Backup & Recovery

The current architecture of NCP on-premise backend involves database engines. One of the engines, TimescaleDB supports storage and access of time series telemetry data while the other, PostgreSQL, supports the same for aggregated CRUD data. The current deployment form factor of these engines is a single server.

In this document, we will first present a few unavailability scenarios relevant to the NCP application. For each unavailability scenario, we will highlight the impact in terms of mean time between occurrences, mean time to recovery, and last but not least, whether the scenario would require a data recovery/migration aspect. Finally, we will propose a standard set of solutions that are recommended by both TimescaleDB and Postgres to handle data and service recovery in such situations.

Scenarios Considered

Scenario	Impact	Mean Time Between Occurrences	Mean Time of Unavailability	Mean Time to Recover	Service Loss Impact	Data Loss Impact	Need for Data Recovery/Migration
NCP Application Upgrade	Low	Occasional, between releases (major as well as patch)	Low	Not applicable	Low	No	Not required
NCP TimescaleDB and Postgres Upgrade	Low	Occasional, when application needs new database features to be enabled	Low	Not applicable	Low	No	Not required
Application Crashes	Low	Occasional	Mostly low	Mostly low	Low	No	Not required
Database Instance Crashes (Recoverable)	Medium	Proven COTS components, very infrequent	Low	Low	Medium	No	Not required
Database Instance Crashes (Irrecoverable mainly due to data corruption)	High (depends on existing data volume)	Proven COTS components, very infrequent	Very High	Very High	High	No	Required within the same server
Media Failure (Recoverable)	High (depends on existing data volume)	Very infrequent	High	High	High	No	No
Media Failure (Irrecoverable)	High (depends on existing data volume)	Very infrequent	Very High	Very High	High	Yes	Required across different servers
Data Center Disaster	High (depends on existing data volume)	Extremely Infrequent	Extremely High	Extremely High	Extremely High	Extremely High	Cross location
Data Migration due to DB product replacement	High (depends on existing data volume)	Extremely Infrequent	Very High	Very High	High	Yes	Required

From the above table, it is evident that even though the scenarios that would require data migration/recovery are infrequent, the availability impact is high. The rest of the document presents a set of solutions, pros and cons of each of them, assumptions, and end-user side requirements

NCP Solution

NCP provide DB backup service and performs periodic backups to the remote NFS backup server endpoint provided by the customer. In disaster scenarios, the customers can use our SRE teams to recover the data from backups.

In future, NCP will implement the backup and restore using a DB backup service which performs a transparent migration from single server Postgres/Timescale to distributed Timescale handling both fresh installs and upgrades.