Skip to content

Commit

Permalink
docs: addresses comments on HA article
Browse files Browse the repository at this point in the history
Signed-off-by: mbshields <[email protected]>
  • Loading branch information
mbshields committed Jun 24, 2024
1 parent 77cd805 commit fbed6c8
Showing 1 changed file with 16 additions and 15 deletions.
31 changes: 16 additions & 15 deletions docs/articles/high-availability.md
Original file line number Diff line number Diff line change
Expand Up @@ -6,27 +6,17 @@ In the zot configuration, the `sync` extension allows a zot instance to mirror a

Two failover configurations are possible:

* Active/active

Registry requests are load-balanced between two zot instances, each of which mirrors the other.

* Active/standby

Registry requests are sent by the load balancer to the active zot instance, while a standby instance mirrors the active. If the load balancer detects a failure of the active instance, it then sends requests to the standby instance.

> :pencil2: The highly available zot registry described in this article differs from [zot clustering](clustering.md). Although zot clustering provides a level of high availability, the instances share common storage, whose failure would affect all instances. In the method described in this article, each instance has its own storage, providing an additional level of safety.
For details of configuring the `sync` extension, see [OCI Registry Mirroring With zot](mirroring.md).

## Configuring an active/active registry
* Active/active

An active/active zot registry can be implemented between two zot instances by configuring the `sync` extension in each instance to point to the other instance. In this scheme:
Registry requests are load-balanced between two zot instances, each of which mirrors the other.

* a load balancer such as HAProxy or a [DNS-based routing](https://coredns.io/plugins/loadbalance/) scheme is deployed for round-robin load balancing between zot instances
* each zot instance is configured as a standalone registry with its own storage
* each zot instance has its `sync` extension enabled to periodically synchronize with the other instance
> :pencil2: The highly available zot registry described in this article differs from [zot clustering](clustering.md). Although zot clustering provides a level of high availability, the instances share common storage, whose failure would affect all instances. In the method described in this article, each instance has its own storage, providing an additional level of safety.
With periodic synchronization, a window of failure exists between synchronization actions. For example, if an image is posted to instance A soon after instance B has synchronized with instance A, and then instance A fails, instance B will not have the new image. To minimize this exposure, we recommend keeping the synchronization period as small as practical.
For details of configuring the `sync` extension, see [OCI Registry Mirroring With zot](mirroring.md).


## Configuring an active/standby registry
Expand All @@ -37,4 +27,15 @@ An active/standby zot registry can be implemented between two zot instances by c
* each zot instance is configured as a standalone registry with its own storage
* the standby zot instance has its `sync` extension enabled to periodically synchronize with (mirror) the active instance

With periodic synchronization, this scheme is also susceptible to the window of failure described in the active/active case. To minimize this exposure, we recommend keeping the synchronization period as small as practical.
With periodic synchronization, a window of failure exists between synchronization actions. For example, if an image is posted to the active instance soon after the standby has synchronized with the active, and then the active fails, the standby will not have the new image. To minimize this exposure, we recommend keeping the synchronization period as small as practical.

## Configuring an active/active registry

An active/active zot registry can be implemented between two zot instances by configuring the `sync` extension in each instance to point to the other instance. In this scheme:

* a load balancer such as HAProxy or a [DNS-based routing](https://coredns.io/plugins/loadbalance/) scheme is deployed for load balancing between zot instances
* [path-based routing](https://www.haproxy.com/blog/path-based-routing-with-haproxy) must be implemented
* each zot instance is configured as a standalone registry with its own storage
* each zot instance has its `sync` extension enabled to periodically synchronize with the other instance

With periodic synchronization, a window of failure exists between synchronization actions. For example, if an image is posted to instance A soon after instance B has synchronized with instance A, and then instance A fails, instance B will not have the new image. To minimize this exposure, we recommend keeping the synchronization period as small as practical.

0 comments on commit fbed6c8

Please sign in to comment.