Worker Configuration¶

The worker is a kubernetes pod which is responsible for the actual execution of polling, processing trap messages and sending data to Splunk.

Worker types¶

SC4SNMP has two base functionalities: monitoring traps and polling. These operations are handled by 3 types of workers:

trap worker consumes all the trap related tasks produced by the trap pod.
poller worker consumes all the tasks related to polling.
sender worker handles sending data to splunk. You need to always have at least one sender pod running.

Worker configuration file¶

Worker configuration is kept in values.yaml file in the worker section. worker has 3 subsections: poller, sender or trap, that refer to the workers’ types. values.yaml is being used during the installation process for configuring Kubernetes values. The worker default configuration is:

worker:
  # There are 3 types of workers 
  trap:
    # replicaCount: number of trap-worker pods which consumes trap tasks
    replicaCount: 2
    #autoscaling: use it instead of replicaCount in order to make pods scalable by itself
    #autoscaling:
    #  enabled: true
    #  minReplicas: 2
    #  maxReplicas: 10
    #  targetCPUUtilizationPercentage: 80
  poller:
    # replicaCount: number of poller-worker pods which consumes polling tasks
    replicaCount: 2
    #autoscaling: use it instead of replicaCount in order to make pods scalable by itself
    #autoscaling:
    #  enabled: true
    #  minReplicas: 2
    #  maxReplicas: 10
    #  targetCPUUtilizationPercentage: 80
  sender:
    # replicaCount: number of sender-worker pods which consumes sending tasks
    replicaCount: 1
    # autoscaling: use it instead of replicaCount in order to make pods scalable by itself
    #autoscaling:
    #  enabled: true
    #  minReplicas: 2
    #  maxReplicas: 10
    #  targetCPUUtilizationPercentage: 80
  # udpConnectionTimeout: timeout in seconds for SNMP operations
  #udpConnectionTimeout: 5
  logLevel: "INFO"

All parameters are described in Worker parameters section.

Worker scaling¶

You can adjust number of worker pods to your needs in two ways: setting fixed value in replicaCount or enabling autoscaling which scales pods automatically.

Real life scenario: I use SC4SNMP for only trap monitoring, I want to use my resources effectively¶

If you don’t use polling at all, would be the best to set worker.poller.replicaCount to 0. Remember, that if you’ll want to use polling in the future you need to increase replicaCount, otherwise it won’t work. To monitor traps, adjust worker.trap.replicaCount depending on your needs and worker.sender.replicaCount to send traps to splunk. Usually you need much less sender pods than trap ones.

This is the example of values.yaml without using autoscaling:

worker:
  trap:
    replicaCount: 4
  sender:
    replicaCount: 1
  poller:
    replicaCount: 0
  logLevel: "WARNING"

This is the example of values.yaml with autoscaling:

worker:
  trap:
    autoscaling:
      enabled: true
      minReplicas: 4
      maxReplicas: 10
      targetCPUUtilizationPercentage: 80
  sender:
    autoscaling:
      enabled: true
      minReplicas: 2
      maxReplicas: 5
      targetCPUUtilizationPercentage: 80
  poller:
    replicaCount: 0
  logLevel: "WARNING"

In the example above both trap and sender pods are autoscaled. During an upgrade process minReplicas number of pods is created, and then new ones are created only if CPU threshold exceeds targetCPUUtilizationPercentage which by default is 80%. This solution helps you to keep resources usage adjusted to what you actually need.

After helm upgrade process, you will see horizontalpodautoscaler in microk8s kubectl get all -n sc4snmp:

NAME                                                                             REFERENCE                                               TARGETS   MINPODS   MAXPODS   REPLICAS   AGE
horizontalpodautoscaler.autoscaling/snmp-mibserver                               Deployment/snmp-mibserver                               1%/80%    1         3         1          97m
horizontalpodautoscaler.autoscaling/snmp-splunk-connect-for-snmp-worker-sender   Deployment/snmp-splunk-connect-for-snmp-worker-sender   1%/80%    2         5         2          28m
horizontalpodautoscaler.autoscaling/snmp-splunk-connect-for-snmp-worker-trap     Deployment/snmp-splunk-connect-for-snmp-worker-trap     1%/80%    4         10        4          28m

If you see <unknown>/80% in TARGETS section instead of the CPU percentage, you probably don’t have metrics-server addon enabled. Enable it using: microk8s enable metrics-server.

Real life scenario: I have a significant delay in polling¶

Sometimes when polling is configured to be run frequently and on many devices, workers get overloaded and there is a delay in delivering data to splunk. To avoid such situations we can scale poller and sender pods. Because of the walk cycles (walk is a costly operation ran once for a while), poller workers require more resources for a short time. For this reason, enabling autoscaling is recommended.

This is the example of values.yaml with autoscaling:

worker:
  trap:
    autoscaling:
      enabled: true
      minReplicas: 4
      maxReplicas: 10
      targetCPUUtilizationPercentage: 80
  sender:
    autoscaling:
      enabled: true
      minReplicas: 2
      maxReplicas: 5
      targetCPUUtilizationPercentage: 80
  poller:
    autoscaling:
      enabled: true
      minReplicas: 2
      maxReplicas: 20
      targetCPUUtilizationPercentage: 80
  logLevel: "WARNING"

Remember, that the system won’t scale itself infinitely, there is a finite amount of resources that you can allocate. By default, every worker has configured following resources:

    resources:
      limits:
        cpu: 500m
      requests:
        cpu: 250m

I have autoscaling enabled and experience problems with Mongo and Redis pod¶

If MongoDB and Redis pods are crushing, and some of the pods are in infinite Pending state, that means you’re over your resources and SC4SNMP cannot scale more. You should decrease number of maxReplicas in workers, so that it’s not going beyond available CPU.

I don’t know how to set autoscaling parameters and how many replicas I need¶

The best way to see if pods are overloaded is to run command:

microk8s kubectl top pods -n sc4snmp

NAME                                                          CPU(cores)   MEMORY(bytes)   
snmp-mibserver-7f879c5b7c-nnlfj                               1m           3Mi             
snmp-mongodb-869cc8586f-q8lkm                                 18m          225Mi           
snmp-redis-master-0                                           10m          2Mi             
snmp-splunk-connect-for-snmp-scheduler-558dccfb54-nb97j       2m           136Mi           
snmp-splunk-connect-for-snmp-trap-5878f89bbf-24wrz            2m           129Mi           
snmp-splunk-connect-for-snmp-trap-5878f89bbf-z9gd5            2m           129Mi           
snmp-splunk-connect-for-snmp-worker-poller-599c7fdbfb-cfqjm   260m         354Mi           
snmp-splunk-connect-for-snmp-worker-poller-599c7fdbfb-ztf7l   312m         553Mi           
snmp-splunk-connect-for-snmp-worker-sender-579f796bbd-vmw88   14m           257Mi           
snmp-splunk-connect-for-snmp-worker-trap-5474db6fc6-46zhf     3m           259Mi           
snmp-splunk-connect-for-snmp-worker-trap-5474db6fc6-mjtpv     4m           259Mi

Here you can see how much CPU and Memory is being used by the pods. If the CPU is close to 500m (which is the limit for one pod by default), you should enable autoscaling/increase maxReplicas or increase replicaCount with autoscaling off.

Here you can read about Horizontal Autoscaling and how to adjust maximum replica value to the resources you have: Horizontal Autoscaling.

Worker parameters¶

variable	description	default
worker.taskTimeout	task timeout in seconds (usually necessary when walk process takes a long time)	2400
worker.poller.replicaCount	number of poller worker replicas	2
worker.poller.autoscaling.enabled	enabling autoscaling for poller worker pods	false
worker.poller.autoscaling.minReplicas	minimum number of running poller worker pods when autoscaling is enabled	2
worker.poller.autoscaling.maxReplicas	maximum number of running poller worker pods when autoscaling is enabled	40
worker.poller.autoscaling.targetCPUUtilizationPercentage	CPU % threshold that must be exceeded on poller worker pods to spawn another replica	80
worker.poller.resources.limits	the resources limits for poller worker container	{}
worker.poller.resources.requests	the requested resources for poller worker container	{}
worker.trap.replicaCount	number of trap worker replicas	2
worker.trap.autoscaling.enabled	enabling autoscaling for trap worker pods	false
worker.trap.autoscaling.minReplicas	minimum number of running trap worker pods when autoscaling is enabled	2
worker.trap.autoscaling.maxReplicas	maximum number of running trap worker pods when autoscaling is enabled	40
worker.trap.autoscaling.targetCPUUtilizationPercentage	CPU % threshold that must be exceeded on trap worker pods to spawn another replica	80
worker.trap.resources.limits	the resources limits for poller worker container	{}
worker.trap.resources.requests	the requested resources for poller worker container	{}
worker.sender.replicaCount	number of sender worker replicas	2
worker.sender.autoscaling.enabled	enabling autoscaling for sender worker pods	false
worker.sender.autoscaling.minReplicas	minimum number of running sender worker pods when autoscaling is enabled	2
worker.sender.autoscaling.maxReplicas	maximum number of running sender worker pods when autoscaling is enabled	40
worker.sender.autoscaling.targetCPUUtilizationPercentage	CPU % threshold that must be exceeded on sender worker pods to spawn another replica	80
worker.sender.resources.limits	the resources limits for poller worker container	{}
worker.sender.resources.requests	the requested resources for poller worker container	{}