Use MongoDB Atlas in production

[johnml1135]

We should use it. Let's hope it's under 1k/year.

[ddaspit]

We should investigate AWS DocumentDB as well. It might be cheaper.

[johnml1135]

Sandbox is free, serverless looks like it would be good enough for us and not too much: https://www.mongodb.com/pricing. A brief guess:

• Read: $0.1/million => $0.1/month
• Write: $1/million => $1/month
• Storage: $0.25/GB-month => 1GB -> $0.25/month
• Backup: $0.20/GB-month => 1GB -> $0.20/month
• Data Transfer: $0.01/GB => 10GB -> $0.1/month
• Projected total per month: $2? Also, we can bill through AWS: https://aws.amazon.com/marketplace/pp/prodview-pp445qepfdy34?trk=a0273619-f6ca-40ef-aa7e-571da2f8ab6f&sc_channel=el.

[johnml1135]

We may need to go with the dedicated server, that is, M10 for 0.08$ per hour ~= 700$/year. M10 is the only one with a SLA which is 99.995%, or 131 seconds per month.

[ddaspit]

If we use MongoDB Atlas, we want to run it on AWS.

[johnml1135]

Here is a migration plan:

• Overall plan

  • Put qa, int_qa and docker-compose on personal M0 Mongo Atlas (free teir) - use prefixes to enable the database names to not conflict.
  • Put production on a M10 instance (normal naming)
  • Migrate the data with a series of backup and restore commands, renaming the database names as I go along (where needed).

• Detailed steps to migrate:

  • Draft the serval k8s charts to use Mongo Atlas (specify the correct databases)
  • Using port forwarding, establish a connection to the mongoDB locally - check it with Compass
  • Port Forwarding breaks with k8s for long streams? Whatever the case, instead of doing port forwarding, dump the database and then copy.
  • Uninstall the serval k8s chart (Mongo chart is independent and will still be up)
  • Log onto the terminal on container and mongodump the old mongoDB to tmp/dump.tar.gz
  • tar -czvf dump.tar.gz dump
  • copy the zipped dump locally
  • kubectl -n serval get pods
  • `kubectl cp serval/:/tmp/dump.tar.gz ./dump.tar.gz
  • tar -xf dump.tar.gz
  • Rename the folders under the dump folder to have the correct (renamed) database names
  • Check with Compass that the Atlas DB is ready (and does not have the databases)
  • mognorestore the Atlas DB
  • Check with Compass that the data was sucessfully uploaded
  • Apply the updated serval k8s chart
  • Confirm that the database connections are correct and that there was no data loss (can request engines)

[johnml1135]

We probably can go with Serverless: https://www.mongodb.com/community/forums/t/is-mongodb-serverless-suitable-for-production/184480

[johnml1135]

The mogodump is breaking from an esoteric error when it lasts too long. The issue appears to be here: https://github.com/kubernetes/kubernetes/issues/74551