Elastic media server
All the replicas for the media server are always up and running which incurs unnecessary cost to customers. The basic media server pod power management (Elastic media server) feature provides Auto scaling of media server replicas based on the CPU and memory usage to reduce the cost.
For enabling/disabling the auto scaling feature, following media server CR inputs are required:
: Describes the maximum number of replicas that the media server can scale up to.
: Describes the minimum number of replicas of the media server running. This is an optional field. If not specified, the value for field will be set to the default value of 1.
To enable the elasticity of media server, the value of must be more than value of .
To disable the autoscaling feature of media server, ensure that the value of is equal to the value of .
Note:
The values of and must be greater than 0 to enable the elasticity of media server.
Following table describes the ElasticityAttributes that describes the attributes associated with the media server autoscaler. These attributes are only applicable if autoscaler is running.
Fields | Description |
|---|---|
ExpectedReplicas | Indicates the ideal number of replicas computed by media server autoscaler that must be running. Default value is 0. It will be 0 if media server autoscaler is disabled. Note: If autoscaler is enabled (after that autscaler is tuned off) the value would be set to the value of . It will be even if media server autoscaler is disabled. |
ActiveReplicas | Indicates the actual number of replicas that must be running to complete the ongoing operations on the media servers. Default value is 0. It will be 0 if media server autoscaler is disabled. Note: If autoscaler is enabled (after that autscaler is tuned off) the value would be set to the value of . It will be even if media server autoscaler is disabled. |
NextIterationTime | Indicates the next iteration time of the media server autoscaler that is, the media server autoscaler will run after NextIterationTime only. Default value is empty. |
NextCertificateRenewalTime | Next time to scale up all registered media servers for certificate renewal. |
ConfigMap
A new ConfigMap with name
nbu-media-autoscaler-configmapis created during deployment and the key-value pairs would be consumed for tuning the media server autoscaler. This ConfigMap is common to all the media server CR objects and supports the following keys:Parameters
Description
memory-low-watermark-in-percent
Low watermark for memory usage.
memory-high-watermark-in-percent
High watermark for memory usage.
cpu-low-watermark-in-percent
Low watermark for CPU usage.
cpu-high-watermark-in-percent
High watermark for CPU usage.
scaling-interval-in-seconds
Interval after which media server autoscaler should run.
stabilitywindow-time-in-seconds
CPU and memory usage is calculated between two time intervals. This key indicates the time interval to be considered for collecting usage.
stability-count
CPU and memory usages are calculated by averaging out on multiple readings. This key indicates the number of readings to be considered.
graceful-shutdown-interval-in-seconds
The time interval after which the media server autoscaler should run incase it is not able to scale in due to running jobs on media server pods.
delayed-scalein-notifications-interval-in-minutes
The time interval between two successive notifications in the event that a scale in does not occur.
Media server scaling
Parameters
Description
Scale-out
If the CPU or memory consumption is above the specified value provided in configMap but if any existing media server is idle that is, no jobs are running on it, then scale-out will not be performed. If all the existing media servers which are ready have jobs running on them, media server autoscaler will scale out a media server pod.
Scale-in
If the CPU and memory consumption is below the specified values provided in configMap, media server autoscaler will scale in the media server pods. Ensure that the running jobs are completed.
Note:
Media server autoscaler scales out single pod at a time but can scale in multiple pods at a time.
Note:
If the scale-in does not happen due to background processes running on the media server, a notification would be sent on NetBackup Web UI after regular time interval as configured in the autoscaler ConfigMap. For more details, see the following section:
The time taken for media server scale depends on the value of configuration parameter. During this interval, the jobs would be served by existing media server replicas based on NetBackup throttling parameters. For example, Maximum concurrent jobs in storage unit, Number of jobs per client, and so on.
Cluster's native autoscaler takes some time as per attribute, which decides on the time a node should be unneeded before it is eligible to be scaled down. By default this is 10 minutes. To change this parameter, edit the cluster-autoscaler's current deployment settings using the following commands and then edit the existing value:
AKS: az aks update -g $RESOURCE_GROUP_NAME -n
EKS: kubectl -n kube-system edit deployment cluster-autoscaler
Note the following:
For scaled in media servers, certain resources and configurations are retained to avoid reconfiguration during subsequent scale out.
Kubernetes services, persistent volume claims and persistent volumes are not deleted for scaled in media servers.
Host entries for scaled in media servers are not removed from NetBackup primary server. Hence scaled in media server entries will be displayed on Web UI / API.
For scaled down media servers, the deleted media servers are also displayed on Web UI/API during the credential validation for database servers.
In order to ensure that the certificates are renewed on the media server replicas for which the pods were scaled in by media server autoscaler for longer duration, the NetBackup operator would scale out all the media server replicas once in a month for which certificate was issued. The scaled-out media server replicas would be then scaled in as per media server autoscaler.
More Information