Google Deployment Manager - database creation fails - google-cloud-platform

Question

I'm trying to create a CloudSQL instance and two databases using Google Deployment Manager. I can't get a reliable first-time deployment where both databases create successfully. Instead, each time I run it, one (or both!) fail with the status "FAILED_PRECONDITION", error message "Bad Request" and no further explanation as to which precondition has failed or how to fix it. Has anyone else come across this before or have any clues on how I can find the issue?
The properties {{ SQL_NAME }} etc. are all defined at the top of my jinja template, but I've ommitted them for clarity.
resources:
- name: {{ SQL_NAME }}
type: sqladmin.v1beta4.instance
properties:
backendType: SECOND_GEN
instanceType: CLOUD_SQL_INSTANCE
region: {{ properties["region"] }}
databaseVersion: {{ properties["dbType"] }}
settings:
tier: db-n1-standard-1
dataDiskSizeGb: 10
dataDiskType: PD_SSD
storageAutoResize: true
replicationType: SYNCHRONOUS
locationPreference:
zone: {{ properties['zone']}}
ipConfiguration:
privateNetwork: {{ properties["network"] }}
- name: {{ DB_NAME }}
type: sqladmin.v1beta4.database
properties:
name: db1
instance: $(ref.{{ SQL_NAME }}.name)
charset: utf8
collation: utf8_general_ci
metadata:
dependsOn:
- {{ SQL_NAME }}
- name: {{ DB2_NAME }}
type: sqladmin.v1beta4.database
properties:
name: db2
instance: $(ref.{{ SQL_NAME }}.name)
charset: utf8
metadata:
dependsOn:
- {{ SQL_NAME }}
- name: {{ USER_NAME }}
type: sqladmin.v1beta4.user
properties:
name: dbroot
host: "%"
instance: $(ref.{{ SQL_NAME }}.name)
password: {{ properties['password'] }}
metadata:
dependsOn:
- {{ SQL_NAME }}

Answer 1

So, I found the answer. It turns out Google is even less helpful with their error message than I thought when I came across the issue. What it seems to be (I still have no concrete evidence that this was the precondition, but the below seems to solve it) is that you can't create two databases at the same time on the same CloudSQL instance, and Deployment Manager tries to because they're both only dependant on the CloudSQL instance itself. I solved the issue by adding dependencies on each successive resource:
- name: {{ DB_NAME }}
type: sqladmin.v1beta4.database
properties:
name: db1
instance: $(ref.{{ SQL_NAME }}.name)
charset: utf8
collation: utf8_general_ci
metadata:
dependsOn:
- {{ SQL_NAME }}
- name: {{ DB2_NAME }}
type: sqladmin.v1beta4.database
properties:
name: db2
instance: $(ref.{{ SQL_NAME }}.name)
charset: utf8
metadata:
dependsOn:
- {{ SQL_NAME }}
- {{ DB_NAME }}
- name: {{ USER_NAME }}
type: sqladmin.v1beta4.user
properties:
name: dbroot
host: "%"
instance: $(ref.{{ SQL_NAME }}.name)
password: {{ properties['password'] }}
metadata:
dependsOn:
- {{ SQL_NAME }}
- {{ DB2_NAME }}

Answer 2

BTW, my coworker just discovered today that as of right meow, explicitly setting backendType to SECOND_GEN and databaseVersion to MYSQL_5_6 also yielded a 400 error. You can use this combination in the console. This is a very recent API break. Just a heads up.