FAQ on Avi Controller Cluster
How many nodes are supported in an Avi Controller cluster?
Avi Controller clusters may include either 1 (standalone mode) or 3 Controller nodes in a cluster.
How many nodes are needed for the Avi Controller cluster to be operational?
A three-node cluster requires a quorum (majority) of the nodes to be present for the cluster to be operational. This means at least 2 nodes must be up.
Can you explain how the 3 nodes in an Avi Controller cluster are used during normal operation?
Among the 3 nodes, a leader node is elected and orchestrates the process startup or shutdown across all the active members of the cluster. Configuration and metrics databases are configured to be active on the leader node and placed in standby mode on the follower nodes. Streaming replication is configured to synchronize the active database to the standby databases. Analytics functionality is shared among all the nodes of the cluster. A given virtual service streams the logs and metrics to one of the nodes of the cluster.
What happens if a follower node goes down?
This node is removed from the active member list and the work that was performed on this node is re-distributed among the remaining nodes in the Avi Controller cluster.
What happens if the leader node goes down?
One of the follower nodes is promoted as a leader. This triggers a warm restart of the processes among the remaining nodes in the cluster. The warm standby is required to promote the standby configuration and metrics databases from standby to active on the new leader.
Note: During warm restart, the Avi Controller REST API is unavailable for a period of 2-3 minutes.
What happens if two nodes go down?
The entire cluster becomes non-operational until at least one of the nodes is brought up. A quorum (2 out of 3) of the nodes must be up for the cluster to be operational.
Will there be a disruption to traffic during cluster convergence (single or multiple nodes go down and come back)?
During the time the cluster is non-operational, the Avi Service Engines (SEs) continue to forward traffic for the configured virtual services (This is sometimes referred to as “headless mode”). During this time, SEs attempt to buffer client logs, providing sufficient disk space is available. When the Avi Controller cluster is once again operational, the buffered client logs become available for retrieval. Metric data points are not buffered on the SE during headless operation, however incremental metrics such as “bytes sent” continue to be accounted on the SE so that aggregate information remains correct once metric collection by the Controller resumes.
Recover a non-operational cluster where two of the three nodes are permanently down and not recoverable
Please click here.
Recover the system if all 3 Avi Controller nodes are permanently down and not recoverable
Please click here.
Reconfigure the Avi Controller cluster membership
Please click here.
Can the Avi Controller nodes in a cluster have different resource allocations (memory, CPU, and drive space)?
It is recommended to give all 3 nodes within the Avi Controller cluster the same allocations of these resources.
Over the course of time, there may be a need to re-size (change resource allocations of) the Avi Controllers in the cluster, to accommodate growth. During such a time, it is acceptable to change resource allocations on one Avi Controller node at a time. However, it is expected that eventually all Avi Controllers within the cluster will be re-sized to the same allocations.
Can the Avi Controllers in a cluster be in different subnets?
Yes. This configuration is supported and may even be preferred for certain topologies for improved fault tolerance.
However, a limitation with placing the Avi Controllers in separate subnets is that the cluster IP address is not supported. The cluster IP address requires all Avi Controller nodes to be in the same subnet.
Can the cluster leader be changed manually?
No. Currently, this is not a supported operation. The leader can be chosen during initial deployment of the cluster or when recovering a fully down cluster. However, the leader cannot be manually changed while the cluster is operational.
Avi Controllers participate in election of their leader, and automatically decide who the new leader should be in cases of failure.
Is the Avi Controller restart required for the leader change in the cluster set-up?
To restart the Avi Controller for the leader change in a cluster set-up in an optional step, not mandatory. The other option is to stop the process supervisor by running the following bash command.
systemctl stop process-supervisor
What timers are used during cluster failover? Are the cluster failover timers configurable?
- Leader failure: If the leader Avi Controller of the cluster fails, this triggers an internal warm restart of the Avi Controller processes. Typically, this takes around 2-3 minutes after it is detected that the leader has failed.
- Graceful reboot of leader: Failure detection is almost instantaneous in cases of a graceful reboot.
- Ungraceful power-off of leader: In the case of an ungraceful power-off of the leader, it may take up to 30 seconds to detect that the leader has failed.
These timers have been tuned based on testing but are not configurable.
Can configuration changes be made directly on follower Avi Controllers?
Yes. Nearly any configuration changes can be made on any of the Avi Controller nodes.
- Configuration of the cluster itself (node IP addresses and cluster address)
- Upgrade, which can be performed only on the leader node
What is the recommended cluster deployment option for multiple data centers in different regions?
Avi recommends deploying the Avi Controller cluster only within a single region. However, the member nodes can be deployed across multiple availability zones within the region. If there are multiple regions, it is recommended to deploy one Avi Controller cluster per region. (For more information, click here.)
How to import and export Avi metrics database?
Prior to upgrading to version 22.1.2, it is recommended to export the metrics database.
To export the metrics database, copy the
/var/lib/postgresql/10/pg_metrics/ directory directory and store it outside the Controller VM.
In case you need to roll back to a previous release, the exported metrics can be imported. Thus, preventing the loss of data.
To import the metrics database,
- Stop the process supervisor by running the following command first on the follower nodes and then on the leader.
systemctl stop process-supervisor
Move the copied pg_metrics directory to
/var/lib/postgresql/10/pg_metrics/on the leader.
- Run the following to first start the process supervisor on leader and then on follower so that the imported
pg_metricsdata is replicated from leader to the followers.
python3 /opt/avi/scripts/pg_rec_fix_cluster.py --nodelist leader_ip follower_1_ip follower_2_ip
Note: It will take some time for the cluster to become HA_ACTIVE based on the metrics database size.
- Controller Cluster IP
- Deploying an Avi Controller Cluster
- Updating the Configuration Following Avi Controller IP Address Change
- Controller Cluster Deployment Across Two Availability Zones
- Avi Controller Cluster Configuration in AWS
- API - Configuring the Avi Controller Cluster
- Impact of a Controller Failure
- Cluster Configuration with DNS Hostnames