Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-62095

"ulimit -n" just 1024 (instead of 1048576)

XMLWordPrintable

    • Icon: Bug Bug
    • Resolution: Unresolved
    • Icon: Critical Critical
    • 4.20.0
    • 4.20
    • Node / CRI-O
    • None
    • Quality / Stability / Reliability
    • False
    • Hide

      None

      Show
      None
    • None
    • None
    • Yes
    • None
    • Rejected
    • None
    • Proposed
    • Known Issue
    • Hide
      * Starting with OCP 4.20, there is a decrease in the default maximum open files soft limit for containers. As a consequence, end users may experience application failures. To work around this problem, increase the container runtimes (CRI-O) ulimit configuration. (OCPBUGS-62095).
      Show
      * Starting with OCP 4.20, there is a decrease in the default maximum open files soft limit for containers. As a consequence, end users may experience application failures. To work around this problem, increase the container runtimes (CRI-O) ulimit configuration. ( OCPBUGS-62095 ).
    • None
    • None
    • None
    • None

      Description of problem:

      On recent OCP 4.20 buids, the containes seem to have tighter limits on max open files (ulimit -n) than previous OCP version (1024 vs 1048576)
      
      This breaks existing applications.

      Version-Release number of selected component (if applicable):

          4.20.0-0.nightly-2025-09-19-231410

      How reproducible:

          Always

      Steps to Reproduce:

      # Create
      
      oc create -f - <<EOF
      apiVersion: batch/v1
      kind: Job
      metadata:
        name: ulimit-check
      spec:
        template:
          spec:
            containers:
            - name: ulimit
              image: quay.io/fedora/fedora:42
              command: ["ulimit", "-n"]
            restartPolicy: Never
        backoffLimit: 0
      EOF
      
      # Check logs of the container

      Actual results:

          1024

      Expected results:

          1048576

      Additional info:

      This seem to break things, specifically,  I have noticed this while testing the Serverless Operator, its kourier-gateway crashes due to its inability to open new sockets even during parallel e2e test runs, not just performance tests)

              sgrunert@redhat.com Sascha Grunert
              maschmid@redhat.com Marek Schmidt
              None
              None
              Min Li Min Li
              Katie Drake Katie Drake
              Votes:
              0 Vote for this issue
              Watchers:
              13 Start watching this issue

                Created:
                Updated: