agent does not report inodes usage

[nioncode]

Describe the bug

ops-agent seems to not collect the system.filesystem.inodes.usage metric.

To Reproduce
Steps to reproduce the behavior:

1. Run a GCE VM with Ops Agent --version=2.48.0 and default config.
2. Wait for metrics to show up in Cloud Monitoring.
3. See that there are metrics for disk usage, but NOT inodes usage

Expected behavior
Inodes usage should be collected and available in Cloud Monitoring.

Environment (please complete the following information):

• VM distro / OS: Ubuntu 24.04
• Ops Agent version: 2.48.0
• Ops Agent configuration: default

Additional context
It seems like the metric is explicitly filtered out in the ops-agent (see

ops-agent/apps/hostmetrics.go

Line 325 in 1246ead

"system.filesystem.inodes.usage",

). Is there any reason for this? Can this be overridden with a custom config?

[github-actions]

This issue was marked stale due to lack of activity. It will be closed in 14 days.

[nioncode]

Still relevant to us.

[jefferbrecht]

Hi @nioncode,

That the metric is missing is intentional. The short answer is that there's no platform-side agent.googleapis.com/... metric schema defined yet for an inode metric (although there are comparable ones for e.g. GKE platform metrics); the longer answer is that we can accept this as a feature request and work to resolve that blocker so that we can get it added to the Ops Agent. We'll add an internal tracker for this and see whether we can get it prioritized.

Apologies for the lack of response so far.

[diegopvj]

@nioncode did you solved this by any other way? I'm trying to monitor inode by creating an alert but I'm stuck in this ops agents problem too.

[nioncode]

@diegopvj yes, we created a custom python script to publish the inodes data of the mountpoint that was relevant to us. We then call this script on our GCP VMs via a systemd timer every minute.

import argparse
import subprocess
import time

from google.api import metric_pb2 as ga_metric
from google.cloud import monitoring_v3

INODE_USAGE_METRIC_TYPE = "workload.googleapis.com/inodes_percent_used"


def create_inode_usage_metric(project_id):
    client = monitoring_v3.MetricServiceClient()
    project_name = f"projects/{project_id}"

    descriptor = ga_metric.MetricDescriptor()
    descriptor.display_name = "Inode Usage"
    descriptor.description = "Tracks the percentage of used inodes on a disk"
    descriptor.type = INODE_USAGE_METRIC_TYPE
    descriptor.metric_kind = ga_metric.MetricDescriptor.MetricKind.GAUGE
    descriptor.value_type = ga_metric.MetricDescriptor.ValueType.DOUBLE
    descriptor.unit = "%"

    client.create_metric_descriptor(name=project_name, metric_descriptor=descriptor)


def report_inode_usage(project_id, instance_id, instance_zone, used_inodes_percentage):
    client = monitoring_v3.MetricServiceClient()
    project_name = f"projects/{project_id}"

    series = monitoring_v3.TimeSeries()
    series.metric.type = INODE_USAGE_METRIC_TYPE
    series.resource.type = "gce_instance"
    series.resource.labels["instance_id"] = instance_id
    series.resource.labels["zone"] = instance_zone

    interval = monitoring_v3.TimeInterval({"end_time": {"seconds": int(time.time())}})
    point = monitoring_v3.Point(
        {"interval": interval, "value": {"double_value": used_inodes_percentage}}
    )
    series.points = [point]
    client.create_time_series(name=project_name, time_series=[series])


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description="Send the 'inode usage' metric to Google Cloud Monitoring."
    )
    parser.add_argument("project_id", type=str, help="GC Project ID")
    parser.add_argument("instance_id", type=str, help="GC Instance ID")
    parser.add_argument("instance_zone", type=str, help="GC Instance zone")
    parser.add_argument(
        "-c", "--create", action="store_true", help="Create the 'inode usage' metric"
    )

    args = parser.parse_args()

    if args.create:
        create_inode_usage_metric(args.project_id)
    else:
        space_usage = (
            subprocess.check_output(
                "df /mnt/storage --output=itotal,iused | tail -n 1",
                shell=True,
            )
            .decode()
            .strip()
        )
        total_inodes = int(space_usage.split()[0])
        used_inodes = int(space_usage.split()[1])
        inode_usage = (used_inodes / total_inodes) * 100
        report_inode_usage(
            args.project_id, args.instance_id, args.instance_zone, inode_usage
        )

[github-actions]

This issue was marked stale due to lack of activity. It will be closed in 14 days.

[nioncode]

Still relevant to us.