Netbox HA

[vladsker]

Hello, folks,

I'm relying on Netbox for my automation, and different scripts and robots are using it more and more frequently.

So, the natural question came - how are you managing highly available installations? It does not look like there is much information on this matter.

Thanks for creating such a great product!

Any advice is much appreciated!

Kind regards,
Vlad

[candlerb]

80% of making Netbox HA is making Postgres HA. Once you've done that, you can have multiple frontends all pointing to the same Postgres database or cluster, stick them behind a load-balancer, and they'll happily share the load or take over from each other.

10% is making Redis HA - since multiple netbox and netbox-rq workers will need to point to the same Redis instance.

The remaining 10% is sharing the media directory (/opt/netbox/netbox/media), and the reports and scripts directories. The media directory can be shared using NFS (in which case you will need a HA NFS server), or you can configure Netbox to store its media on S3 instead. The reports and scripts directories can be periodically rsync'd from a master location, or pushed out using your configuration management system.

If you just want disaster recovery rather than HA, then it's somewhat simpler. Configure your master postgres server to replicate to a backup postgres server in another data centre, and sync the media/reports/scripts directories. When you need to cutover in disaster, then promote the replica database to master. You may lose any in-flight background tasks or webhook events in Redis; if that's important you could configure Redis leader-follower replication too. However to be honest, if the whole primary data centre goes down, and you're only doing asynchronous replication between sites, you may lose the most recently-written data anyway.

Alternatively you could outsource this, and use one of the cloud-managed Postgres services (e.g. AWS RDS) and Redis services (e.g. AWS MemoryDB). Soon you'll be able to outsource your entire Netbox instance to NS1 :-)

[robertlynch3]

I was looking through this and I don't mind an active/backup setup, so I am really only looking to get Postgres in HA mode. That being said, how does this work with upgrading the two instances. I am sure instance 1 will upgrade no problem, but what about instance two?

[candlerb]

As long as you don't upgrade both instances at once, you should be fine. The second one will notice that the database migrations have already been applied, and not apply them again.

Alternatively, you could modify the ./upgrade.sh script on the second host to comment out those steps which touch the database.

If the upgrade is going to take a while, then make sure you shut down the second instance before you start - because you don't want to be running an old version of Netbox which is talking to a database which has been migrated to a newer schema.

[robertlynch3]

Awesome, thanks for this info!

[robertlynch3]

For any future readers. If you configure Netbox with Postgres in a hot/standby config, you need to add the following config so that the backup system will allow user sign in. (credits to #3118 and #3196 )

In netbox/netbox/netbox/configuration.py

MAINTENANCE_MODE=True # this allows the server to connect to the local read-only db
SESSION_FILE_PATH="/opt/netbox/netbox/sessionFile" # this will have netbox write sessions to local storage instead of the database. you will need to run `mkdir /opt/netbox/netbox/sessionFile` which needs to be owned by the netbox user

In netbox/netbox/netbox/ldap_config.py
And if you have LDAP configured, you will need to add

AUTH_LDAP_ALWAYS_UPDATE_USER = False #stops ldap from updating the database, needed for readonly databases
``

[chicks-net]

Is this config only for the backup host and not the primary host? If so, does anything special need to be configured on the primary host?

If /opt/netbox/netbox/sessionFile is a directory it might be less confusing to name it /opt/netbox/netbox/sessions or something else that leaves the File out of a directory name.

[candlerb]

Is this config only for the backup host and not the primary host?

He is describing a situation where:

• There is postgres-level replication from a primary database to a secondary database (and therefore the secondary database is read-only)
• The primary host points at the primary database
• The secondary host points at the secondary database
• You want to be able to login to the secondary host (e.g. for testing DR)

In that situation, you make those special config changes to the secondary host only. In the event of a disaster, you'd promote the secondary database to primary, and turn off maintenance mode.

I'd suggest a more conventional approach would be to point the secondary Netbox host at the primary database. In this case, active/active works fine. In the event of a disaster, you'd repoint the secondary host to the secondary database, and promote the secondary database to primary.

[robertlynch3]

If /opt/netbox/netbox/sessionFile is a directory it might be less confusing to name it /opt/netbox/netbox/sessions or something else that leaves the File out of a directory name.

Yes you can call it whatever you'd like. You need to make sure that it is copied between updates or it will complain.

He is describing a situation where:

• There is postgres-level replication from a primary database to a secondary database (and therefore the secondary database is read-only)
• The primary host points at the primary database
• The secondary host points at the secondary database
• You want to be able to login to the secondary host (e.g. for testing DR)

Sorry I didn't make this clearer before. My use case is a failure at the main site (where Netbox is located) and we need Netbox up to bring said site back online. In my case, we are only going to use the backup image for that one situation, all day to day interactions with Netbox will be with the primary instance.

I didn't want to overcomplicate it as my use case is strictly for disaster recovery, but mainly that critical time to bring the main core infrastructure up.