Container volume snapshot backup and restore

[joshenders]

• Backup server state (/srv/minecraft/world) every N minutes (5min?) to ... S3? Ensure data is safe to snapshot via rcon?
• On-startup, if backup exists, restore backup instead of re-initializing. Offer forced re-initialization via env variable
• Tickrate FF on resume

[joshenders]

After some research in EFS/EBS/S3. EBS volumes seem like the right approach here. They support snapshots (aiming to do like 6-12/hour?) and the snapshots themselves can be backed up to S3, for portability. S3 is a fall back but EFS doesn't seem like a great fit.

[joshenders]

• Backup server state (/srv/minecraft/world) every N minutes (5min?) to ... S3? Ensure data is safe to snapshot via rcon?

/auto-save's every 5 mins by default to EFS. Want durable S3 backups after every session. Wrote lambdas to idle check and safe shutdown but now thinking this may be bettered handled as a healthcheck script since there's a timelimit on lambdas. If so, may use micropython and rewrite entrypoint.sh in python.

• On-startup, if backup exists, restore backup instead of re-initializing. Offer forced re-initialization via env variable

Done. ENV variable is FORCE_CREATE_NEW_WORLD

• Tickrate FF on resume

Not possible without a mod. Skip for now.

[joshenders]

Need to solve several problems here:

1. Detect when server is idle. (solution?: lambda supervisor)
2. If server is idle, save, stop server, compress backup and upload /data to S3. (Solution: Can be done in entrypoint script.)

• Implement process supervisor as a lambda function;
• Implement process supervisor (or components of it) within docker healthcheck; drawback: 60s max execution time/bad separation of duties/design.
• Implement process supervisor as another process within container; drawback: multi-process container is bad design
• Implement process supervisor as a sidecar. Non-starter; more added cost.

Implement process supervisor as a lambda function

Pros

• Code is already implemented.

Cons

• Undecided how to trigger an EFS backup to S3, lambda can run for 15m but usage may exceed free tier for large backups. Can be added to entrypoint script
• Code has to be uploaded to S3 separately, not the end of the world but one more moving part
• If the container crashes, there doesn't seem to be a good event that is emitted? (Maybe there is and I just need to research this more?) (TODO: tmp security rule cleanup within supervisor controller)