pip install pulselog

📚 Table of Contents

• Performance Tiers
• Benchmarks
• Quick Start
• Why pulselog?
• Dashboard
• API
• stdlib logging Integration
• Configuration
• Production Usage
• Examples
• Design Notes
• Requirements
• License

🚀 Performance Tiers

The Rust-backed native extension delivers 2× throughput by moving the hot path into compiled code. Same API, zero code changes — just a different install method.

# Pure Python (default)
pip install pulselog

# Rust extension (macOS / Linux)
pip install maturin
git clone https://github.com/your-repo/pulselog
cd pulselog
maturin build --release
pip install dist/*.whl

📊 Benchmarks

Pure Python

(Windows, Python 3.12, dashboard=False)

Rust Extension

(macOS, release build)

Latency Under Concurrent Load

p99 with 4 background threads flooding the queue (Pure Python):

p99 is lower under load than idle — per-thread sharding means concurrent producers create zero interference with each other.

Version Improvements (v0.1.2 → v2.0.0)

A typical ML training loop logs 10–100 records/sec. PulseLog handles 2,164× that load before any issues.

⚡ Quick Start

from pulselog import Logger

log = Logger("my-app", checkpoint_path=":memory:")

log.info("training started", epoch=1)
log.warning("learning rate too high", lr=0.1)
log.save("epoch-1", {"acc": 0.91, "loss": 0.23}, status="DONE", progress=33)

log.shutdown()

A browser tab opens at http://localhost:5678 and streams every log in real time.

🤔 Why pulselog?

Standard logging blocks the calling thread on every write — waiting for a file, a socket, or a database. In tight loops (ML training, data pipelines, inference servers) this adds up fast.

pulselog never blocks. Every log call enqueues a record in O(1) and returns immediately. A daemon worker drains the queue every 10ms and pushes batches to the dashboard over WebSocket.

log.info()             ← O(1), ~1.8µs, never blocks
      │
      ▼
  ShardedLogQueue       ← per-thread deques, zero cross-thread contention
      │                    each thread writes to its own private deque
      │
      ▼  every 10ms (adaptive — halves under load)
BackgroundWorker        ← daemon thread, fan-drains all shards
      │
      ├──▶ DashboardServer.broadcast()  ← WebSocket → live browser
      │
      └──▶ (custom handlers)

Why per-thread sharding matters: A single shared queue means all producer threads compete for the same lock on every put(). At 8 threads, that bottleneck cut throughput from 356k to 41k/sec — an 87% collapse. Per-thread sharding eliminates the shared state entirely. Each thread appends to its own deque (a GIL-atomic operation) and the worker fan-drains all shards once per cycle. Result: multi-thread throughput matches single-thread.

🖥️ Dashboard

Single self-contained HTML file served over WebSocket — no build step, no CDN, no npm.

Logs Tab

• 🎨 Colour-coded by level — DEBUG gray · INFO blue · WARNING amber · ERROR/CRITICAL red
• 🔍 Full-text search — filter logs by message content in real time
• 🎚️ Level filter — toggle DEBUG, INFO, WARNING, ERROR, CRITICAL visibility
• 📜 Virtual list rendering — 100k+ logs with zero browser lag
• ⏬ Auto-scroll with manual scroll override
• 📤 Export session as JSON

Checkpoints Tab

• 📈 Progress bar per checkpoint (overall = average across all checkpoints)
• 🔎 Expandable JSON data viewer
• 🏷️ Status badges — DONE ✅ · IN_PROGRESS 🟡 · FAILED 🔴 · SKIPPED ⚫
• 🔍 Search checkpoints by name

🔧 API

Logger

log = Logger(
    name             = "my-app",
    host             = "localhost",
    port             = 5678,          # auto-increments if taken
    auto_open        = True,          # open browser on start
    dashboard        = True,          # False for CI / production
    checkpoint_path  = ".pulselog/checkpoints.db",  # ":memory:" for in-memory
    level            = "DEBUG",
    worker_interval  = 0.01,          # drain interval in seconds (default 10ms)
    queue_size       = 100_000,       # max records before oldest evicted
    overflow         = "drop",        # "drop" | "block" | "raise"
)

Logging

log.debug("msg", **extra)
log.info("msg", **extra)
log.warning("msg", **extra)
log.error("msg", **extra)
log.critical("msg", **extra)

# kwargs appear as structured metadata in the dashboard
log.info("request handled", user_id=42, latency_ms=12, status=200)

# exception() captures the current traceback automatically
try:
    result = model.predict(x)
except Exception:
    log.exception("prediction failed", input_shape=str(x.shape))

Zero-Allocation Fast Paths

When you call log.info("msg", key=val), Python builds the {"key": val} dict before the function is entered — in the C layer, before any pulselog code runs. At 216k/sec that's 216k dict allocations/sec you cannot avoid with **kwargs syntax.

For calls where you don't need per-record metadata, use the fast-path variants:

log.info_fast("step done")       # ~455k/sec — no dict allocated, ever
log.debug_fast("heartbeat")
log.warning_fast("queue high")
log.error_fast("connection lost")
log.critical_fast("out of memory")

When to use which:

# Tight loop — no metadata needed → use fast path
for step in range(100_000):
    log.info_fast("step")               # 455k/sec

# Need metadata → use standard API
log.info("step", loss=loss, acc=acc)    # 216k/sec — kwargs cost is unavoidable

Checkpoints

Checkpoints persist structured data with progress tracking — ideal for ML training, data pipelines, and long-running jobs.

# Save a checkpoint
log.save(
    name      = "epoch-5",
    data      = {"loss": 0.31, "acc": 0.94},
    status    = "DONE",        # "DONE" | "IN_PROGRESS" | "FAILED" | "SKIPPED"
    note      = "best so far",
    progress  = 50             # 0–100, shown as progress bar in dashboard
)

# Load a checkpoint (returns dict or None — never raises)
result = log.load("epoch-5")
# → {"loss": 0.31, "acc": 0.94, "status": "DONE", "note": "best so far", "progress": 50}

# List all checkpoints (most recent first)
names = log.checkpoints()
# → ["epoch-5", "epoch-4", "epoch-3"]

# Delete a checkpoint
log.delete_checkpoint("epoch-3")

In-memory mode for tests and ephemeral runs:

log = Logger("test", checkpoint_path=":memory:")
log.save("step-1", {"value": 42}, status="DONE", progress=50)
log.load("step-1")  # → works immediately, no disk I/O

Context and Grouping

# Tag groups subsequent logs under a label (per-thread — safe for concurrent use)
log.tag("training")

# Context manager — restores the previous tag on exit, even on exception
with log.context(tag="validation"):
    log.info("val loss", loss=0.41)
# tag is restored here

# Visual divider in the dashboard stream
log.divider("epoch boundary")

Utilities

log.flush(timeout=2.0)   # drain queue synchronously — returns False if timeout hit
log.shutdown()            # graceful teardown (also called automatically on exit)

stats = log.stats()
# {
#   "records_dropped":  int,
#   "drop_rate":        float,   # e.g. 0.04 = 4%
#   "queue_size":       int,
#   "queue_capacity":   int,
#   "queue_fill_pct":   float,
#   "checkpoints_saved": int,
#   "dashboard_clients": int,
#   "uptime_seconds":   float,
# }

🔌 stdlib logging Integration

Drop-in bridge — all structured fields (lineno, filename, funcName, exc_info) are forwarded to the dashboard.

import logging
from pulselog.handler import PulseHandler

logging.getLogger().addHandler(PulseHandler("my-app"))

logging.info("this appears in the dashboard")
logging.error("with traceback", exc_info=True)  # traceback preserved

⚙️ Configuration

Priority (highest → lowest): Logger() kwargs → env vars → pulselog.toml → defaults

PULSELOG_DASHBOARD=false
PULSELOG_HOST=0.0.0.0
PULSELOG_PORT=8080
PULSELOG_AUTO_OPEN=false
PULSELOG_CHECKPOINT_PATH=/data/checkpoints.db
PULSELOG_LEVEL=INFO
PULSELOG_WORKER_INTERVAL=0.01

[pulselog]
host            = "0.0.0.0"
port            = 8080
auto_open       = false
level           = "INFO"
worker_interval = 0.01

🏭 Production Usage

# Disable dashboard, keep checkpoints, log to stderr on drop
log = Logger(
    "prod",
    dashboard        = False,
    checkpoint_path  = "/data/checkpoints.db",
    overflow         = "drop",   # never block — warn on stderr instead
)

With dashboard=False:

• ✅ No threads started beyond the background worker, no port bound, no browser opened
• ✅ Checkpoint reads/writes still work
• ✅ CI environments (CI=true) disable the dashboard automatically

📖 Examples

from pulselog import Logger
import time

log = Logger("resnet-training", checkpoint_path=":memory:")

log.tag("training")
for i in range(10):
    loss = 1.0 - i * 0.08
    acc = 0.6 + i * 0.035

    log.info(f"epoch {i+1}", loss=round(loss, 3), acc=round(acc, 3))

    log.save(
        f"epoch-{i+1}",
        {"loss": loss, "acc": acc},
        status   = "DONE",
        progress = (i + 1) * 10,
    )

    if i > 0 and loss > prev_loss * 1.5:
        log.warning("loss spike", epoch=i+1, loss=loss)

    prev_loss = loss
    time.sleep(0.5)

log.shutdown()

from pulselog import Logger

log = Logger("etl-pipeline")

with log.context("ingestion"):
    log.info("loading source", table="events", rows=1_200_000)
    records = ingest()
    log.info("ingestion complete", rows=len(records))

with log.context("validation"):
    errors = validate(records)
    if errors:
        log.warning("schema errors found", count=len(errors))
    log.save("validation", {"errors": len(errors), "rows": len(records)},
             status="DONE", progress=40)

with log.context("feature_engineering"):
    for feat in ["activity_7d", "churn_score", "ltv_estimate"]:
        features = compute_feature(feat, records)
        log.info("feature computed", name=feat, coverage=features.coverage)
        if features.null_rate > 0.03:
            log.warning("high null rate", feat=feat, null_rate=features.null_rate)

with log.context("warehouse_write"):
    rows_written = write_to_warehouse(features)
    log.info("write complete", rows=rows_written, target="bigquery://features")
    log.save("pipeline_run", {"rows": rows_written, "features": 3},
             status="DONE", progress=100)

log.shutdown()

🧠 Design Notes

📦 Requirements

• Python ≥ 3.8
• websockets ≥ 11.0 (only needed with dashboard=True)

pip install pulselog            # includes websockets

📄 License

MIT