Server Monitoring

Sysmos provides deep server observability through a lightweight agent that uses eBPF (extended Berkeley Packet Filter) to collect kernel-level metrics without modifying your applications.

What Gets Collected

The agent collects metrics across several categories:

System Metrics

• CPU — Per-core utilization, user vs system time, load averages
• Memory — Used, available, cached, swap usage
• Disk — Read/write IOPS, throughput, latency, space utilization
• Network — Bytes in/out, packets, errors, connection counts

eBPF-Powered Metrics

• TCP connections — Active connections, new/closed per second, latency distribution
• Process monitoring — Top processes by CPU, memory, and I/O
• DNS resolution — Query latency and failure rates
• File I/O — Per-file read/write latency at the kernel level

Anomaly Detection

The agent includes built-in EWMA (Exponentially Weighted Moving Average) anomaly detection. It learns the baseline behavior of your metrics and alerts when values deviate significantly.

This runs locally on the agent — no data needs to leave your server for anomaly detection to work.

Offline Resilience

If the network connection to Sysmos is interrupted, the agent buffers metrics locally using BadgerDB (up to 500 MB by default). When connectivity is restored, buffered data is automatically flushed to the backend.

Performance

The agent is designed for minimal impact:

• Target: Less than 2% host CPU overhead
• Memory: Typically 30-50 MB RSS
• Disk: Configurable buffer size (default 500 MB)

All eBPF resources are properly cleaned up on shutdown with deferred close handlers.