flowwatch-client 2.1.3

Python client SDK for FlowWatch — feature flags, durable workflows, tracing, and error capture

Python client SDK for FlowWatch — durable workflows, feature flags, request tracing, and error capture.

---

Before you start: the sidecar is required

This package is a thin HTTP client. It does not connect to any database and does not run any background processes. All the actual logic lives in the FlowWatch Node.js sidecar, which must be running and reachable before any method in this SDK will work.

What this package does:

• Wraps the five sidecar REST endpoints in a clean Python API
• Handles auth headers, JSON serialization, and HTTP errors for you
• Provides both sync (FlowwatchClient) and async (AsyncFlowwatchClient) variants

What this package does NOT do:

• Does not connect to Postgres or Redis
• Does not run workflows, evaluate flags, or store traces by itself
• Does not start the sidecar — you must start it separately

---

Step 1 — Start the FlowWatch sidecar

The sidecar is a Node.js process. In your infrastructure (Docker container, separate server, or just another terminal):

// sidecar.ts  — requires Node.js and @pranshulsoni/flowwatch installed
import { createFlowwatch, startSidecar } from "@pranshulsoni/flowwatch";

const fw = await createFlowwatch({
  db: { connectionString: process.env.DATABASE_URL },
  redis: { url: process.env.REDIS_URL },
  migrations: { autoRun: true },
});

startSidecar(fw, { port: 9400, token: process.env.SIDECAR_TOKEN });
// API available at http://localhost:9400/api/*

The sidecar handles Postgres, Redis, the workflow queue, and the dashboard. Your Python app talks to it over HTTP.

---

Step 2 — Install the Python SDK

pip install flowwatch-client

Requires Python 3.8+. The only dependency is httpx.

---

Step 3 — Use it

from flowwatch import FlowwatchClient

client = FlowwatchClient(
    base_url="http://localhost:9400",
    token="your-sidecar-token",   # must match the token set in startSidecar
)

---

Authentication

If your sidecar was started with a token, every request must include it. Pass it to the client constructor:

client = FlowwatchClient("http://localhost:9400", token="my-secret-token")

The SDK sends it automatically as Authorization: Bearer <token> on every request. If your sidecar has no token (dev only), omit the token argument.

---

API Reference

evaluate_flag(key, context=None) → bool

Returns True if the feature flag is enabled for the given user context. The flag rules and percentage rollout are evaluated by the sidecar, not this SDK.

enabled = client.evaluate_flag("new-checkout-flow", {
    "userId": "user_123",
    "plan": "pro",
    "email": "[e-mail]",
})
if enabled:
    use_new_checkout()

context is optional. Omit it for flags with no targeting rules.

---

trigger_workflow(name, input=None) → dict

Starts a durable workflow execution. The workflow steps run in the Node.js process — this call just enqueues the job and returns immediately.

result = client.trigger_workflow("order-fulfillment", {
    "orderId": "order_456",
    "userId": "user_123",
})
print(result["executionId"])   # track progress in the dashboard

The workflow must be registered in the Node.js sidecar with fw.workflow(name, steps).

---

log_trace_span(name, type, duration_ms, status="ok", metadata=None)

Submits a pre-timed performance span. You are responsible for measuring the duration.

import time

start = time.monotonic()
rows = db.execute("SELECT * FROM products")
duration_ms = (time.monotonic() - start) * 1000

client.log_trace_span(
    name="fetch-products",
    type="database",
    duration_ms=duration_ms,
    metadata={"rowCount": len(rows)},
)

type can be any string: "database", "http", "function", "cache", etc. It's used for grouping in the dashboard.

---

trace_span(name, type, metadata=None) — context manager

Auto-times the wrapped block and calls log_trace_span on exit. Sets status="error" automatically if an exception is raised.

with client.trace_span("send-email", "http", metadata={"to": "[e-mail]"}):
    send_email(to="[e-mail]", subject="Your order shipped")

---

capture_error(message, name="Error", stack=None, source=None, **options)

Reports an error to FlowWatch. Use in except blocks to get stack traces and context visible in the dashboard.

import traceback

try:
    process_payment(order)
except Exception as e:
    client.capture_error(
        message=str(e),
        name=type(e).__name__,
        stack=traceback.format_exc(),
        source="python-payment-service",
        category="payment",
        level="error",
    )
    raise

Do not swallow the exception unless you intend to — capture_error just records it, it does not re-raise.

---

health() → dict

Checks whether the sidecar is reachable and healthy. Returns the sidecar's status object.

status = client.health()
print(status)
# {"status": "ok", "postgres": "connected", "redis": "connected", ...}

Use this for readiness checks or startup validation.

---

Async variant

All methods above are available on AsyncFlowwatchClient with await:

import asyncio
from flowwatch import AsyncFlowwatchClient

async def main():
    async with AsyncFlowwatchClient("http://localhost:9400", token="secret") as client:
        enabled = await client.evaluate_flag("dark-mode", {"userId": "u1"})
        result = await client.trigger_workflow("send-welcome-email", {"userId": "u1"})
        status = await client.health()

asyncio.run(main())

AsyncFlowwatchClient uses httpx.AsyncClient under the hood. Use async with or call aclose() manually to release the connection pool.

---

Deployment

The sidecar and your Python app are separate processes. A common setup:

# docker-compose.yml
services:
  flowwatch-sidecar:
    image: node:20
    command: node sidecar.js
    environment:
      - DATABASE_URL=postgres://...
      - SIDECAR_TOKEN=my-secret-token
    ports:
      - "9400:9400"

  my-python-app:
    build: .
    environment:
      - FLOWWATCH_URL=http://flowwatch-sidecar:9400
      - SIDECAR_TOKEN=my-secret-token

import os
client = FlowwatchClient(os.environ["FLOWWATCH_URL"], token=os.environ["SIDECAR_TOKEN"])

---

Error handling

Every method raises httpx.HTTPStatusError on a non-2xx response. Wrap calls that can fail:

from httpx import HTTPStatusError

try:
    client.log_trace_span("slow-query", "database", 450.0)
except HTTPStatusError as e:
    print(f"FlowWatch sidecar returned {e.response.status_code}")
except Exception:
    pass  # sidecar unreachable — decide whether this should fail your app

FlowWatch is observability infrastructure. A down sidecar should generally not crash your application — catch and swallow errors from this SDK at the call sites that are not critical.