Clone and redeploy an AWS RDS instance to propagate a production dataset to downstream environments like staging or demo. This allows staging to maintain data parity with production in a throw-away environment.
Suppose for testing or sales purposes it is necessary to maintain an independent application stack with it's own database, which needs to periodically refresh data from the production database. To simplify the illustration, this will focus on the changes to the AWS RDS database replication graphs, and omit the application and other services it may depend on.
Consider two independent application stacks, production and demo, with primary databases mitosis-prod
and mitosis-demo
respectively. Each stack has a replication graph where a primary database is followed by one replica, ie mitosis-prod
replicates to mitosis-prod-replica
and mitosis-demo
replicates to mitosis-demo-replica
.
Stack mitosis will clone mitosis-prod
into a new replica temp-mitosis-demo
. The newly created replica will have the same data as mitosis-prod
, but will copy instance attributes like tags, VPC security groups, and db parameter groups from the original mitosis-demo
.
It then promotes temp-mitosis-demo
to disconnect from the replication graph of mitosis-prod
, and enables backups so it can act as a replication source.
Once temp-mitosis-demo
is an independent replication graph, create a new replica of it called temp-mitosis-demo-replica
, which will have identical data to temp-mitosis-demo
, but instance attributes copied from mitosis-demo-replica
.
It will then rename the existing mitosis-demo
graph to prefix with old-
.
Once that is complete, it's safe to rename the temp-
prefixed clones back to mitosis-demo
and mitosis-staging-demo
.
However, as this is a DNS swap, the application is likely still connected to the original old-mitosis-demo
. By specifying a restart script, stack-mitosis can force the demo application to restart, and connect to the newly created mitosis-demo
with fresh data from production. Once it has restarted the application successfully, it deletes the old-
prefixed database instances from the original demo replication graph.
Note that this replication graph is a simple case, but it supports replacing arbitrarily complex replication graphs on RDS and has been verified with mysql and postgres database engines. The postgres engine on RDS only allows multiple replicas of a single primary, but the Mysql engine on RDS allows cascading replicas of replicas. See the AWS documentation for working with RDS read replicas for more information on these limitations.
In the case where source and target instances live in different VPCs, or in case --restore-snapshot
is used, the first instance (temp-mitosis-demo
here) is created by restoring the latest available snapshot, instead of using replication. This also skips the promote replica step. All other steps remain the same.
After installing a JDK, follow the clojure install
instructions for your environment
to ensure clj
and clojure
are in path.
Stack mitosis uses aws-api to interact with AWS. That uses the same credentials preferences as the AWS Java client. However, stack-mitosis also supports using STS and prompting for a MFA token to authorize use of a limited role for the duration of the operation.
In order to support that the following environment variables need to be present;
AWS_CONFIG_FILE
, AWS_CREDENTIAL_PROFILES_FILE
should be set to specify any
credentials other than .aws/credentials
. For the initial handshake with AWS,
credentials are needed from those files, or from environment variables,
AWS_ACCESS_KEY_ID
, AWS_SECRET_ACCESS_KEY
, and AWS_REGION
. In order to
select an assumed role using an MFA token, for now a role.edn
file should be
specified with the following values:
{:mfa-serial "arn:aws:iam::1234:mfa/username"
:role-arn "arn:aws:iam::1234:role/sudo"
:region "us-west-1"}
Hopefully in the future this can be parsed directly from the AWS_CONFIG
file.
clj -m stack-mitosis.cli \
--source mitosis-production --target mitosis-staging \
--restart "./restart-service.sh"
[--credentials resources/role.edn]
[--plan]
[--iam-policy]
[--restore-snapshot]
The --plan
flag will give a flight plan showing the expected list of API calls it's planning on executing against the Amazon API.
$ clj -m stack-mitosis.cli --source mitosis-prod --target mitosis-demo --plan
Flight plan:
:CreateDBInstanceReadReplica temp-mitosis-demo
{:Port 5430, :DBInstanceClass "db.t3.micro", :Tags [{:Key "Service", :Value "Mitosis"}], :OptionGroupName "default:mysql-5-7", :SourceDBInstanceIdentifier "mitosis-prod", :StorageType "gp2", :MultiAZ false}
:PromoteReadReplica temp-mitosis-demo
{}
:ModifyDBInstance temp-mitosis-demo
{:ApplyImmediately true, :PreferredMaintenanceWindow "tue:07:02-tue:08:00", :PreferredBackupWindow "05:30-06:20", :DBParameterGroupName "default.mysql5.7", :BackupRetentionPeriod 1}
:CreateDBInstanceReadReplica temp-mitosis-demo-replica
{:Port 5431, :DBInstanceClass "db.t3.micro", :Tags [{:Key "Service", :Value "Mitosis"} {:Key "IsReplica", :Value "true"}], :OptionGroupName "default:mysql-5-7", :SourceDBInstanceIdentifier "temp-mitosis-demo", :StorageType "gp2", :MultiAZ false}
:ModifyDBInstance temp-mitosis-demo-replica
{:ApplyImmediately true, :PreferredMaintenanceWindow "mon:07:30-mon:08:00", :PreferredBackupWindow "06:40-07:10", :DBParameterGroupName "default.mysql5.7"}
:ModifyDBInstance mitosis-demo-replica
{:ApplyImmediately true, :NewDBInstanceIdentifier "old-mitosis-demo-replica"}
:ModifyDBInstance mitosis-demo
{:ApplyImmediately true, :NewDBInstanceIdentifier "old-mitosis-demo"}
:ModifyDBInstance temp-mitosis-demo-replica
{:ApplyImmediately true, :NewDBInstanceIdentifier "mitosis-demo-replica"}
:ModifyDBInstance temp-mitosis-demo
{:ApplyImmediately true, :NewDBInstanceIdentifier "mitosis-demo"}
:DeleteDBInstance old-mitosis-demo-replica
{:SkipFinalSnapshot true}
:DeleteDBInstance old-mitosis-demo
{:SkipFinalSnapshot true}
Note that for many cases, even if the clone process is interrupted, the flight plan will show steps it will try to execute again, and steps it will skip because it has detected that the instance has already been created or modified to the right attribute values. In other words, it tries to pickup where it left-off if there is a failure.
Stack-mitosis can also generate an IAM policy for an automated user to update a particular environment. The policy uses the database names from the planned changeset to calculate these minimal permissions. While they have been elided from the example below, the ARNs are locked to the specific account & region used.
$ clj -m stack-mitosis.cli --source mitosis-prod --target mitosis-demo --iam-policy
{"Version":"2012-10-17",
"Statement":
[{"Effect":"Allow",
"Action":["rds:DescribeDBInstances", "rds:ListTagsForResource"],
"Resource":["arn:aws:rds:*:*:db:*"]},
{"Effect":"Allow",
"Action":["rds:CreateDBInstanceReadReplica"],
"Resource":
["arn:aws:rds:*:*:og:*", "arn:aws:rds:*:*:pg:*",
"arn:aws:rds:*:*:subgrp:*",
"arn:aws:rds:*:*:db:mitosis-prod",
"arn:aws:rds:*:*:db:temp-mitosis-demo",
"arn:aws:rds:*:*:db:temp-mitosis-demo-replica"]},
{"Effect":"Allow",
"Action":["rds:AddTagsToResource"],
"Resource":
["arn:aws:rds:*:*:db:temp-mitosis-demo",
"arn:aws:rds:*:*:db:temp-mitosis-demo-replica"]},
{"Effect":"Allow",
"Action":["rds:PromoteReadReplica"],
"Resource":
["arn:aws:rds:*:*:db:temp-mitosis-demo"]},
{"Effect":"Allow",
"Action":["rds:ModifyDBInstance"],
"Resource":
["arn:aws:rds:*:*:og:*", "arn:aws:rds:*:*:pg:*",
"arn:aws:rds:*:*:secgrp:*", "arn:aws:rds:*:*:subgrp:*",
"arn:aws:rds:*:*:db:temp-mitosis-demo",
"arn:aws:rds:*:*:db:temp-mitosis-demo-replica",
"arn:aws:rds:*:*:db:mitosis-demo-replica",
"arn:aws:rds:*:*:db:old-mitosis-demo-replica",
"arn:aws:rds:*:*:db:mitosis-demo",
"arn:aws:rds:*:*:db:old-mitosis-demo"]},
{"Effect":"Allow",
"Action":["rds:RebootDBInstance"],
"Resource":
["arn:aws:rds:*:*:db:temp-mitosis-demo",
"arn:aws:rds:*:*:db:temp-mitosis-demo-replica",
"arn:aws:rds:*:*:db:mitosis-demo-replica",
"arn:aws:rds:*:*:db:mitosis-demo"]},
{"Effect":"Allow",
"Action":["rds:DeleteDBInstance"],
"Resource":
["arn:aws:rds:*:*:db:old-mitosis-demo-replica",
"arn:aws:rds:*:*:db:old-mitosis-demo"]}]}
This ensures that a continuous integration or cronjob server like Jenkins can clone production to demo environments on a weekly basis restricted to the minimal permissions necessary. If a user needs to run stack-mitosis for multiple environments (demo, staging, random developer test environment), then a policy can be attached for each environment.
bin/kaocha # basic unit tests
bin/kaocha --plugin cloverage # with coverage output
Cloudformation and Terraform are wonderful tools focused on declarative architecture transformation from one steady state to another. Stack-mitosis is focused on safely cloning the contents of a database in one environment to another without changing from one steady state to another. As example, for an environment with production and demo environments, they both exist in the correct configuration before running stack-mitosis, and then after running stack-mitosis the configuration remains the same but the demo environment has a fresh copy of the data from production.
I suspect this could also be accomplished using one of these declarative infrastructure tools by transitioning through multiple intervening states, but have not found any examples of anyone doing that.
Copyright © 2019-2021 Charles L.G. Comstock
Distributed under the BSD-3 Clause License (see LICENSE file)