Cloud Functions "Connection Refused" when connecting to Cloud SQL

I was having the same issue and after much back and forth with Google, this finally solved it for me.

The sample shows the following try clause:

if not mysql_conn: try: mysql_conn = pymysql.connect(**mysql_config) except OperationalError: # If production settings fail, use local development ones mysql_config['unix_socket'] = f'/cloudsql/{CONNECTION_NAME}' mysql_conn = pymysql.connect(**mysql_config) 

I changed it to:

if not mysql_conn: mysql_config['unix_socket'] = f'/cloudsql/{CONNECTION_NAME}' mysql_conn = pymysql.connect(**mysql_config) 

And used (data, context) as function arguments rather than (request)

This is what I ultimately ended up with as a test function:

from os import getenv import pymysql from pymysql.err import OperationalError CONNECTION_NAME = getenv( 'INSTANCE_CONNECTION_NAME', 'connection_name') DB_USER = getenv('MYSQL_USER', 'username') DB_PASSWORD = getenv('MYSQL_PASSWORD', 'password') DB_NAME = getenv('MYSQL_DATABASE', 'database') mysql_config = { 'user': DB_USER, 'password': DB_PASSWORD, 'db': DB_NAME, 'charset': 'utf8mb4', 'cursorclass': pymysql.cursors.DictCursor, 'autocommit': True } if getenv('NODE_ENV', '') == 'production': mysql_config['unix_socket'] = f'/cloudsql/{CONNECTION_NAME}' # Create SQL connection globally to enable reuse # PyMySQL does not include support for connection pooling mysql_conn = None def __get_cursor(): """ Helper function to get a cursor PyMySQL does NOT automatically reconnect, so we must reconnect explicitly using ping() """ try: return mysql_conn.cursor() except OperationalError: mysql_conn.ping(reconnect=True) return mysql_conn.cursor() def mysql_demo(data, context): global mysql_conn # Initialize connections lazily, in case SQL access isn't needed for this # GCF instance. Doing so minimizes the number of active SQL connections, # which helps keep your GCF instances under SQL connection limits. if not mysql_conn: mysql_conn = pymysql.connect(**mysql_config) # Remember to close SQL resources declared while running this function. # Keep any declared in global scope (e.g. mysql_conn) for later reuse. with __get_cursor() as cursor: cursor.execute('SELECT NOW() as now') results = cursor.fetchone() return str(results['now']) 

Same problem, but

1. on App Engine Standard and with PostgreSQL
2. Cloud SQL running within a different project B than the app code that is running in project A.

The underlying issue looks to be very much the same.

Following the tutorial to the dot (again, this is for Postgres, not MySQL), and thus using sample code from here, the solution that made things work almost immediately (no configuration delay noticed):

• Look up the main service account's email address in project A's IAM & Admin section; it should have the format <Your project ID>@appspot.gserviceaccount.com
• In project B, add a line under IAM & Admin with this email address as principal and role Cloud SQL Client

The tutorial talks about

If the authorizing service account belongs to a different project than the Cloud SQL instance, the Cloud SQL Admin API and IAM permissions will need to be added for both projects.

I did add Cloud SQL API and Cloud SQL Admin API to both projects, but did NOT give the service account in project A any Cloud SQL-related permissions -- if that's what the tutorial is talking about.

The above implies that if both Cloud SQL and the app code run within the same project, adding the Cloud SQL Client role to the project's service account should be all that's needed. (Role Cloud SQL Admin, as @Zou Dino suggests isn't required, but should work, too.)