fix(ddb): use two instances, select primary as read instance

We previously went to one instance thinking that there was a problem
with stale reads. This turned out to be CloudFront caching api traffic.

With a single instance and no secondaries, mongoose tries forever to
reach the secondary, in vain.

This should be solved by using `readPreference=primary` in the conn
string, but to allow for longevity of the setup, we will bump ddb back
to 2 instances.

We will also specify a read preference of `primaryPreferred`, to
better assure some consistency

lib/formsg-on-cdk-stack.ts

@@ -108,7 +108,7 @@ export class FormsgOnCdkStack extends cdk.Stack {
       // providing 750 machine hours free per month
       // See https://aws.amazon.com/documentdb/free-trial/
       instanceType: ec2.InstanceType.of(ec2.InstanceClass.BURSTABLE3, ec2.InstanceSize.MEDIUM),
-      instances: 1,
+      instances: 2,
       engineVersion: '4.0',
       parameterGroup: new cdk.aws_docdb.ClusterParameterGroup(this, 'ddb-parameter-group', {
         dbClusterParameterGroupName: 'disabled-tls-parameter2',
@@ -126,7 +126,7 @@ export class FormsgOnCdkStack extends cdk.Stack {
         secretName: 'ddb-connstring',
         removalPolicy: cdk.RemovalPolicy.DESTROY,
         secretStringValue: cdk.SecretValue.unsafePlainText(
-          `mongodb://root:${ddbPassSecret.secretValue.unsafeUnwrap()}@${db.clusterEndpoint.socketAddress}/form?replicaSet=rs0&retryWrites=false`
+          `mongodb://root:${ddbPassSecret.secretValue.unsafeUnwrap()}@${db.clusterEndpoint.socketAddress}/form?replicaSet=rs0&readPreference=primaryPreferred&retryWrites=false`
         ),
       })
     )