Server Mode

Lynguine Server Mode provides fast repeated access by keeping lynguine loaded in memory, avoiding the ~2 second startup cost for each call.

Performance: 156.3x speedup for repeated operations (1.532s → 9.8ms)

Note

Server mode is designed for applications that call lynguine multiple times via subprocesses (like lamd). It is not needed for long-running processes like Jupyter kernels (referia).

Overview

When to Use Server Mode:

✅ Multiple lynguine calls (loops, batch processing)

✅ Subprocess-based applications (like lamd)

✅ Slow startup times (~2s per call)

✅ Single-user, local development

Benefits:

156.3x speedup for repeated calls
Auto-start: Zero setup required
Auto-restart: Crash recovery
Auto-shutdown: Idle timeout
Retry logic: Network resilience

Quick Start

from lynguine.client import ServerClient

# Zero-setup client (auto-starts server)
client = ServerClient(
    auto_start=True,        # Starts server automatically
    idle_timeout=300,       # 5-minute timeout
    max_retries=3,          # Retry on failures
    retry_delay=1.0         # Exponential backoff
)

# Use just like direct mode
df = client.read_data(interface_file='config.yml')
client.close()

Contents

Performance

Scenario	Direct Mode	Server (First)	Server (Subsequent)	Speedup
Single call	1.947s	1.947s	–	1x
10 calls	19.47s	1.947s + 98ms	98ms	198x
100 calls	194.7s	1.947s + 980ms	980ms	67x

Implementation Phases

Phase 1: Proof of Concept ✅: Basic server/client, instance checking, benchmarks
Phase 2: Core Features ✅: New endpoints, idle timeout, auto-start, diagnostics
Phase 3: Robustness ✅: Retry logic, crash recovery, comprehensive documentation
Phase 4: Remote Access (optional): Authentication, TLS, multi-user support