automated tests, target compilation, package/signature, browser, installed-device behavior, real POC/BOS and capacity as separate evidence. A passing fake-PUT test or a small metadata response is not production throughput certification. Historical acceptance reports apply to the revisions and installations they name. ## Required preflight - Select POC explicitly; do not use Fanyu as a target for this branch. - Upgrade API/coordinator/indexes first. Verify the authenticated capabilities response has `features.sessionHeartbeat=true`, `features.previewSessions=true` and `features.credentialActivation=true`, `features.metadataRecovery=true`, `features.missingFileSkip=true`, `features.safeClean=true`, Session heartbeat works, and Session status includes `progress.totalItems`/`progress.terminalItems`. The client must reject a missing feature before scanning/creating a Session; no compatibility fallback is expected. - Start a dedicated empty daemon namespace and confirm both installed binaries with `drosync version`. The runner does not start a daemon through a CLI alias or replace credentials on a daemon already managing other mappings. - Use the daemon to migrate schema 5; validate old journal cancellation/rescan before testing resume. Never delete source data or live WAL/SHM to prepare acceptance. - Use a small isolated fixture first. Real capacity work requires its separate cost/quota confirmation and storage budget. ## Automated and local integration checks ```bash cargo fmt --manifest-path tools/drosync/Cargo.toml --all -- --check cargo clippy --manifest-path tools/drosync/Cargo.toml --workspace --all-targets -- -D warnings cargo test --manifest-path tools/drosync/Cargo.toml --workspace go test -race ./internal/api ./internal/store ./internal/objectsync npm run lint:web npm run test:frontend python3 scripts/generate_drosync_contract.py --check python3 tools/drosync/packaging/tests/test_installation.py ``` Tests should exercise the business result and failure path, not merely the existence of an entry: | Area | Required observations | |---|---| | Single entry | Old CLI/IPC entries rejected; logged-in platform report/admin-abort routes 404; no report modal/export/stop buttons; local Web is read-only. | | CLI results | Sync failed/expired/aborted/partial/paused exit 50, conflicted exit 30; successful sync/preview exit 0 and preserve structured output. | | Pagination | Local status cursor/limit boundary; platform Session previous/next pages; conflicts fetched only after selecting a Session, with next cursors retained and cancelled stale requests. | | Scanning/integrity | SHA256/part-MD5 and upload correspond to the same source snapshot; same-size mutation is rejected; stable identity can reuse the source cache; weak identity does not bypass content validation. | | Scheduling | Continuous temporary-file writes do not starve an already stable file; initially missing/reconnected directories are watched again; event buffering and mapping concurrency are bounded. | | SQLite | Schema migration and read-only rejection; latest-file projection; keyset pages; concurrent open/metadata/receipt writes avoid deferred read-to-write lock upgrade failures. | | Cancellation | In-flight fake PUT closes before its full body, remote abort is confirmed, no later complete/commit; create-response loss replays the original key/body before abort; uncertain cancellation preserves mapping/journal. | | Retry and deadlines | Complete-response loss retries completion rather than re-uploading an already-uploaded item; Retry-After is honored; HTTP and planning/commit phases have bounded deadlines. | | Coordinator | Bounded parallel planning and commits; blocked planning does not halt later commits; planning is not announced complete early; claim loss cancels work and rejects stale terminal writes; failures appear in metrics. | | Metadata progress | Hot cached polls, dirty-page invalidation, revision races and expiry; authoritative completion recount; required indexes; fresh/stale queue snapshot semantics. | | Reports | Upload/report filters agree, .done markers excluded appropriately, read errors propagate, changed/missing source identity cannot show complete, same-day scope names differ, malicious names render as text. | | Updates/packages | Check does not download; service only stages; CLI authorization uses a privileged private copy with fixed digest/native revalidation; missing/unknown/non-native format markers and partial release matrices rejected; signing precedes final checksum; service restart/enable behavior. | Use local fake servers, independent SQLite/Mongo namespaces and temporary source fixtures for repeatable failure injection. Preserve commands, raw outcomes and artifact paths. A local probe whose assertion is deliberately red is diagnostic evidence, not a passing regression. Release CI includes format/Clippy/tests and packaging-script checks plus four target build jobs; it does not by itself run a real million-file/20TiB BOS acceptance. Long fuzz campaigns, Windows execution/linking, authorization dialogs and native signing/installation still require their own records. For browser checks use desktop and a mobile viewport: correct route/title, meaningful content, no framework overlay or unexplained console errors, a screenshot and an exercised action/result. Mocked API data proves the UI interaction, not backend authorization or real service readiness. ## Real POC functional runner The functional runner creates an isolated Space/prefix and scoped Key/Secret, drives the installed client, checks excluded entries, downloads accepted fixtures for SHA256 verification, confirms remote no-delete and records related sessions/items/audit. ```bash npm run acceptance:drosync-poc -- \ --base-url http://120.48.164.19 \ --connection-id osc_example \ --bucket acceptance-bucket ``` HTTP and HTTPS are accepted according to the selected control-plane policy. HTTP evidence must record `encrypted=false` and the exposure of credentials/metadata; add `--require-https` or `DROSYNC_ACCEPTANCE_REQUIRE_HTTPS=true` when TLS is required. Public upload/update artifact transport remains HTTPS-only. With 1.0.7, ordinary acceptance synchronization must finish successfully. An intentional conflict test must explicitly accept exit 30 and inspect the returned conflicted Session rather than treating every nonzero code as transport failure or success. Partial/aborted results must fail acceptance. Cleanup must confirm abort/remove while its device credential is still usable, then revoke/disable the isolated resources. A cleanup failure changes the outcome to FAIL; do not erase journals to hide it. Scripts and their installed binaries must be reviewed against this contract before reuse; an older script run does not certify 1.0.7. Runner evidence is stored under timestamped `.codex-artifacts/` paths with credential values redacted. Final objects remain for the agreed review/retention window. Production/customer source cleanup is outside the acceptance runner. ## Real BOS evidence Record request IDs, object evidence and failure/retry outcomes for conditional create/update and 412, direct/multipart staging, server-side multipart copy above 5GB, two devices targeting one key, disconnection/response loss/process kill, URL/lease/session expiry, credential revocation, external drift and audit/release linkage. Verify final receipts against independently downloaded SHA256 and BOS HEAD size/ETag/version. A generic object listing, local report or UI success badge is not enough. A keep-both case requires both devices/sessions, increasing fencing evidence, independently verified original/conflict objects and their audit records. BOS object publication and Mongo bookkeeping remain separate operations; test recovery across their interruption boundary. ## Capacity certification The capacity runner refuses insufficient local free space (minimum 100GiB), insecure Secret-file permissions, missing price/quota evidence or absent explicit real-cost confirmation. Reduced targets are diagnostic only (`DIAGNOSTIC_PASS`), not formal PASS. ```bash npm run acceptance:drosync-capacity -- \ --base-url http://120.48.164.19 \ --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 ``` Formal targets are one million files, a 1TiB object and a 20TiB session. Preserve the bill/quota record, elapsed time, client/server peak resources, retries, final checksums and source-retention decisions. Synthetic local SQLite/Mongo scale tests isolate metadata cost and are useful regressions; they are not real disk/network/BOS capacity evidence. Record stable-file-to-first-upload p95, useful MiB/s and files/s, scan/hash time, planning/commit queue age, metadata calls/bytes, Mongo keys/docs examined, daemon/Web RSS, cancellation latency and recovery correctness. Interpret queue age only with an available and fresh sample. Compare final source/image/binary versions, not only protocol `v1` or a health endpoint. Current end-to-end user workflows and remaining real-environment gates: [1.0.6 closure manual](../../doc/drosync功能闭环使用与验收-2026-10-08.md). ## 1.0.5 regression additions Use `DROSYNC_CLOSURE_MONGO_URI=mongodb://127.0.0.1: go test ./internal/store -run TestClosure -v` for real-Mongo credential activation/no-op replay, scoped recovery discovery, task pagination/retry and release CAS. The test refuses nonlocal/authenticated endpoints and owns only unique temporary databases. API tests include pending-token restrictions, immutable v2 signature/policy, Rust-to-Go signature fixture, preview upload denial, foreign-lease no-mutation and no partial audit success. Publisher tests simulate storage/API readback; they do not publish. Native acceptance must include initial trusted baseline capture, interrupted authorization, checksum change during approval, nonzero installer results, target daemon readiness timeout, exactly one baseline recovery, offline rollback, reboot-required, and late attempt receipts. CLI JSON must remain parseable with child installer output. Test all four real OS packages separately; a mocked installer or a cross-check is not that evidence. 1.0.6 metadata, timeout and BOS recovery changes: [repair and verification](../../doc/drosync群反馈修复与验证-2026-10-08.md). 1.0.7 当前新增能力和配置迁移以 [V1.1 需求逐项实现](../../doc/drosync每日数据上传V1.1需求审查与实现-2026-10-08.md) 和 [运行配置完整清单](runtime-configuration.md) 为准;本页引用的 1.0.6 验证数量仍仅代表历史版本。