AzureMonitorAgent: disk space check respects volumes mounted under AMA paths

[luckylucab0]

Problem

check_disk_space_availability() in AzureMonitorAgent/agent.py calls os.statvfs() on /var, /etc and /opt. statvfs always reports stats for the filesystem that owns the given path — i.e. the top-level mount — so any filesystem mounted deeper (e.g. a dedicated volume at /var/opt/microsoft/azuremonitoragent) is invisible to the check.

In setups where dedicated volumes are mounted at the AMA target directories (common in environments with constrained /var, /etc or /opt, or with per-application volume policies), the check evaluates the wrong filesystem. The top-level mount may have less than 700/500/500 MB free even though the actual AMA volumes have several GB available, causing the extension to abort with exit code 52 (MissingDependency) and the message <operation> failed due to low disk space.

Fix

Two small, behavior-preserving changes in AzureMonitorAgent/agent.py:

1. check_disk_space_availability() now checks the AMA target directories themselves (/var/opt/microsoft/azuremonitoragent, /etc/opt/microsoft/azuremonitoragent, /opt/microsoft/azuremonitoragent) with the same size thresholds (700 / 500 / 500 MB).
2. get_free_space_mb() walks up the path to the nearest existing ancestor before calling os.statvfs(). This keeps the check working on a fresh install where the AMA directories do not exist yet (it transparently falls back to /var, /etc, /opt), and lets it measure the correct filesystem once a dedicated volume is mounted at the AMA subdirectory.

Also adds a one-line log indicating which path was short on space and by how much — useful when diagnosing exactly this scenario.

Behavior matrix

Exit code, error message, list of destructive operations skipped, and the thresholds (700 / 500 / 500 MB) are unchanged.

Testing

• python -m py_compile AzureMonitorAgent/agent.py — passes.
• Manual: created a small loopback filesystem, mounted it at /var/opt/microsoft/azuremonitoragent, ran the extension on a VM with /var near full → previously failed with code 52, now succeeds.
• Manual: removed the AMA directories entirely, confirmed the walk-up falls back to /var, /etc, /opt and reproduces the previous behavior.

Risk

Low. The change is local to two helper functions, preserves the exit code contract (MissingDependency = 52), and falls back to the previous code paths whenever the AMA subdirectories are absent.

[luckylucab0]

@microsoft-github-policy-service agree company="Swisscom"

[simathih]

can

[luckylucab0]

Should be fixed now, sorry for circumstances :)

[simathih]

thanks @luckylucab0 for the contribution!