Configuring external services

Server v4.4

Server Admin

PostgreSQL

Best practices for PostgreSQL

Your externalized PostgreSQL instance needs to be version 12.1 or higher.

Consider running at least two PostgreSQL replicas to allow recovery from primary failure and for backups. The table below shows the recommended specifications for PostgreSQL machines:

# of Daily Active Users	# of PostgreSQL Replicas	CPU	RAM	Disk	NIC Speed
<50	2	8 Cores	16 GB	100 GB	1 Gbps
50 - 250	2	8 Cores	16 GB	200 GB	1 Gbps
250 - 1000	3	8 Cores	32 GB	500 GB	10 Gbps
1000 - 5000	3	8 Cores	32 GB	1 TB	10 Gbps
5000+	3	8 Cores	32 GB	1 TB	10 Gbps

Migrating from an internal PostgreSQL to an externalized source

If you are doing a fresh install of CircleCI server, then you can skip this section and head to Connecting your external PostgreSQL instance to CircleCI server

When a CircleCI server instance is deployed, Postgres is deployed internally by default via its helm chart. However, as an operator, you may wish to externalize this database to have better control over scalability and availability. Once you have configured your external Postgres, you may use the guide below to migrate your Postgres data to your external database.

This process requires downtime.

1. Disable the application

Disable the CircleCI server application by scaling down the application layer pods. No Data is lost in this process, but the application will be unreachable.

Scale down your application layer pods:

namespace=<your-server-namespace>
kubectl -n "$namespace" scale deploy -l "layer=application" --replicas="0"

Running kubectl -n "$namespace" get pods will show most of your pods scaling to down, leaving your database pods running including PostgreSQL.

2. Validate access to your external PostgreSQL from within the cluster (optional)

Confirm that pods within your CircleCI server cluster can access your external PostgreSQL. You can do this from within your internal PostgreSQL.
```
PG_POD=$(kubectl -n "$namespace" get pods | grep postgresql | tail -1 | awk '{print $1}')
kubectl exec -it -n "$namespace" "$PG_POD" -- bash
```

While still connected to the pod run:

psql -h <your-external-postgres-host> -U postgres -p <your-external-postgres-port>

You should be able to connect to your external Postgres at this point. If not, resolve any issues before proceeding.

You may use helm upgrade … to restore your CircleCI server instance to a running state.

3. Generate export of your internal PostgreSQL

Retrieve your internal Postgres credentials:
```
PG_PASSWORD=$(kubectl -n "$namespace" get secrets postgresql -o jsonpath="{.data.postgresql-password}" | base64 --decode)
```
The username for your internal Postgres is postgres. The password is randomly generated unless directly set at installation.
Connect to your Postgres pod and perform a Postgres dump:
```
kubectl -n "$namespace" exec -it "$PG_POD" -- bash -c "export PGPASSWORD='$PG_PASSWORD' && pg_dumpall -U postgres -c" > circle.sql
```
This backup is created in the filesystem used by the Postgres pod. If you wish to store it locally, you may use kubectl cp -n "$namespace" "$PG_POD":circle.sql /local/dir/circle.sql
Clean up the Postgres Dump. Your internally deployed Postgres uses the username postgres. However, during the restore, the Postgres dump will drop all resources before trying to create new ones, including the postgres user. Access the Postgres pod where the dump is stored and run the following commands on the Postgres dump file to remove the lines that would delete the Postgres user.
```
PG_POD=$(kubectl -n "$namespace" get pods | grep postgresql | tail -1 | awk '{print $1}')
kubectl exec -it -n "$namespace" "$PG_POD" -- bash

sed -i".bak" '/DROP ROLE postgres/d' circle.sql
sed -i".bak" '/CREATE ROLE postgres/d' circle.sql
sed -i".bak" '/ALTER ROLE postgres WITH SUPERUSER INHERIT CREATEROLE CREATEDB LOGIN REPLICATION BYPASSRLS PASSWORD/d' circle.sql
```

4. Restore your data in your external PostgreSQL

While still connected to your the internally deployed Postgres, restore the dumped data to your external Postgres:

psql -h <your-external-postgres-host> -U postgres -p <your-external-postgres-port> < circle.sql

Now your external Postgres will have your CircleCI server data. In the next section you will update CircleCI server to point to your external Postgres.

Connecting your external PostgreSQL instance to CircleCI server

Once you have set up your external PostgreSQL instance, add the following to your values.yaml file so that your CircleCI server instance can access it.

postgresql:
  internal: false
  postgresqlHost: <domain> # The domain or IP address of your PostgreSQL instance
  postgresqlPort: <port> # The port of your PostgreSQL instance

postgresql.internal: false will remove any previously deployed PostgreSQL instance deployed internally

The changes will take effect upon running helm install/upgrade. If you are completing a migration to an externalized PostgreSQL instance then when you perform helm upgrade, the scaled down pods will be scaled back to their replica numbers as defined by your values.yaml.

Back up PostgreSQL

PostgreSQL provides official documentation for backing up and restoring your PostgreSQL 12 install, which can be found here.

We strongly recommend the following:

Take daily backups
Keep at least 30 days of backups
Use encrypted storage for backups as databases might contain sensitive information
Perform a backup before each upgrade of CircleCI server

MongoDB

If using your own MongoDB instance, it needs to be version 3.6 or higher.

Migrating from an internal MongoDB to an externalized source

If you are doing a fresh install of CircleCI server, then you can skip this section and head to Connecting your external MongoDB instance to CircleCI server

When a CircleCI server instance deployed, MongoDB is deployed internally by default via its helm chart. However, as an operator, you may wish to externalize this database to have better control over scalability and availability. Once you have configured your external MongoDB, you may use the guide below to migrate your Mongo data to your external database.

This process requires downtime.

1. Disable the application

Disable the CircleCI server application by scaling down the application layer pods. No Data is lost in this process, but the application will be unreachable.

Scale down your application layer pods:

namespace=<your-server-namespace>
kubectl -n "$namespace" scale deploy -l "layer=application" --replicas="0"

Running kubectl -n "$namespace" get pods will show most of your pods scaling to down, leaving your database pods running, including Mongo.

2. Validate access to your external MongoDB from within the cluster (optional)

Confirm that pods within your CircleCI server cluster can access your external MongoDB. You can do this from within your internal MongoDB pod:
```
MONGO_POD="mongodb-0"
kubectl exec -it -n "$namespace" "$MONGO_POD" -- bash
```

While still connected to the pod run the following:

mongo --username <username> --password --authenticationDatabase admin --host <external-mongodb-host> --port <external-mongodb-port>

You should be able to connect to your external MongoDB at this point. If not, resolve any issues before proceeding.

You may use helm upgrade … to restore your CircleCI server instance to a running state.

3. Generate export of your internal MongoDB

Retrieve your internal MongoDB credentials:

MONGO_POD="mongodb-0"
MONGODB_USERNAME="root"
MONGODB_PASSWORD=$(kubectl -n "$namespace" get secrets mongodb -o jsonpath="{.data.mongodb-root-password}" | base64 --decode)

Create a backup directory in your MongoDB pod:

kubectl -n "$namespace" exec "$MONGO_POD" -- mkdir -p /tmp/backups/

Generate a MongoDB database dump to the backup directory you just created:

kubectl -n "$namespace" exec -it "$MONGO_POD" -- bash -c "mongodump -u '$MONGODB_USERNAME' -p '$MONGODB_PASSWORD' --authenticationDatabase admin --db=circle_ghe --out=/tmp/backups/"

4. Restore your data in your external MongoDB

Use the generated MongoDB backup to restore the data to your external MongoDB:

kubectl -n "$namespace" exec "$MONGO_POD" -- mongorestore --drop -u "$MONGODB_USERNAME" -p "$MONGODB_PASSWORD" --host <external-mongodb-host> --port <external-mongodb-port> --authenticationDatabase admin /tmp/backups/circle_ghe;

Now your external MongoDB will have your CircleCI server data. In the next section you will update CircleCI server to point to your external MongoDB.

Connecting your external MongoDB instance to CircleCI server

Once you have configured your external MongoDB instance, add the following to your values.yaml file to connect your CircleCI server instance.

mongodb:
  internal: false
  hosts: <hostname:port> # this can be a comma-separated list of multiple hosts for sharded instances
  ssl: <ssl-enabled>
  # If using an SSL connection with custom CA or self-signed certs, set this
  # to true
  tlsInsecure: false
  # Any other options you'd like to append to the MongoDB connection string.
  # Format as query string (key=value pairs, separated by &, special characters
  # need to be URL encoded)
  options: <additional-options>
  auth:
    database: <authentication-source-database
    mechanism: SCRAM-SHA-1

The changes will take effect upon running helm install/upgrade. If you are completing a migration to an externalized MongoDB instance then when you perform helm upgrade, the scaled down pods will be scaled back to their replica numbers as defined by your values.yaml.