SuperAPI AI Directory

azuresql-db-auth

Data & analytics

View repository

Updated 10-16 06:47

About this skill

Connect securely to the Azure SQL Database container (least-privilege user, auth, secrets)

sa is a bootstrap/admin login for provisioning, not what your application should connect as. This skill wires the app to a least-privilege user, picks the auth method per environment (SQL locally, Microsoft Entra or managed identity in the cloud, changing only the connection string), secures the connection, and keeps the secret out of source control.

Re-measured on 2026-09-19 against the container image tag 18.0.226_4_147, reporting EngineEdition 5, Edition SQL Azure, build 12.0.2000.8. All seven executable checks behind this skill pass, and three of the numbers this skill used to print were wrong: a contained user is refused with Msg 33233 and not Msg 15007, SET CONTAINMENT = PARTIAL is refused with Msg 12824 and not Msg 12844, and CREATE USER ... FROM EXTERNAL PROVIDER on a container started without Entra configuration is refused with Msg 33134 and not Msg 37525. Each statement is still refused, so every instruction here is unchanged; only the numbers to expect are. The fixed database roles this skill grants were confirmed on the same run. The cloud side of this guidance, Azure Key Vault and managed identity, was not measured and comes from Microsoft Learn.

Load-bearing facts (inlined; full engine detail in azuresql-db-container)

  • This is the Azure SQL Database engine (Private Preview), not the SQL Server image mcr.microsoft.com/mssql/server. SERVERPROPERTY('EngineEdition') returns 5, Edition returns 'SQL Azure'.
  • Image: sqldbpreview-dpgaeqhmgphzd4bk.azurecr.io/azure-sql/db-dev:latest (x64; on a non-x64 host add --platform linux/amd64). Required env ACCEPT_EULA=Y + a complex MSSQL_SA_PASSWORD. Engine listens on 1433.
  • The engine does NOT auto-create databases. CREATE DATABASE appdb on a master connection first; do not USE to switch databases (a user-database session returns Msg 40508); select the database in the connection string.
  • Apps read one SQL_CONNECTION_STRING env var; strings use User Id= / Password= / Database= and TrustServerCertificate=true for the local self-signed cert.
  • Container-specific and verified: a SQL contained user (CREATE USER ... WITH PASSWORD) does not work on the container today, and you cannot turn it on. CREATE USER ... WITH PASSWORD fails with Msg 33233 ("You can only create a user with a password in a contained database"). ALTER DATABASE ... SET CONTAINMENT = PARTIAL fails with Msg 12824, which asks for the contained database authentication setting to be 1, and sp_configure does not exist on this engine (Msg 2812), so there is nothing to set. Create a SQL app identity as a server login mapped to a database user instead. This is the inverse of Azure SQL Database in the cloud, where the contained user is the norm. Measured 2026-09-19 on image tag 18.0.226_4_147.
  • Entra has to be configured on the engine before you can use it. On a container started with no Microsoft Entra ID configuration, CREATE USER [name] FROM EXTERNAL PROVIDER is refused with Msg 33134, "Principal '...' could not be resolved. Error message: 'Unable to query Azure AD certificate from local cert store.'" That is a missing engine configuration, not a broken statement: enable Entra first (see references/entra-auth.md in the azuresql-db-container skill for how to enable Entra Authentication) and the same statement then works. Measured 2026-09-19 on image tag 18.0.226_4_147.

Step 1: create a least-privilege user (not sa)

Do provisioning as sa, then give the app its own identity with only the roles it needs. The working recipe differs by environment, but the app code does not (the app just connects with a username and password, or an Entra token).

Local container (SQL auth): create a server login on master, map a database user to it in appdb, and grant only the roles the app needs.

Set AZURE_SQL_APP_PASSWORD from a secret source to a fresh, deployment-unique value that differs from the sa password. Require it before provisioning and substitute it for <AZURE_SQL_APP_PASSWORD> in the clearly labeled non-executable SQL templates. Render the templates at execution time using AZURE_SQL_APP_PASSWORD from the configured secret source.

The app then connects as applogin, never sa.

Cloud (Azure SQL Database) or Entra anywhere: prefer a contained user. For Entra (which works on the container too, once enabled), use CREATE USER [name] FROM EXTERNAL PROVIDER in appdb. Enable Entra on the engine first by following references/entra-auth.md in the azuresql-db-container skill before running the statement; without that configuration the statement is refused with Msg 33134. In the cloud with SQL auth, CREATE USER ... WITH PASSWORD is the norm there. Full recipes for every path are in references/auth-and-secrets.md when you need a complete provisioning or secret-handling recipe.

Step 2: pick the auth method per environment (only the connection string changes)

  • Local: SQL auth. sa bootstraps; the app connects as the least-privilege applogin. Server=localhost,1433;Database=appdb;User Id=applogin;Password=...;Encrypt=true;TrustServerCertificate=true.
  • Cloud (Azure SQL Database): prefer a token-based identity over a password. In production, use Authentication=Active Directory Managed Identity rather than Active Directory Default: Default walks a credential chain (DefaultAzureCredential) that is slower and ambiguous under load, while a specific method skips the chain. Microsoft.Data.SqlClient caches the token, so refresh is occasional, not per-connection. This is still a connection-string-only change, so the app code does not change (see the azuresql-db-local-to-cloud skill).

Step 3: secure the connection

  • Encrypt=true everywhere (the default in modern drivers). Encrypt the TLS channel in both local and cloud.
  • TrustServerCertificate=true only locally, to accept the container's self-signed cert. Never set it against Azure SQL Database in the cloud, where the certificate is real and validating it is the point.

Step 4: keep the secret out of source

The connection string carries a credential. Never commit it or the SA password.

  • Read it from one env var, SQL_CONNECTION_STRING; put local values in a .env that is git-ignored (or dotnet user-secrets for .NET).
  • In the cloud, store it in Azure Key Vault and reference it, or use managed identity so there is no password to store at all.

For a .NET project the local value goes in the secret store rather than in a file that can be committed:

Store the complete rendered connection string with .NET user-secrets or the IDE secret manager under the key your app already reads. Keep the real password out of source and shell history; see references/auth-and-secrets.md.

Open references/auth-and-secrets.md when you need the per-stack handling for Key Vault, user-secrets or .env.

Validation rules

  • The app connects as a least-privilege identity, not sa; it has only the roles it needs (no db_owner/admin).
  • On the container, the SQL app identity is a server login + mapped database user (CREATE LOGIN on master, then CREATE USER ... FOR LOGIN in appdb), not a contained CREATE USER ... WITH PASSWORD (which fails on the container). The database user and its role grants are created on the appdb connection, not via USE.
  • Every connection string sets Encrypt=true; TrustServerCertificate=true appears only for the local container, never for the cloud.
  • The connection string / SA password is read from a secret store or a git-ignored env var, never committed.
  • Cloud auth prefers managed identity / a specific Entra method over Active Directory Default in production; only the connection string changes.
  • 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"

Check it worked

  • Positive verification: Reconnect with the application identity and confirm the expected database user, reader/writer role memberships, and allowed application query.
  • Cleanup verification: Remove temporary logins/users and confirm the application connection and test objects no longer exist.
  • Report success only after both observations pass.

Do not

  • Do not connect the application as sa; sa is for provisioning only.
  • Do not try to create a SQL contained user with CREATE USER ... WITH PASSWORD on the container; it fails with Msg 33233. Do not try to turn partial containment on either: ALTER DATABASE ... SET CONTAINMENT = PARTIAL fails with Msg 12824, and the sp_configure setting it asks for does not exist on this engine. Use a server login plus a mapped database user locally.
  • Do not run CREATE USER ... FROM EXTERNAL PROVIDER against a container with no Entra configuration; it is refused with Msg 33134. Configure Entra on the engine first.
  • Do not grant the app db_owner or server admin when read/write roles suffice.
  • Do not commit the connection string or the SA password; use a secret store or a git-ignored env var.
  • Do not set TrustServerCertificate=true against Azure SQL Database in the cloud; that disables cert validation on a real certificate.
  • Do not lean on DefaultAzureCredential's full chain in a hot production path; pick a specific auth method (managed identity) so token acquisition is fast and predictable.
  • Do not use the SQL Server image mcr.microsoft.com/mssql/server; this is the Azure SQL engine.

References

  • references/auth-and-secrets.md: open it when creating least-privilege users, selecting an environment-specific connection string, or configuring Azure Key Vault, dotnet user-secrets, or .env.

Staying current

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

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.