lmxopcua/docs/plans/evidence/2026-11-10-opcua2/backend-matrix.md

Local historian backend matrix — 2026-11-10

Candidate a8e1082d69a112ea0e16788f023067979698527a (SDK 2.0.1) compared with baseline 83ed880e44 (SDK 1.5.378.182). This is the existing direct backend integration test matrix over real local SQL Server and QuestDB. It is not an OPC UA wire test, AVEVA validation, or stress/soak approval. No source edits, expected-value changes, new skips, suppressions, issue updates, pushes, or commits were made for this task.

Version	Backend	Passed	Failed	Skipped	Build / test / guard exit
candidate	SqlServer	195	0	0	0 / 0 / 0
candidate	QuestDb	233	0	1	0 / 0 / 0
baseline	SqlServer	195	0	0	0 / 0 / 0
baseline	QuestDb	233	0	1	0 / 0 / 0

Every invocation was unfiltered, used a fresh SHA/project run directory and TRX, preserved the dotnet test exit status, and ran the repository MIN_EXECUTED=1 guard. Project builds were serial within each worktree; test invocations were serial against the same isolated fixture endpoints. Exact commands, UTC start/end, SDK dependency versions and test/SDK binary SHA-256 hashes are in each result.json.

Candidate positive coverage includes, on each backend, 13 raw-read, 14 AtTime, 11 modified-value, 7 event-read, 12 update, 64 processed-read, 6 bounding-value and 4 quality-contract conformance cases, plus backend-specific schema, write, provisioning, retention (SQL), tombstone/index/TTL (QuestDB), and alarm-writer tests.

SqlServer: baseline/candidate display names and outcomes identical = True across 195 discovered cases. Baseline failures: 0; candidate failures: 0. QuestDb: baseline/candidate display names and outcomes identical = True across 234 discovered cases. Baseline failures: 0; candidate failures: 0.

No PercentGood failure appeared in these selected integration suites. Their empty failure sets match where both versions have completed; there is no failing message pair to compare. This does not resolve or replace the separate standalone SDK-oracle PercentGood issue #796.

The existing QuestDB LiveWriteReadRaceStressTests.Per_sample_reads_of_a_tag_being_written_never_fault_and_never_return_a_valueless_non_bad_result case skips because HISTORIAN_QUESTDB_STRESS_SECONDS is unset. It was not enabled, shortened, reclassified, or replaced by a normal passing case. No five-million-row workload or two-hour stress/soak was run. Those gates remain pending.

Fixture ownership and exact images

SQL Server: existing otopcua-dev-sql-1, endpoint 127.0.0.1:14330. Its documented local dev credential was read from docs/Historian.md and supplied only via environment; the password and full secret-bearing connection string were not printed or included in the report. SqlHistorianFixture was verified before execution: it forces master only for CREATE/DROP, generates OtOpcUaHistorian_T_, directs all test schemas into that new database, clears its pools, and drops only its generated name. The fixture source is unchanged between baseline and candidate. Five preexisting OtOpcUaHistorian_T_GUID databases were recorded and were never dropped by this task.

SQL container ID: 74ea75e5904c647f18e89b8b14c9ac385d7149c0484fbf13f6c16ec00cf3e24f; image ID: sha256:a2fbff321505724b8a271a214d4dec020a8e7a745c83eede0b9d762b51a9ab34.

QuestDB: new task-owned otopcua-sdk2-questdb, loopback-only PG port 32770 and HTTP port 32771, new storage volume otopcua-sdk2-questdb-data-20261110t170141z. No existing QuestDB service or volume was used. Connection settings use ServerCompatibilityMode=NoTypeLoading and Maximum Pool Size=16; the fixture applies its existing one-second idle pruning. QuestDbFixture assigns unique t<12 hex>_ prefixes, validates them, locates owned tables with starts_with (not wildcard LIKE), and drops only those prefixes. The same unchanged fixture source and image were used by both versions.

QuestDB image: questdb/questdb:10.0.1; immutable image ID sha256:41ceca901724ad10a0035da0cb0c1f99f88bf3692abdc9090baabdfff1d215f9; digest questdb/questdb@sha256:67eaed863ebb2383227919ea9a5499a4a39ceca3862c20b974c84a10e11cdf89.

Run evidence

Version / backend	UTC start	UTC end	Directory
candidate SqlServer	2026-11-10T17:03:02.101289+00:00	2026-11-10T17:03:52.950418+00:00	/Users/dohertj2/.codex/worktrees/opcua2-test-convergence/OtOpcUa/artifacts/opcua2/a8e1082d/backend-matrix-20261110T170141Z/SqlServer
candidate QuestDb	2026-11-10T17:03:58.392642+00:00	2026-11-10T17:11:13.275488+00:00	/Users/dohertj2/.codex/worktrees/opcua2-test-convergence/OtOpcUa/artifacts/opcua2/a8e1082d/backend-matrix-20261110T170141Z/QuestDb
baseline SqlServer	2026-11-10T17:11:19.005717+00:00	2026-11-10T17:12:11.817250+00:00	/Users/dohertj2/.codex/worktrees/opcua2-test-convergence/OtOpcUa/artifacts/opcua2/83ed880e/backend-matrix-20261110T170141Z/SqlServer
baseline QuestDb	2026-11-10T17:12:16.867323+00:00	2026-11-10T17:19:37.746023+00:00	/Users/dohertj2/.codex/worktrees/opcua2-test-convergence/OtOpcUa/artifacts/opcua2/83ed880e/backend-matrix-20261110T170141Z/QuestDb

Session index / fixture metadata / aggregate results / exact failure-set comparison:

/Users/dohertj2/.codex/worktrees/opcua2-test-convergence/OtOpcUa/artifacts/opcua2/a8e1082d/backend-matrix-20261110T170141Z

Each project directory contains build.log, test.log, a fresh TRX, guard.log and result.json. Candidate directories also contain class-counters.json. The session has fixture-metadata.json and readiness/WAL snapshots; cleanup.json and questdb-container.log are written after final cleanup.

Cleanup

Completed at 2026-11-10T17:19:44.751608+00:00. SQL GUID-database inventory unchanged: True; SQL container ID unchanged: True. QuestDB tables remaining after all tests: 0. Task-owned container absent: True; task-owned volume absent: True. Only the captured task container ID and its verified task-labelled volume were removed. No other container, volume or database was removed.

Reproduction and remaining gates

Completed scripts (kept in /tmp; they create new timestamped evidence directories and protect ownership):

python3 /tmp/opcua2-backend-setup.py
python3 /tmp/opcua2-backend-matrix.py
python3 /tmp/opcua2-backend-cleanup.py
python3 /tmp/opcua2-backend-report.py

For each version/backend the actual commands were dotnet build <owning-project.csproj> -c Release, then dotnet test <owning-project.csproj> -c Release --no-build --no-restore --logger trx --results-directory , with only the appropriate real connection environment variable set. Guard: MIN_EXECUTED=1 bash scripts/ci/assert-not-all-skipped.sh /*.trx. The orchestration preserves nonzero test and guard statuses instead of overwriting them. SQL connection acquisition is from the documented local dev recipe without echo; QuestDB endpoints are obtained from the newly created task container.

AVEVA/HistorianGateway live read/write/alarm/provisioning and licensed backend validation remain external and pending. The separate SDK-oracle issue #796, wire/security interoperability, five-million-row workload, two-hour stress/soak and release/performance gates are not completed by these local backend runs.