• /
  • EnglishEspañol日本語한국어Português
  • Log inStart now

Send traces from your product

Tip

This procedure is a part of course that teaches you how to build a quickstart. If you haven't already, checkout the course introduction.

Each procedure in this course builds on top of the last one, so make sure you've completed the last procedure, send logs from your product before proceeding with this one.

Traces capture details of a single request as it moves through a system. They're composed of spans, which are data structures that represent individual operations in the flow of execution.

New Relic, provides you a variety of ways to instrument your application to send traces to our Trace API.

In this lesson, you learn to send traces from your product using our telemetry software development kit (SDK).

import os
import random
import datetime
from sys import getsizeof
import psutil
from newrelic_telemetry_sdk import MetricClient, GaugeMetric, CountMetric, SummaryMetric
from newrelic_telemetry_sdk import EventClient, Event
from newrelic_telemetry_sdk import LogClient, Log
metric_client = MetricClient(os.environ["NEW_RELIC_LICENSE_KEY"])
event_client = EventClient(os.environ["NEW_RELIC_LICENSE_KEY"])
log_client = LogClient(os.environ["NEW_RELIC_LICENSE_KEY"])
db = {}
stats = {
"read_response_times": [],
"read_errors": 0,
"read_count": 0,
"create_response_times": [],
"create_errors": 0,
"create_count": 0,
"update_response_times": [],
"update_errors": 0,
"update_count": 0,
"delete_response_times": [],
"delete_errors": 0,
"delete_count": 0,
"cache_hit": 0,
}
last_push = {
"read": datetime.datetime.now(),
"create": datetime.datetime.now(),
"update": datetime.datetime.now(),
"delete": datetime.datetime.now(),
}
def read(key):
print(f"Reading...")
if random.randint(0, 30) > 10:
stats["cache_hit"] += 1
stats["read_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["read_errors"] += 1
stats["read_count"] += 1
try_send("read")
def create(key, value):
print(f"Writing...")
db[key] = value
stats["create_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["create_errors"] += 1
stats["create_count"] += 1
try_send("create")
def update(key, value):
print(f"Updating...")
db[key] = value
stats["update_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["update_errors"] += 1
stats["update_count"] += 1
try_send("update")
def delete(key):
print(f"Deleting...")
db.pop(key, None)
stats["delete_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["delete_errors"] += 1
stats["delete_count"] += 1
try_send("delete")
def try_send(type_):
print("try_send")
now = datetime.datetime.now()
interval_ms = (now - last_push[type_]).total_seconds() * 1000
if interval_ms >= 2000:
send_metrics(type_, interval_ms)
send_event(type_)
send_logs()
def send_metrics(type_, interval_ms):
print("sending metrics...")
keys = GaugeMetric("fdb_keys", len(db))
db_size = GaugeMetric("fdb_size", getsizeof(db))
errors = CountMetric(
name=f"fdb_{type_}_errors",
value=stats[f"{type_}_errors"],
interval_ms=interval_ms
)
cache_hits = CountMetric(
name=f"fdb_cache_hits",
value=stats["cache_hit"],
interval_ms=interval_ms
)
response_times = stats[f"{type_}_response_times"]
response_time_summary = SummaryMetric(
f"fdb_{type_}_responses",
count=len(response_times),
min=min(response_times),
max=max(response_times),
sum=sum(response_times),
interval_ms=interval_ms,
)
batch = [keys, db_size, errors, cache_hits, response_time_summary]
response = metric_client.send_batch(batch)
response.raise_for_status()
print("Sent metrics successfully!")
clear(type_)
def send_event(type_):
print("sending event...")
count = Event(
"fdb_method", {"method": type_}
)
response = event_client.send_batch(count)
response.raise_for_status()
print("Event sent successfully!")
def send_logs():
print("sending log...")
process = psutil.Process(os.getpid())
memory_usage = process.memory_percent()
log = Log("FlashDB is using " + str(round(memory_usage * 100, 2)) + "% memory")
response = log_client.send(log)
response.raise_for_status()
print("Log sent successfully!")
def clear(type_):
stats[f"{type_}_response_times"] = []
stats[f"{type_}_errors"] = 0
stats["cache_hit"] = 0
stats[f"{type_}_count"] = 0
last_push[type_] = datetime.datetime.now()
db.py

Use our SDK

We offer an open source telemetry SDK in several of the most popular programming languages such as Python, Java, Node/TypeScript. These send data to our data ingest APIs, including our Trace API.

In this lesson, you learn how to install and use the Python telemetry SDK to report your first span to New Relic.

Report your first span

Change to the send-traces/flashDB direcrory of the course repository.

bash
$
cd ../../send-traces/flashDB

If you haven't already, install the newrelic-telemetry-sdk package.

bash
$
pip install newrelic-telemetry-sdk

Open db.py file in the IDE of your choice and configure the SpanClient.

import os
import random
import datetime
from sys import getsizeof
import psutil
from newrelic_telemetry_sdk import MetricClient, GaugeMetric, CountMetric, SummaryMetric
from newrelic_telemetry_sdk import EventClient, Event
from newrelic_telemetry_sdk import LogClient, Log
from newrelic_telemetry_sdk import SpanClient
metric_client = MetricClient(os.environ["NEW_RELIC_LICENSE_KEY"])
event_client = EventClient(os.environ["NEW_RELIC_LICENSE_KEY"])
log_client = LogClient(os.environ["NEW_RELIC_LICENSE_KEY"])
span_client = SpanClient(os.environ["NEW_RELIC_LICENSE_KEY"])
db = {}
stats = {
"read_response_times": [],
"read_errors": 0,
"read_count": 0,
"create_response_times": [],
"create_errors": 0,
"create_count": 0,
"update_response_times": [],
"update_errors": 0,
"update_count": 0,
"delete_response_times": [],
"delete_errors": 0,
"delete_count": 0,
"cache_hit": 0,
}
last_push = {
"read": datetime.datetime.now(),
"create": datetime.datetime.now(),
"update": datetime.datetime.now(),
"delete": datetime.datetime.now(),
}
def read(key):
print(f"Reading...")
if random.randint(0, 30) > 10:
stats["cache_hit"] += 1
stats["read_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["read_errors"] += 1
stats["read_count"] += 1
try_send("read")
def create(key, value):
print(f"Writing...")
db[key] = value
stats["create_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["create_errors"] += 1
stats["create_count"] += 1
try_send("create")
def update(key, value):
print(f"Updating...")
db[key] = value
stats["update_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["update_errors"] += 1
stats["update_count"] += 1
try_send("update")
def delete(key):
print(f"Deleting...")
db.pop(key, None)
stats["delete_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["delete_errors"] += 1
stats["delete_count"] += 1
try_send("delete")
def try_send(type_):
print("try_send")
now = datetime.datetime.now()
interval_ms = (now - last_push[type_]).total_seconds() * 1000
if interval_ms >= 2000:
send_metrics(type_, interval_ms)
send_event(type_)
send_logs()
def send_metrics(type_, interval_ms):
print("sending metrics...")
keys = GaugeMetric("fdb_keys", len(db))
db_size = GaugeMetric("fdb_size", getsizeof(db))
errors = CountMetric(
name=f"fdb_{type_}_errors",
value=stats[f"{type_}_errors"],
interval_ms=interval_ms
)
cache_hits = CountMetric(
name=f"fdb_cache_hits",
value=stats["cache_hit"],
interval_ms=interval_ms
)
response_times = stats[f"{type_}_response_times"]
response_time_summary = SummaryMetric(
f"fdb_{type_}_responses",
count=len(response_times),
min=min(response_times),
max=max(response_times),
sum=sum(response_times),
interval_ms=interval_ms,
)
batch = [keys, db_size, errors, cache_hits, response_time_summary]
response = metric_client.send_batch(batch)
response.raise_for_status()
print("Sent metrics successfully!")
clear(type_)
def send_event(type_):
print("sending event...")
count = Event(
"fdb_method", {"method": type_}
)
response = event_client.send_batch(count)
response.raise_for_status()
print("Event sent successfully!")
def send_logs():
print("sending log...")
process = psutil.Process(os.getpid())
memory_usage = process.memory_percent()
log = Log("FlashDB is using " + str(round(memory_usage * 100, 2)) + "% memory")
response = log_client.send(log)
response.raise_for_status()
print("Log sent successfully!")
def clear(type_):
stats[f"{type_}_response_times"] = []
stats[f"{type_}_errors"] = 0
stats["cache_hit"] = 0
stats[f"{type_}_count"] = 0
last_push[type_] = datetime.datetime.now()
db.py

Important

This example expects an environment variable called $NEW_RELIC_LICENSE_KEY.

Instrument your app to report a span to New Relic.

import os
import random
import datetime
from sys import getsizeof
import psutil
import time
from newrelic_telemetry_sdk import MetricClient, GaugeMetric, CountMetric, SummaryMetric
from newrelic_telemetry_sdk import EventClient, Event
from newrelic_telemetry_sdk import LogClient, Log
from newrelic_telemetry_sdk import SpanClient, Span
metric_client = MetricClient(os.environ["NEW_RELIC_LICENSE_KEY"])
event_client = EventClient(os.environ["NEW_RELIC_LICENSE_KEY"])
log_client = LogClient(os.environ["NEW_RELIC_LICENSE_KEY"])
span_client = SpanClient(os.environ["NEW_RELIC_LICENSE_KEY"])
db = {}
stats = {
"read_response_times": [],
"read_errors": 0,
"read_count": 0,
"create_response_times": [],
"create_errors": 0,
"create_count": 0,
"update_response_times": [],
"update_errors": 0,
"update_count": 0,
"delete_response_times": [],
"delete_errors": 0,
"delete_count": 0,
"cache_hit": 0,
}
last_push = {
"read": datetime.datetime.now(),
"create": datetime.datetime.now(),
"update": datetime.datetime.now(),
"delete": datetime.datetime.now(),
}
def read(key):
print(f"Reading...")
if random.randint(0, 30) > 10:
stats["cache_hit"] += 1
stats["read_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["read_errors"] += 1
stats["read_count"] += 1
try_send("read")
def create(key, value):
print(f"Writing...")
db[key] = value
stats["create_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["create_errors"] += 1
stats["create_count"] += 1
try_send("create")
def update(key, value):
print(f"Updating...")
db[key] = value
stats["update_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["update_errors"] += 1
stats["update_count"] += 1
try_send("update")
def delete(key):
print(f"Deleting...")
db.pop(key, None)
stats["delete_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["delete_errors"] += 1
stats["delete_count"] += 1
try_send("delete")
def try_send(type_):
print("try_send")
now = datetime.datetime.now()
interval_ms = (now - last_push[type_]).total_seconds() * 1000
if interval_ms >= 2000:
send_metrics(type_, interval_ms)
send_event(type_)
send_logs()
def send_metrics(type_, interval_ms):
print("sending metrics...")
keys = GaugeMetric("fdb_keys", len(db))
db_size = GaugeMetric("fdb_size", getsizeof(db))
errors = CountMetric(
name=f"fdb_{type_}_errors",
value=stats[f"{type_}_errors"],
interval_ms=interval_ms
)
cache_hits = CountMetric(
name=f"fdb_cache_hits",
value=stats["cache_hit"],
interval_ms=interval_ms
)
response_times = stats[f"{type_}_response_times"]
response_time_summary = SummaryMetric(
f"fdb_{type_}_responses",
count=len(response_times),
min=min(response_times),
max=max(response_times),
sum=sum(response_times),
interval_ms=interval_ms,
)
batch = [keys, db_size, errors, cache_hits, response_time_summary]
response = metric_client.send_batch(batch)
response.raise_for_status()
print("Sent metrics successfully!")
clear(type_)
def send_event(type_):
print("sending event...")
count = Event(
"fdb_method", {"method": type_}
)
response = event_client.send_batch(count)
response.raise_for_status()
print("Event sent successfully!")
def send_logs():
print("sending log...")
process = psutil.Process(os.getpid())
memory_usage = process.memory_percent()
log = Log("FlashDB is using " + str(round(memory_usage * 100, 2)) + "% memory")
response = log_client.send(log)
response.raise_for_status()
print("Log sent successfully!")
def send_spans():
print("sending span...")
with Span(name="sleep") as span:
time.sleep(0.5)
response = span_client.send(span)
response.raise_for_status()
print("Span sleep sent successfully!")
def clear(type_):
stats[f"{type_}_response_times"] = []
stats[f"{type_}_errors"] = 0
stats["cache_hit"] = 0
stats[f"{type_}_count"] = 0
last_push[type_] = datetime.datetime.now()
db.py

Here, you instrument your platform to send a simple sleep span to New Relic.

Amend the try_send module to send the span every 2 second.

import os
import random
import datetime
from sys import getsizeof
import psutil
import time
from newrelic_telemetry_sdk import MetricClient, GaugeMetric, CountMetric, SummaryMetric
from newrelic_telemetry_sdk import EventClient, Event
from newrelic_telemetry_sdk import LogClient, Log
from newrelic_telemetry_sdk import SpanClient, Span
metric_client = MetricClient(os.environ["NEW_RELIC_LICENSE_KEY"])
event_client = EventClient(os.environ["NEW_RELIC_LICENSE_KEY"])
log_client = LogClient(os.environ["NEW_RELIC_LICENSE_KEY"])
span_client = SpanClient(os.environ["NEW_RELIC_LICENSE_KEY"])
db = {}
stats = {
"read_response_times": [],
"read_errors": 0,
"read_count": 0,
"create_response_times": [],
"create_errors": 0,
"create_count": 0,
"update_response_times": [],
"update_errors": 0,
"update_count": 0,
"delete_response_times": [],
"delete_errors": 0,
"delete_count": 0,
"cache_hit": 0,
}
last_push = {
"read": datetime.datetime.now(),
"create": datetime.datetime.now(),
"update": datetime.datetime.now(),
"delete": datetime.datetime.now(),
}
def read(key):
print(f"Reading...")
if random.randint(0, 30) > 10:
stats["cache_hit"] += 1
stats["read_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["read_errors"] += 1
stats["read_count"] += 1
try_send("read")
def create(key, value):
print(f"Writing...")
db[key] = value
stats["create_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["create_errors"] += 1
stats["create_count"] += 1
try_send("create")
def update(key, value):
print(f"Updating...")
db[key] = value
stats["update_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["update_errors"] += 1
stats["update_count"] += 1
try_send("update")
def delete(key):
print(f"Deleting...")
db.pop(key, None)
stats["delete_response_times"].append(random.uniform(0.5, 1.0))
if random.choice([True, False]):
stats["delete_errors"] += 1
stats["delete_count"] += 1
try_send("delete")
def try_send(type_):
print("try_send")
now = datetime.datetime.now()
interval_ms = (now - last_push[type_]).total_seconds() * 1000
if interval_ms >= 2000:
send_metrics(type_, interval_ms)
send_event(type_)
send_logs()
send_spans()
def send_metrics(type_, interval_ms):
print("sending metrics...")
keys = GaugeMetric("fdb_keys", len(db))
db_size = GaugeMetric("fdb_size", getsizeof(db))
errors = CountMetric(
name=f"fdb_{type_}_errors",
value=stats[f"{type_}_errors"],
interval_ms=interval_ms
)
cache_hits = CountMetric(
name=f"fdb_cache_hits",
value=stats["cache_hit"],
interval_ms=interval_ms
)
response_times = stats[f"{type_}_response_times"]
response_time_summary = SummaryMetric(
f"fdb_{type_}_responses",
count=len(response_times),
min=min(response_times),
max=max(response_times),
sum=sum(response_times),
interval_ms=interval_ms,
)
batch = [keys, db_size, errors, cache_hits, response_time_summary]
response = metric_client.send_batch(batch)
response.raise_for_status()
print("Sent metrics successfully!")
clear(type_)
def send_event(type_):
print("sending event...")
count = Event(
"fdb_method", {"method": type_}
)
response = event_client.send_batch(count)
response.raise_for_status()
print("Event sent successfully!")
def send_logs():
print("sending log...")
process = psutil.Process(os.getpid())
memory_usage = process.memory_percent()
log = Log("FlashDB is using " + str(round(memory_usage * 100, 2)) + "% memory")
response = log_client.send(log)
response.raise_for_status()
print("Log sent successfully!")
def send_spans():
print("sending span...")
with Span(name="sleep") as span:
time.sleep(0.5)
response = span_client.send(span)
response.raise_for_status()
print("Span sleep sent successfully!")
def clear(type_):
stats[f"{type_}_response_times"] = []
stats[f"{type_}_errors"] = 0
stats["cache_hit"] = 0
stats[f"{type_}_count"] = 0
last_push[type_] = datetime.datetime.now()
db.py

Your platform will now report this span every 2 seconds.

Navigate to the root of your application at build-a-quickstart-lab/send-traces/flashDB.

Run your services to verify that it is reporting the span.

bash
$
python simulator.py
Writing...
try_send
Reading...
try_send
Reading...
try_send
Writing...
try_send
Writing...
try_send
Reading...
sending metrics...
Sent metrics successfully!
sending event...
Event sent successfully!
sending log...
Log sent successfully!
sending span...
Span sleep sent successfully!

Alternative Options

If the language SDK doesn't fit your needs, try out one of our other options:

  • Existing Zipkin instrumentation: if you've an existing Zipkin implementation, you can simply change the endpoint to New Relic to report your data. Read our documentation to report data from existing Zipkin instrumentation.

  • Manual Implementation: If the previous options don't fit your requirements, you can always manually instrument your own library to make a POST request to the New Relic Trace API.

Your platform is now reporting data to New Relic. Next, you observe this data in New Relic using dashboard.

Tip

This procedure is a part of course that teaches you how to build a quickstart. Continue to next lesson, create a dashboard.

Copyright © 2024 New Relic Inc.

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.