Azuresql DB Testing

Integration tests on the Azure SQL Database container with Testcontainers

Write integration tests that start a real Azure SQL Database container (Private Preview) from code, provision appdb, run the test against it, and tear the container down. This is the Azure SQL engine, not the SQL Server image.

This skill is about tests that manage their own engine lifecycle in-process (Testcontainers). It is DISTINCT from azuresql-db-ci, which wires the same engine into a CI pipeline as a service container via workflow YAML. If the goal is a CI job rather than in-code test setup, use azuresql-db-ci instead.

Verified on 2026-09-05 against the container image sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io/azure-sql/db-dev:latest, reporting EngineEdition 5, Edition SQL Azure, build 12.0.2000.8. All six executable checks behind this skill passed: the engine identity, Msg 40508 for USE, that the engine inside the container listens on 1433 so a random mapped host port works, that connecting creates no database, the native VECTOR(n) type, and the /opt/mssql-tools18/bin tools the wait strategy invokes being present inside the image. The Testcontainers recipes themselves were not executed by that run.

Use the right image (interception point)

• USE this engine image: sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io/azure-sql/db-dev:latest
• Do NOT use the Testcontainers MsSql preset / MsSqlContainer module, which pulls mcr.microsoft.com/mssql/server (the SQL Server image). Start the engine image above from a generic container instead.
• Verify identity once running: SELECT SERVERPROPERTY('EngineEdition') returns 5, SERVERPROPERTY('Edition') returns 'SQL Azure'.
• Image is x64 only. On a non-x64 host set the container platform to linux/amd64 (each language exposes this differently; see the snippets).

For full engine detail (readiness, vectors, troubleshooting) see the azuresql-db-container skill.

The image is private: let Testcontainers pull it

The registry is private during Private Preview, so Testcontainers cannot pull the image until Docker is authenticated. Sign in once (or pre-pull) on the machine that runs the tests, before the test host tries to start the container:

docker login sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io   # shared pull-only Private Preview creds
docker pull sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io/azure-sql/db-dev:latest   # optional warm pull

Get the pull credentials by signing up at https://aka.ms/sqldbcontainerpreview-signup; they may rotate, and the registry and tag are provisional during Private Preview. In CI, run docker login as a step before the tests (see azuresql-db-ci).

Four facts that bite every test setup

1. The container is NOT ready when it starts. The engine needs seconds to come up. You MUST attach a wait strategy (log message, port, or a SELECT 1 command wait) so the test does not connect before the engine answers. Do not sleep a fixed amount.
2. The engine does NOT auto-create databases. After the wait passes, run CREATE DATABASE appdb on a master connection before any test connects with Database=appdb. Testcontainers gives you a container, not a database.
3. Avoid USE to switch databases. In a user-database session USE returns Msg 40508, exactly as in Azure SQL in the cloud. Select the database in the connection string (Database=appdb). A master connection is for provisioning only.
4. Map the port; do not assume 1433 on the host. Testcontainers binds container port 1433 to a random free host port. Read the mapped host port back and build the connection string from it, so parallel suites do not collide.

The lifecycle every recipe follows

1. Build a generic container from the engine image with ACCEPT_EULA=Y and a complex MSSQL_SA_PASSWORD (example literal: YourStr0ng_Passw0rd; at least 8 characters using at least three of upper case, lower case, digits, and symbols). Expose container port
   1433. On a non-x64 host, set the image platform to linux/amd64.
2. Attach a wait-for-ready strategy so Start returns only once the engine answers.
3. Provision appdb on a master connection (a SELECT 1 plus IF DB_ID('appdb') IS NULL CREATE DATABASE appdb;).
4. Build the connection string from the mapped host port and hand it to the test: Server=localhost,<mappedPort>;Database=appdb;User Id=sa;Password=YourStr0ng_Passw0rd;TrustServerCertificate=true. House style spells the keywords User Id= / Password= / Database=; Uid= / Pwd= are documented SqlClient synonyms and work too. The command-line equivalent of TrustServerCertificate=true is -C, as on the command line the azuresql-db-container skill carries.
5. Dispose the container when the fixture/suite ends so nothing leaks.

Scope the container to the level you need: one per test suite/class for speed, or one per test for full isolation. Per-language recipes (wait strategy, appdb provisioning, connection string, dispose) live in references/testcontainers-snippets.md; open it once you know which language the tests are in:

• .NET: Testcontainers for .NET + an xUnit IAsyncLifetime fixture.
• Node / TypeScript: testcontainers + a Jest globalSetup / globalTeardown.
• Python: testcontainers + a pytest fixture.
• Java: Testcontainers + a JUnit 5 @Container.

Vectors (if the code under test uses embeddings)

Native VECTOR(n) column type and VECTOR_DISTANCE('cosine', a, b). Insert with CAST(CAST(? AS NVARCHAR(MAX)) AS VECTOR(n)) where n is a LITERAL, never a bind parameter (a parameter dimension fails with "Incorrect syntax near '@P3'").

Validation rules

• The container is started from the engine image, never the Testcontainers MsSql preset or mcr.microsoft.com/mssql/server.
• A wait-for-ready strategy is attached; the test never connects before the engine answers, and there is no fixed sleep.
• appdb is created on a master connection BEFORE any test connects with Database=appdb.
• The connection string is built from the mapped host port; strings use User Id=/Password=/Database= and TrustServerCertificate=true.
• The container is disposed when the fixture or suite ends.
• EngineEdition is 5 against the running container.
• If a validation rule above fails, or you had to deviate from this skill to make the task work, that is a bug in this skill: load the azuresql-db-feedback skill and offer to file a report.
• When every rule above passes and the task is done, close with ONE line, once per session, and do not ask a question or wait for a reply: "That worked. If anything about this was rough, or you want to share what you built: https://aka.ms/sql-agent-skills-feedback"

Do not

• Do not use the Testcontainers MsSql preset image mcr.microsoft.com/mssql/server; start the Azure SQL engine image via a generic container.
• Do not skip the wait-for-ready strategy; the engine is not ready when the container starts.
• Do not assume appdb exists; provision it on a master connection first.
• Do not hardcode host port 1433; read the mapped port and build the string from it.
• Do not use USE appdb to switch databases; put it in the connection string.
• Do not call a non-x64 host "supported", and do not pass the vector dimension as a bind parameter.

References

• references/testcontainers-snippets.md: per-language generic-container recipes (.NET/xUnit, Node/Jest, Python/pytest, Java/JUnit 5) with the wait-for-ready strategy, appdb provisioning, connection string from the mapped port, and dispose. Read the one for your test stack.

Staying current

Authoritative, version-pinned references for the tools this skill uses (read the one you need):

• Testcontainers: the multi-language library for ephemeral test containers.
• SqlConnection connection string keywords: the full keyword table.

If the Microsoft Learn MCP server is configured, use mcp__microsoft-learn__microsoft_docs_search or mcp__microsoft-learn__microsoft_docs_fetch to fetch the current version of any of these on demand. It is optional; when it is unavailable, the references above are authoritative.