接受查询作为参数的存储过程

Question

I'm trying to write a MySQL stored procedure that accepts a SELECT query and executes it against a list of databases. Is this even possible with MySQL?

Each customer in our application has their own database on the same instance. All customer databases are identical in their schema structure. Sometimes, I need to execute a simple SELECT query but across all the customer databases (eg SELECT COUNT(*) FROM users ).

I have stored procedures for the common recurring "instance-wide" queries like the user count but I don't want to create more of those for one-time queries (eg querying for corrupted records due to by buggy code, querying for columns that we are planning to deprecate, etc.).

My current solution is that I have a node script that I run locally to generate SELECT queries for each database and then concatenate them all using UNION to generate a giant query that I then execute on the database instance.

SELECT 'customerdb1' AS customer,
       COUNT(*) AS user_count
FROM customerdb1.users
UNION
SELECT 'customerdb2' AS customer,
       COUNT(*) AS user_count
FROM customerdb2.users
UNION
SELECT 'customerdb3' AS customer,
       COUNT(*) AS user_count
FROM customerdb3.users

Answer 1

You can do it in a stored procedure with PREPARE and EXECUTE , but it's considered a security vulnerability to run arbitrary SQL in this way. You said the procedure would only be used by a special account, but just allowing the procedure to exist is a risk. What if privileges are modified and allow anyone to run the procedure?

It will also be pretty slow to run this query across many schemas, because MySQL only runs one thread per query. It will have to run the queries serially, collecting the result in a temp table.

Also you have a limit on the length of any single SQL query, which is max_allowed_packet . This should be pretty large, but you can still exceed the length if you have enough UNION terms.

I worked on a site like the one you describe, where there were many schemas, one per customer, with identical tables. When we wanted to run a query across all schemas, I would run a simple query (no UNION ) in many parallel threads, and collect the results in application code.

---

See https://thedailywtf.com/articles/For-the-Ease-of-Maintenance for a fun story about stored procedures that allow input of arbitrary SQL.