Garbage Collector Configuration

Overview

The garbage collector is a maintenance service that terminates user desktop pods that have been idle for longer than a configurable duration. It is invoked via an HTTP request to the pyos API endpoint /API/manager/garbagecollector.

API Parameters

Parameter	Type	Required	Default	Description
expirein	integer (seconds)	Yes	—	Sessions idle for longer than this duration are candidates for deletion.
force	boolean	No	False	If True, deletes sessions regardless of current connection status. If False, active sessions are preserved.
nodename	string	No	None	Restricts deletion to pods running on the specified node. Useful during node maintenance or kubectl drain operations.

The endpoint returns a JSON array listing all deleted pod names.

Warning: Calling /API/manager/garbagecollector?expirein=0&force=true immediately terminates all active user sessions.

Deploying the Garbage Collector as a Kubernetes CronJob

The recommended deployment pattern is a Kubernetes CronJob that invokes the garbage collector API on a scheduled interval.

Create a file named cronjob.yaml:

apiVersion: batch/v1
kind: CronJob
metadata:
  name: abcdesktop-cleaner
spec:
  schedule: "*/5 * * * *"
  failedJobsHistoryLimit: 2
  jobTemplate:
    spec:
      activeDeadlineSeconds: 60
      backoffLimit: 0
      template:
        metadata:
          labels:
            name: abcdesktop-cleaner
            netpol/pyos: 'true'
            netpol/dns: 'true'
        spec:
          containers:
          - name: abcdesktop-cleaner
            image: busybox
            args:
            - /bin/sh
            - -c
            - wget -q -t 3 -O- "http://${PYOS_ADDR}:${PYOS_PORT}/API/manager/garbagecollector?expirein=${EXPIREIN}&force=${FORCE}"
            env:
            - name: PYOS_ADDR
              value: 'pyos.abcdesktop.svc.cluster.local'
            - name: PYOS_PORT
              value: '8000'
            - name: EXPIREIN
              value: '900'
            - name: FORCE
              value: 'false'
          restartPolicy: OnFailure
          dnsPolicy: ClusterFirst

Default Values

• Schedule: */5 * * * * — runs every five minutes.
• expirein: 900 seconds (15 minutes) — sessions idle for longer than 900 seconds are candidates for deletion.
• force: false — active sessions are never deleted.

Deletion Logic

Condition	force	Result
User is not connected AND idle time > expirein	any	Pod is deleted
User is connected AND idle time > expirein	true	Pod is deleted
User is connected AND idle time > expirein	false	Pod is preserved
Idle time ≤ expirein	any	No action

Applying the CronJob

Deploy the CronJob to your namespace:

kubectl apply -f cronjob.yaml -n abcdesktop

Expected output:

cronjob.batch/abcdesktop-cleaner configured

Monitoring CronJob Execution

After a few minutes, verify the job execution history:

kubectl get jobs -n abcdesktop

Expected output:

NAME                          STATUS     COMPLETIONS   DURATION   AGE
abcdesktop-cleaner-29350635   Complete   1/1           5s         13m
abcdesktop-cleaner-29350640   Complete   1/1           5s         8m56s
abcdesktop-cleaner-29350645   Complete   1/1           5s         3m56s

List completed cleaner pods:

kubectl get pods -n abcdesktop

Expected output:

NAME                                READY   STATUS      RESTARTS   AGE    IP           NODE       NOMINATED NODE   READINESS GATES
abcdesktop-cleaner-29350665-lnk6g   0/1     Completed   0          12m    10.0.2.199   abc3ws03   <none>           <none>
abcdesktop-cleaner-29350670-c8dtj   0/1     Completed   0          7m9s   10.0.2.252   abc3ws03   <none>           <none>
abcdesktop-cleaner-29350675-hlj8z   0/1     Completed   0          2m9s   10.0.2.72    abc3ws03   <none>           <none>
[...]

The garbage collector CronJob is now running every five minutes and automatically reclaiming idle desktop pod resources.