SQL Server as a surface: seven sql_* tools over a container this row owns, or
the sqlcmd already on the machine's PATH. Host-only, zero npm dependencies, no
bundled engine, no route and no browser tab.
sql_status the four facts, and the next move
sql_server status | start | stop | restart | remove | logs
sql_driver status | plan | install (ODBC, machine-wide)
sql_query one statement, parsed
sql_script a .sql file, streamed and streamed-in
sql_database list | create | tables | columns | count
sql_restore a .bak, with its files MOVEed here
This pack could already drive Docker (dsh-docker) and a page could already speak
to a database from JavaScript (dsh-research's ledgers). Neither could answer the
question a person asks when they clone a repository that expects SQL Server:
what do I run on this machine, and why does
pyodbcstill fail after the container is up?
The answer spans four things that fail independently, and every wasted afternoon comes from collapsing them:
| fact | owner | when it is missing |
|---|---|---|
| a server | sql_server (a container) or a native service |
nothing to connect to |
| a client | sqlcmd on PATH, or the one in the image |
no statement can be sent |
| a listening port | the server, once it is really ready | "running" but not accepting logins |
| an ODBC driver | the HOST, machine-wide | the server works and the app cannot connect |
- A development server is a container.
mcr.microsoft.com/mssql/server:2022-latestis 1.7 GB, starts in ~40 s, keeps its databases in the named volumedsh-sql-data(soremoveis not data loss), and is removed with one call. A native instance (MSSQLSERVER,MSSQL$SQLEXPRESS) is DETECTED and reported and deliberately not managed: this row will not stop a service somebody's other work depends on. - Ready means it answered, not that it started. SQL Server opens 1433 before it
accepts a login.
startpolls a realSELECT 1through the image's own client and reports the attempts. - A dump goes through
-i, never stdin. Measured on a real SSMS dump: piped through stdin it loaded 0 of 23 rows and exited 0; with-iit loaded 23 of 23. sqlcmd's stdin reader mangles the very long lines a dump is made of, and a silent partial load is the worst failure a database tool can have. - The ODBC driver is the half a container cannot give you. It is installed by an
MSI that needs an administrator, so
sql_driver action=installrequiresconfirm: "install"and then runs the platform's own command (winget / brew / apt) and reports its exit code instead of assuming success.
Windows ships a driver registered as SQL Server. pyodbc.drivers() lists it, code
finds it, and the connection can fail with "the client and server cannot communicate,
because they do not possess a common algorithm" — it predates TLS 1.2. Measured
both ways: against a stock mssql 2022 container that accepts an unencrypted
connection it does connect; against a server with forced encryption it does not.
So it sorts last and is called out, but it is not refused: a connection that works is
the answer to the question.
- install, start, stop or remove a native SQL Server;
- take a backup (
sql_restorerestores; a backup is a file the server writes); - author schema beyond
CREATE DATABASE; - remove a volume (the container, yes, and only with
confirmequal to its name); - hide a failure: every answer names the transport used, the elapsed time, and the
server's own
Msg <n>, Level <l>line.
| variable | default | meaning |
|---|---|---|
DSH_SQL_HOST |
127.0.0.1 |
server host |
DSH_SQL_PORT |
1433 |
port |
DSH_SQL_USER |
sa |
login |
DSH_SQL_SA_PASSWORD |
— | password (never echoed; SQLCMDPASSWORD to the host client) |
DSH_SQL_CONTAINER |
dsh-sql-server |
container whose sqlcmd is borrowed |
DSH_SQL_CONTAINER_CLI |
docker, else podman |
the container CLI |
DSH_SQL_IMAGE |
mcr.microsoft.com/mssql/server:2022-latest |
the image a start pulls |
DSH_SQL_VOLUME |
dsh-sql-data |
named volume for /var/opt/mssql |
sql-server-ops— the four facts, the lifecycle, the driver on each OS, the connection string an application needs, and how to read a state that looks wrong.sql-script-ingest— loading a dump: encoding from the BOM,-ivs stdin (with the measurement), dropping the dump's ownCREATE DATABASEheader, when one bad batch must not stop the load, verifying with row counts, and the tables a "complete" dump turns out not to contain.
npm run check # node checks/check-sqlserver-node.mjs - 69 assertions, no server
The package check drives every branch through scripted runners: the registration,
the refusal that happens before an install, the password policy before docker run,
a whole start end to end, the sanitiser, the grid parser, the three-part table name,
the MOVE clauses, and the BOM detection against real files.