Disk pressure and delivery failures

Check both bytes and inodes with df -h / and df -i /. Inspect journalctl --disk-usage and findmnt /var/lib/alloy. The Alloy mount must remain bounded, and journals must remain within their configured budget. Do not delete application data or unrelated logs to repair telemetry.

If the public VPS store is nearly full, confirm the dedicated 8 GiB filesystem is mounted. Check VictoriaMetrics free-space behavior and write errors. At the configured minimum-free threshold, writes may be refused; queues are intentionally short. Do not remove the hard size bound to hide the symptom. Investigate unexpected series cardinality and retention behavior.

For gateway errors: 401 indicates an authentication mismatch, 400 indicates invalid public telemetry, 415 indicates unsupported encoding/version, 429 indicates bounded capacity, and 503 indicates unavailable storage. Repeated 400 errors require a collector/catalog mismatch fix; retries alone cannot repair labels.

For private delivery failures, check the dedicated Grafana credentials, endpoints, account quotas, clock and connectivity. Preserve the existing shared VPS monitoring configuration. After repair, verify fresh public and private samples independently and record any lost interval.