Should I check a MySQL connection before every query?

Don't assume a WSGI application always has only one request or thread. That depends on the server configuration and deployment, and multiple worker processes or threads may handle requests concurrently. A single shared connection can cause problems in that situation. Use a connection per request or a connection pool sized for the application's concurrency, and return or close the connection afterward.

If you want database-independent code, put the retry and reconnect policy behind your own database abstraction instead of checking for a connector-specific type throughout the application. Each backend can translate its driver exceptions into a common application-level error. Also, avoid a blanket retry for every exception: syntax errors, constraint violations, and other permanent failures should be reported rather than retried.

is_connected() is useful as a quick indication of the last known state, but it isn't an active health check. It won't necessarily notice a network failure until another operation uses the connection. A common pattern is to catch the database driver's exception, reconnect once, and retry only operations that are safe to repeat.

You generally don't need to check is_connected() before every query. The status can change immediately after the check, so it can't guarantee that the following statement will succeed. Execute the query and handle the connector's exception if the connection has failed. Reconnect when appropriate, but be careful about automatically retrying writes because the server may have processed the statement before the connection dropped.