test: add differential query compatibility oracle

[viggy28]

Summary

Add a purpose-built differential query compatibility harness that treats source Postgres as the oracle and compares it with Streambed through the actual Postgres-wire endpoint.

The suite runs the same deterministic fixture and query matrix against both Iceberg and DuckLake targets. It compares query behavior, column names, Postgres OIDs, nulls, row multiplicity, ordering, and type-aware values.

Closes #78.

Implementation

• Add test/querycompat with a deterministic Postgres fixture and declarative query cases.
• Build and start the real Streambed binary for sync and standalone query serving.
• Wait for the fixture commit LSN instead of using fixed sleeps.
• Connect to source Postgres and Streambed with pgx simple protocol.
• Compare integers/text exactly, decimals with exact rational arithmetic, floats with OID-specific tolerances, timestamps by their SQL semantics, timestamptz in UTC, JSON structurally, and UUID/bytea canonically.
• Run the same cases for Iceberg and DuckLake with explicit supported, unsupported, and known-difference classifications.
• Add the oracle to pull-request and main-branch CI and to scripts/test-integration.sh.

Compatibility findings fixed

Running the oracle against real services exposed three correctness issues that are fixed here:

• Advertise standard_conforming_strings=on so pgx can use the simple query protocol.
• Normalize DuckDB's binary UUID result into canonical Postgres UUID text before pgwire encoding.
• Write Postgres timestamp to Parquet with isAdjustedToUTC=false, while retaining true for timestamptz; this prevents local timezone shifts when Iceberg is queried.

Known storage-type differences remain explicitly classified for numeric, unconstrained numeric, varchar, JSONB, and Iceberg smallint mapping.

Validation

Started fresh Postgres and MinIO containers, built Streambed, synced the fixture separately to Iceberg and DuckLake, started the Streambed query server, and compared every query through pgwire:

docker compose -f test/integration/docker-compose.yml down -v
docker compose -f test/integration/docker-compose.yml up -d postgres minio --wait
docker compose -f test/integration/docker-compose.yml up createbucket
go test -tags integration -v -count=1 -timeout 10m ./test/querycompat

Result: PASS for Iceberg and DuckLake.

Also ran:

go test ./internal/... ./config/...
go test -race ./internal/server ./internal/parquet
git diff --check
bash -n scripts/test-integration.sh

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b98b0547c6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Stop sync before deferred slot cleanup

If this target fails after the sync process starts but before the explicit stop, this t.Cleanup runs only after runTargetCompatibility's deferred cleanupSource; the slot is therefore still active when cleanup filters on active = false, so it is left behind while its publication is removed. Because every run uses a new slot name and the test Postgres permits only four replication slots, repeated failed local runs can exhaust all slots and prevent subsequent tests from starting. Arrange process termination to run before the database cleanup on every exit path.

Useful? React with 👍 / 👎.