File descriptor limit exceeded

Problem

You see the following error message in your eBPF agent logs:

The number of used file descriptors (820) is above the threshold (819). This may cause issues with attaching uprobes. Consider increasing the process FD limit

This error indicates that the eBPF agent has reached the maximum number of file descriptors it's allowed to use, which can prevent it from properly monitoring your applications.

Solution

You must increase the file descriptor limit for the eBPF agent service or pod.

For Linux hosts:

Use the systemd override mechanism to set a higher limit for the agent service.

1. Increase the file descriptor limit temporarily:

   bash

   コピー

   $
   ulimit -n 4096

2. Run the following command to edit the change persistent at /etc/security/limits.conf:

   bash

   コピー

   $
   sudo nano /etc/security/limits.conf

   Add the following lines:

   * soft nofile 4096
      * hard nofile 8192

   コピー

3. Restart the eBPF agent:

   bash

   コピー

   $
   sudo systemctl restart newrelic-ebpf-agent

4. Verify the new limit:

   bash

   コピー

   $
   # Check current limits
   $
      cat /proc/$(pgrep newrelic-ebpf-agent)/limits | grep "Max open files"

For Kubernetes:

1. Edit your values.yaml file used for the Helm deployment:

   # values.yaml
      agent:
      resources:
         limits:
            memory: "2Gi"
      # Add security context if needed
      securityContext:
         capabilities:
            add:
            - SYS_ADMIN

   コピー

2. Apply the changes:

   bash

   コピー

   $
   helm upgrade nr-ebpf-agent newrelic/nr-ebpf-agent -n newrelic -f values.yaml

3. Verify the pods restart and are in a Running state:

   bash

   コピー

   $
   kubectl get pods -n newrelic -w

Alternative solutions

If increasing resource limits is not feasible, you can reduce the agent's file descriptor usage by limiting the scope of what it monitors.

1. Reduce monitoring scope by filtering out namespaces or services:

   • For Linux hosts (/etc/newrelic-ebpf-agent/newrelic-ebpf-agent.conf):

     bash

     コピー

     $
     # Exclude specific processes or entities
     $
        DROP_DATA_FOR_ENTITY="process1,process2"

   • For Kubernetes (values.yaml):

     # Exclude namespaces or services
        dropDataForNamespaces: ["kube-system", "monitoring"]
        dropDataServiceNameRegex: "kube-dns|system-service"

     コピー

2. Disable specific protocol monitoring to reduce the number of probes:

   • For Linux hosts:

     bash

     コピー

     $
     PROTOCOLS_MYSQL_ENABLED="false"
     $
        PROTOCOLS_MONGODB_ENABLED="false"

   • For Kubernetes:

     protocols:
        mysql:
           enabled: false
        mongodb:
           enabled: false

     コピー

Verification

1. Check that the error no longer appears in the agent logs:

   • For Linux hosts:

     bash

     コピー

     $
     sudo journalctl -u newrelic-ebpf-agent -f

   • For Kubernetes:

     bash

     コピー

     $
     kubectl logs -f <ebpf-agent-pod> -n newrelic

2. Confirm the agent is functioning normally by looking for the startup message in the logs:

   [STEP-7] => Successfully started the eBPF Agent.

   コピー

3. Verify data is flowing to New Relic UI by filtering entities with instrumentation.name = nr_ebpf.

Additional notes

• The file descriptor limit error is more common in environments with many running processes or services.
• Each monitored process/service requires file descriptors for eBPF probe attachment.
• The default system limit (often 1024) may be insufficient for large-scale deployments.
• Increasing the limit to 4096 is generally safe and sufficient for most use cases.