Memory Profiling

Hawk uses Python's built-in tracemalloc to track memory allocations. No external dependencies required.

Endpoints

GET /debug/prof/mem/tracemalloc/           # Profile for fixed duration
GET /debug/prof/mem/tracemalloc/start/     # Start memory tracing
GET /debug/prof/mem/tracemalloc/snapshot/  # Take a snapshot (while tracing)
GET /debug/prof/mem/tracemalloc/stop/      # Stop memory tracing

Parameters

Parameter	Default	Description
duration	5	Profiling duration in seconds
format	lineno	Output: lineno, traceback, pickle
frames	30	Stack frames to capture
count	10	Number of top allocations to return
gc	true	Run garbage collection before profiling
cumulative	false	Show cumulative statistics

Output Formats

• lineno: JSON with per-line allocation statistics (file, line, size, count)
• traceback: JSON with full stack traces for each allocation
• pickle: Binary pickle format for offline analysis with Python's tracemalloc module

Examples

# Get top 10 memory allocations (5 seconds)
curl "http://localhost:8000/debug/prof/mem/tracemalloc/"

# Get top 20 allocations with full tracebacks
curl "http://localhost:8000/debug/prof/mem/tracemalloc/?count=20&format=traceback"

# Download for offline analysis
curl "http://localhost:8000/debug/prof/mem/tracemalloc/?format=pickle" > snapshot.pkl

Interval Profiling

For tracking memory growth over time, use manual start/snapshot/stop:

# Start tracing
curl "http://localhost:8000/debug/prof/mem/tracemalloc/start/"

# Take snapshots at intervals to compare
curl "http://localhost:8000/debug/prof/mem/tracemalloc/snapshot/" > t0.json
# ... wait ...
curl "http://localhost:8000/debug/prof/mem/tracemalloc/snapshot/" > t1.json

# Stop tracing
curl "http://localhost:8000/debug/prof/mem/tracemalloc/stop/"

Offline Analysis

Download a pickle snapshot and analyze locally:

import pickle
import tracemalloc

with open("snapshot.pkl", "rb") as f:
    snapshot = pickle.load(f)

# Show top allocations
top_stats = snapshot.statistics("lineno")
for stat in top_stats[:10]:
    print(stat)