LangSmith — tracing not appearing in the dashboard

# In .env or shell (BEFORE running Python)
export LANGCHAIN_TRACING_V2=true
export LANGCHAIN_API_KEY=ls_your_key_here
export LANGCHAIN_PROJECT=myapp-prod
# Self-hosted only:
# export LANGCHAIN_ENDPOINT=https://langsmith.internal.example.com

# Verify from shell before running Python
env | grep LANGCHAIN

# In Python — load .env BEFORE importing langchain
from dotenv import load_dotenv
load_dotenv()   # this must happen FIRST

# ONLY THEN import LangChain
from langchain_openai import ChatOpenAI
from langchain_core.prompts import ChatPromptTemplate

# Verify at runtime
import os
assert os.environ.get("LANGCHAIN_TRACING_V2") == "true", "Tracing not enabled"
assert os.environ.get("LANGCHAIN_API_KEY"), "API key not set"
print(f"Tracing to project: {os.environ.get('LANGCHAIN_PROJECT', 'default')}")

# For docker/systemd, set them in the environment section of the unit
# For k8s, use envFrom + a Secret

from langchain_core.tracers.context import tracing_v2_enabled
from langchain_openai import ChatOpenAI

llm = ChatOpenAI(model="gpt-4o-mini")

# Default project: whatever LANGCHAIN_PROJECT is set to
result = llm.invoke("hello")

# Per-block override — route these traces to a specific project
with tracing_v2_enabled(project_name="experiments-quality"):
    result = llm.invoke("hello")

# Per-invoke override via config
result = llm.invoke("hello", config={"run_name": "smoke-test",
                                     "tags": ["smoke", "prod"],
                                     "metadata": {"user_id": "u_123"}})

# Environment-based projects
import os
os.environ["LANGCHAIN_PROJECT"] = {
    "development": "myapp-dev",
    "staging": "myapp-staging",
    "production": "myapp-prod",
}[os.environ.get("APP_ENV", "development")]

# For agents / chains, tag runs to make filtering easy in LangSmith UI
from langchain.agents import AgentExecutor
executor.invoke(
    {"input": "..."},
    config={
        "tags": ["agent-v2", os.environ.get("APP_ENV")],
        "metadata": {"user_id": "u_123", "conversation_id": "c_456"},
    },
)

# Option A: force synchronous callbacks
import os
os.environ["LANGCHAIN_CALLBACKS_BACKGROUND"] = "false"  # every trace flushed immediately

# Option B: explicit flush before exit (recommended for serverless)
from langsmith import Client

def flush_traces():
    try:
        client = Client()
        client.flush()   # blocks until pending traces upload
    except Exception as e:
        print(f"LangSmith flush failed: {e}")

# For AWS Lambda handler
def lambda_handler(event, context):
    try:
        response = my_langchain_chain.invoke(event)
        return {"statusCode": 200, "body": response}
    finally:
        flush_traces()   # ensure traces upload before Lambda freezes the container

# For CLI tools with atexit
import atexit
atexit.register(flush_traces)

# For long-running services (FastAPI, Django) — do NOT force sync
# The background sender is fine and adds no latency to your requests