Lightweight: - GPU collection moves to a dedicated worker thread that owns the gfxinfo handle for the process lifetime. gfxinfo's active_gpu() runs a full NVML init/teardown (~20ms, blocking) and we were paying it on the async runtime for every collect — measured at ~80% of the agent's entire active CPU on a GPU machine. The handle holds Rc<Nvml> (not Send), so a thread + mpsc/oneshot channel pair confines it; a zero-total-VRAM reply is treated as a dead session (driver reload) and re-probed. - journalctl now runs via tokio::process instead of blocking one of the two runtime workers for the duration of the subprocess. - TtlCell (state.rs) replaces the four hand-rolled static TTL caches; a cached negative result now counts as fresh, so hosts with no matching temp sensor or GPU stop rescanning every request. Single lock+clone on the GPU cache hit path (was two). Correctness: - Process/child CPU times are now microseconds as documented; they were milliseconds, rendering 1000x too small next to (correct) thread times. - Non-Linux per-process CPU%% clamps AFTER dividing by core count; a 4-cores-busy process on an 8-core box reported 12.5% instead of 50%. - Journal timestamps are real RFC 3339 UTC plus an additive timestamp_us field (sorting is now numeric); the old strings were Debug-formatted SystemTime mangled by string replace. - Partition detection uses /sys/block on Linux: whole-disk filesystems on names like nvme0n1 or zram1 are no longer misclassified as partitions. One shared parent_disk_name() replaces two inline copies. - New sampled_at_ms on the metrics payload (additive) records when the snapshot was actually collected, so clients can compute exact rates across the agent's TTL cache. Security/robustness: - key.pem is created 0600 (was umask default 0644, world-readable) and pre-1.51 keys are tightened on startup. - Per-PID detail/journal caches now evict (60s max age, 64 entries max); they previously grew without bound under PID-walking clients. - The two per-PID ws handlers collapse into one generic helper. - /proc/<pid>/stat parsing unified in one comm-safe module. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
socktop_agent (server)
Lightweight on‑demand metrics WebSocket server for the socktop TUI.
Highlights:
- Collects system metrics only when requested (keeps idle CPU <1%)
- Optional TLS (self‑signed cert auto‑generated & pinned by client)
- JSON for fast metrics / disks; protobuf (optionally gzipped) for processes
- Accurate per‑process CPU% on Linux via /proc jiffies delta
- Optional GPU & temperature metrics (disable via env vars)
- Simple token auth (?token=...) support
Run (no TLS):
cargo install socktop_agent
socktop_agent --port 3000
Enable TLS:
SOCKTOP_ENABLE_SSL=1 socktop_agent --port 8443
# cert/key stored under $XDG_DATA_HOME/socktop_agent/tls
Environment toggles:
- SOCKTOP_AGENT_GPU=0 (disable GPU collection)
- SOCKTOP_AGENT_TEMP=0 (disable temperature)
- SOCKTOP_TOKEN=secret (require token param from client)
- SOCKTOP_AGENT_METRICS_TTL_MS=250 (cache fast metrics window)
- SOCKTOP_AGENT_PROCESSES_TTL_MS=1000
- SOCKTOP_AGENT_DISKS_TTL_MS=1000
NOTE ON ENV vars
Generally these have been added for debugging purposes. you do not need to configure them, default values are tuned and GPU will deisable itself after the first poll if not available.
Systemd unit example & full docs: https://github.com/jasonwitty/socktop
WebSocket API Integration Guide
The socktop_agent exposes a WebSocket API that can be directly integrated with your own applications. This allows you to build custom monitoring dashboards or analysis tools using the agent's metrics.
WebSocket Endpoint
ws://HOST:PORT/ws # Without TLS
wss://HOST:PORT/ws # With TLS
With authentication token (if configured):
ws://HOST:PORT/ws?token=YOUR_TOKEN
wss://HOST:PORT/ws?token=YOUR_TOKEN
Communication Protocol
All communication uses JSON format for requests and responses, except for the process list which uses Protocol Buffers (protobuf) format with optional gzip compression.
Request Types
Send a JSON message with a type field to request specific metrics:
{"type": "metrics"} // Request fast-changing metrics (CPU, memory, network)
{"type": "disks"} // Request disk information
{"type": "processes"} // Request process list (returns protobuf)
Response Formats
- Fast Metrics (JSON):
{
"cpu_total": 12.4,
"cpu_per_core": [11.2, 15.7],
"mem_total": 33554432,
"mem_used": 18321408,
"swap_total": 0,
"swap_used": 0,
"hostname": "myserver",
"cpu_temp_c": 42.5,
"networks": [{"name":"eth0","received":12345678,"transmitted":87654321}],
"gpus": [{"name":"nvidia-0","usage":56.7,"memory_total":8589934592,"memory_used":1073741824,"temp_c":65.0}]
}
- Disks (JSON):
[
{"name":"nvme0n1p2","total":512000000000,"available":320000000000},
{"name":"sda1","total":1000000000000,"available":750000000000}
]
- Processes (Protocol Buffers):
Processes are returned in Protocol Buffers format, optionally gzip-compressed for large process lists. The protobuf schema is:
syntax = "proto3";
message Process {
uint32 pid = 1;
string name = 2;
float cpu_usage = 3;
uint64 mem_bytes = 4;
}
message ProcessList {
uint32 process_count = 1;
repeated Process processes = 2;
}
Example Integration (JavaScript/Node.js)
const WebSocket = require('ws');
// Connect to the agent
const ws = new WebSocket('ws://localhost:3000/ws');
ws.on('open', function open() {
console.log('Connected to socktop_agent');
// Request metrics immediately on connection
ws.send(JSON.stringify({type: 'metrics'}));
// Set up regular polling
setInterval(() => {
ws.send(JSON.stringify({type: 'metrics'}));
}, 1000);
// Request processes every 3 seconds
setInterval(() => {
ws.send(JSON.stringify({type: 'processes'}));
}, 3000);
});
ws.on('message', function incoming(data) {
// Check if the response is JSON or binary (protobuf)
try {
const jsonData = JSON.parse(data);
console.log('Received JSON data:', jsonData);
} catch (e) {
console.log('Received binary data (protobuf), length:', data.length);
// Process binary protobuf data with a library like protobufjs
}
});
ws.on('close', function close() {
console.log('Disconnected from socktop_agent');
});
Example Integration (Python)
import json
import asyncio
import websockets
async def monitor_system():
uri = "ws://localhost:3000/ws"
async with websockets.connect(uri) as websocket:
print("Connected to socktop_agent")
# Request initial metrics
await websocket.send(json.dumps({"type": "metrics"}))
# Set up regular polling
while True:
# Request metrics
await websocket.send(json.dumps({"type": "metrics"}))
# Receive and process response
response = await websocket.recv()
# Check if response is JSON or binary (protobuf)
try:
data = json.loads(response)
print(f"CPU: {data['cpu_total']}%, Memory: {data['mem_used']/data['mem_total']*100:.1f}%")
except json.JSONDecodeError:
print(f"Received binary data, length: {len(response)}")
# Process binary protobuf data with a library like protobuf
# Wait before next poll
await asyncio.sleep(1)
asyncio.run(monitor_system())
Notes for Integration
-
Error Handling: The WebSocket connection may close unexpectedly; implement reconnection logic in your client.
-
Rate Limiting: Avoid excessive polling that could impact the system being monitored. Recommended intervals:
- Metrics: 500ms or slower
- Processes: 2000ms or slower
- Disks: 5000ms or slower
-
Authentication: If the agent is configured with a token, always include it in the WebSocket URL.
-
Protocol Buffers Handling: For processing the binary process list data, use a Protocol Buffers library for your language and the schema provided in the
proto/processes.protofile. -
Compression: Process lists may be gzip-compressed. Check if the response starts with the gzip magic bytes (
0x1f, 0x8b) and decompress if necessary.
LLM Integration Guide
If you're using an LLM to generate code for integrating with socktop_agent, this section provides structured information to help the model understand the API better.
API Schema
# WebSocket API Schema for socktop_agent
endpoint: ws://HOST:PORT/ws or wss://HOST:PORT/ws (with TLS)
authentication:
type: query parameter
parameter: token
example: ws://HOST:PORT/ws?token=YOUR_TOKEN
requests:
- type: metrics
format: JSON
example: {"type": "metrics"}
description: Fast-changing metrics (CPU, memory, network)
- type: disks
format: JSON
example: {"type": "disks"}
description: Disk information
- type: processes
format: JSON
example: {"type": "processes"}
description: Process list (returns protobuf)
responses:
- request_type: metrics
format: JSON
schema:
cpu_total: float # percentage of total CPU usage
cpu_per_core: [float] # array of per-core CPU usage percentages
mem_total: uint64 # total memory in bytes
mem_used: uint64 # used memory in bytes
swap_total: uint64 # total swap in bytes
swap_used: uint64 # used swap in bytes
hostname: string # system hostname
cpu_temp_c: float? # CPU temperature in Celsius (optional)
networks: [
{
name: string # network interface name
received: uint64 # total bytes received
transmitted: uint64 # total bytes transmitted
}
]
gpus: [
{
name: string # GPU device name
usage: float # GPU usage percentage
memory_total: uint64 # total GPU memory in bytes
memory_used: uint64 # used GPU memory in bytes
temp_c: float # GPU temperature in Celsius
}
]?
- request_type: disks
format: JSON
schema:
[
{
name: string # disk name
total: uint64 # total space in bytes
available: uint64 # available space in bytes
}
]
- request_type: processes
format: Protocol Buffers (optionally gzip-compressed)
schema: See protobuf definition below
Protobuf Schema (processes.proto)
syntax = "proto3";
message Process {
uint32 pid = 1;
string name = 2;
float cpu_usage = 3;
uint64 mem_bytes = 4;
}
message ProcessList {
uint32 process_count = 1;
repeated Process processes = 2;
}
Step-by-Step Integration Pseudocode
1. Establish WebSocket connection to ws://HOST:PORT/ws
- Add token if required: ws://HOST:PORT/ws?token=YOUR_TOKEN
2. For regular metrics updates:
- Send: {"type": "metrics"}
- Parse JSON response
- Extract CPU, memory, network info
3. For disk information:
- Send: {"type": "disks"}
- Parse JSON response
- Extract disk usage data
4. For process list:
- Send: {"type": "processes"}
- Check if response is binary
- If starts with 0x1f, 0x8b bytes:
- Decompress using gzip
- Parse binary data using protobuf schema
- Extract process information
5. Implement reconnection logic:
- On connection close/error
- Use exponential backoff
6. Respect rate limits:
- metrics: ≥ 500ms interval
- disks: ≥ 5000ms interval
- processes: ≥ 2000ms interval
Common Implementation Patterns
Pattern 1: Periodic Polling
// Set up separate timers for different metric types
const metricsInterval = setInterval(() => ws.send(JSON.stringify({type: 'metrics'})), 500);
const disksInterval = setInterval(() => ws.send(JSON.stringify({type: 'disks'})), 5000);
const processesInterval = setInterval(() => ws.send(JSON.stringify({type: 'processes'})), 2000);
// Clean up on disconnect
ws.on('close', () => {
clearInterval(metricsInterval);
clearInterval(disksInterval);
clearInterval(processesInterval);
});
Pattern 2: Processing Binary Protobuf Data
// Using protobufjs
const root = protobuf.loadSync('processes.proto');
const ProcessList = root.lookupType('ProcessList');
ws.on('message', function(data) {
if (typeof data !== 'string') {
// Check for gzip compression
if (data[0] === 0x1f && data[1] === 0x8b) {
data = gunzipSync(data); // Use appropriate decompression library
}
// Decode protobuf
const processes = ProcessList.decode(new Uint8Array(data));
console.log(`Total processes: ${processes.process_count}`);
processes.processes.forEach(p => {
console.log(`PID: ${p.pid}, Name: ${p.name}, CPU: ${p.cpu_usage}%`);
});
}
});
Pattern 3: Reconnection Logic
function connect() {
const ws = new WebSocket('ws://localhost:3000/ws');
ws.on('open', () => {
console.log('Connected');
// Start polling
});
ws.on('close', () => {
console.log('Connection lost, reconnecting...');
setTimeout(connect, 1000); // Reconnect after 1 second
});
// Handle other events...
}
connect();