netflix-spectator-py 0.2.2

Python library for reporting metrics to the Netflix Atlas Timeseries Database.

Introduction

Python thin-client metrics library for use with Atlas and SpectatorD.

Supports Python >= 3.5. This version is chosen as the baseline, because it is the oldest system Python available in our operating environments.

Local Development

Install pyenv, possibly with Homebrew, and install a recent Python version.

make setup-venv
make test
make coverage

Usage

Installing

Install this library for your project as follows:

pip3 install netflix-spectator-py

Publishing metrics requires a SpectatorD process running on your instance.

Importing

Standard Usage

At Netflix, your initialization script should load the environment, to ensure that the standard variables are available to the Python application.

source /etc/nflx/environment

Importing the GlobalRegistry instantiates a Registry with a default configuration that applies process-specific common tags based on environment variables and opens a socket to the SpectatorD sidecar. The remainder of the instance-specific common tags are provided by SpectatorD.

from spectator import GlobalRegistry

Once the GlobalRegistry is imported, it is used to create and manage Meters.

Logging

This package provides the following loggers:

• spectator.SidecarWriter

When troubleshooting metrics collection and reporting, you should set the SidecarWriter logging to the DEBUG level, before the first metric is recorded. For example:

import logging

# record the human-readable time, name of the logger, logging level, thread id and message
logging.basicConfig(
    level=logging.DEBUG,
    format='%(asctime)s - %(name)s - %(levelname)s - %(thread)d - %(message)s'
)

logging.getLogger('spectator.SidecarWriter').setLevel(logging.DEBUG)

There is approximately a 10% performance penalty in UDP write performance when debug logging is enabled. It may be more, depending on the exact logging configuration (i.e. flushing to slow disk).

Working with IDs

The IDs used for identifying a meter in the GlobalRegistry consist of a name and a set of tags. IDs will be consumed by users many times after the data has been reported, so they should be chosen thoughtfully, while considering how they will be used. See the naming conventions page for general guidelines.

IDs are immutable, so they can be freely passed around and used in a concurrent context. Tags can be added to an ID when it is created, to track the dimensionality of the metric. All tag keys and values must be strings. For example, if you want to keep track of the number of successful requests, you must cast integers to strings.

from spectator import GlobalRegistry

requests_id = GlobalRegistry.counter("server.numRequests", {"statusCode": str(200)})
requests_id.increment()

Meter Types

Age Gauges

The value is the time in seconds since the epoch at which an event has successfully occurred, or 0 to use the current time in epoch seconds. After an Age Gauge has been set, it will continue reporting the number of seconds since the last time recorded, for as long as the SpectatorD process runs. The purpose of this metric type is to enable users to more easily implement the Time Since Last Success alerting pattern.

To set a specific time as the last success:

from spectator import GlobalRegistry

GlobalRegistry.age_gauge("time.sinceLastSuccess").set(1611081000)

To set now() as the last success:

from spectator import GlobalRegistry

GlobalRegistry.age_gauge("time.sinceLastSuccess").set(0)

By default, a maximum of 1000 Age Gauges are allowed per spectatord process, because there is no mechanism for cleaning them up. This value may be tuned with the --age_gauge_limit flag on the spectatord binary.

Counters

A Counter is used to measure the rate at which an event is occurring. Considering an API endpoint, a Counter could be used to measure the rate at which it is being accessed.

Counters are reported to the backend as a rate-per-second. In Atlas, the :per-step operator can be used to convert them back into a value-per-step on a graph.

Call increment() when an event occurs:

from spectator import GlobalRegistry

GlobalRegistry.counter("server.numRequests").increment()

You can also pass a value to increment(). This is useful when a collection of events happens together:

from spectator import GlobalRegistry

GlobalRegistry.counter("queue.itemsAdded").increment(10)

Distribution Summaries

A Distribution Summary is used to track the distribution of events. It is similar to a Timer, but more general, in that the size does not have to be a period of time. For example, a Distribution Summary could be used to measure the payload sizes of requests hitting a server.

Always use base units when recording data, to ensure that the tick labels presented on Atlas graphs are readable. If you are measuring payload size, then use bytes, not kilobytes (or some other unit). This means that a 4K tick label will represent 4 kilobytes, rather than 4 kilo-kilobytes.

Call record() with a value:

from spectator import GlobalRegistry

GlobalRegistry.distribution_summary("server.requestSize").record(10)

Percentile Distribution Summaries

The value tracks the distribution of events, with percentile estimates. It is similar to a Percentile Timer, but more general, because the size does not have to be a period of time.

For example, it can be used to measure the payload sizes of requests hitting a server or the number of records returned from a query.

In order to maintain the data distribution, they have a higher storage cost, with a worst-case of up to 300X that of a standard Distribution Summary. Be diligent about any additional dimensions added to Percentile Distribution Summaries and ensure that they have a small bounded cardinality.

Call record() with a value:

from spectator import GlobalRegistry

GlobalRegistry.pct_distribution_summary("server.requestSize").record(10)

Gauges

A gauge is a value that is sampled at some point in time. Typical examples for gauges would be the size of a queue or number of threads in a running state. Since gauges are not updated inline when a state change occurs, there is no information about what might have occurred between samples.

Consider monitoring the behavior of a queue of tasks. If the data is being collected once a minute, then a gauge for the size will show the size when it was sampled. The size may have been much higher or lower at some point during interval, but that is not known.

Call set() with a value:

from spectator import GlobalRegistry

GlobalRegistry.gauge("server.queueSize").set(10)

Gauges will report the last set value for 15 minutes. This done so that updates to the values do not need to be collected on a tight 1-minute schedule to ensure that Atlas shows unbroken lines in graphs. A custom TTL, may be configured for gauges. SpectatorD enforces a minimum TTL of 5 seconds.

from spectator import GlobalRegistry

GlobalRegistry.gauge("server.queueSize", ttl_seconds=120).set(10)

Timers

A Timer is used to measure how long (in seconds) some event is taking.

Call record() with a value:

from spectator import GlobalRegistry

GlobalRegistry.timer("server.requestLatency").record(0.01)

A stopwatch() method is available which may be used as a Context Manager to automatically record the number of seconds that have elapsed while executing a block of code:

import time
from spectator import GlobalRegistry

t = GlobalRegistry.timer("thread.sleep")

with t.stopwatch():
    time.sleep(5)

Internally, Timers will keep track of the following statistics as they are used:

• count
• totalTime
• totalOfSquares
• max

Percentile Timers

The value is the number of seconds that have elapsed for an event, with percentile estimates.

This metric type will track the data distribution by maintaining a set of Counters. The distribution can then be used on the server side to estimate percentiles, while still allowing for arbitrary slicing and dicing based on dimensions.

In order to maintain the data distribution, they have a higher storage cost, with a worst-case of up to 300X that of a standard Timer. Be diligent about any additional dimensions added to Percentile Timers and ensure that they have a small bounded cardinality.

Call record() with a value:

from spectator import GlobalRegistry

GlobalRegistry.pct_timer("server.requestLatency").record(0.01)

A stopwatch() method is available which may be used as a Context Manager to automatically record the number of seconds that have elapsed while executing a block of code:

import time
from spectator import GlobalRegistry

t = GlobalRegistry.pct_timer("thread.sleep")

with t.stopwatch():
    time.sleep(5)

Writing Tests

To write tests against this library, instantiate a test instance of the Registry and configure it to use the MemoryWriter, which stores all updates in a List. Inspect the last_line() or scan all _messages to verify your metrics updates.

import unittest

from spectator import Registry
from spectator.sidecarconfig import SidecarConfig

class MetricsTest(unittest.TestCase):

    def test_counter(self):
        r = Registry(config=SidecarConfig({"sidecar.output-location": "memory"}))

        c = r.counter("test")
        self.assertTrue(c._writer.is_empty())

        c.increment()
        self.assertEqual("c:test:1", c._writer.last_line())

If you need to keep track of the metric classes, you can do that as follows:

from spectator.counter import Counter, MonotonicCounter
from spectator.distsummary import DistributionSummary
from spectator.gauge import AgeGauge, Gauge, MaxGauge
from spectator.timer import Timer

# https://github.com/Netflix-Skunkworks/spectatord#metric-types
_METRIC_CLASS_MAP = {
    'c': Counter,
    'd': DistributionSummary,
    'g': Gauge,
    'm': MaxGauge,
    't': Timer,
    'A': AgeGauge,
    'C': MonotonicCounter,
    'D': DistributionSummary,
    'T': Timer
}

If you need to override the default output location (udp) of the GlobalRegistry, then you can set a SPECTATOR_OUTPUT_LOCATION environment variable to one of the following values:

• none - disable output
• memory - write to memory (SidecarWriter internal list _messages)
• stdout - write to standard out for the process
• stderr - write to standard error for the process
• file://$path_to_file - write to a file (e.g. file:///tmp/foo/bar)
• udp://$host:$port - write to a UDP socket

Migrating from 0.1.X to 0.2.X

• This library no longer publishes directly to the Atlas backends. It now publishes to the SpectatorD sidecar which is bundled with all standard AMIs and containers. If you must have the previous direct publishing behavior, because SpectatorD is not yet available on the platform where your code runs, then you can pin to version 0.1.18.

• The internal Netflix configuration companion library is no longer required and this dependency may be dropped from your project.

• The API surface area remains unchanged to avoid breaking library consumers, and standard uses of GlobalRegistry helper methods for publishing metrics continue to work as expected. Several helper methods on meter classes are now no-ops, always returning values such as 0 or nan. If you want to write tests to validate metrics publication, take a look at the tests in this library for a few examples of how that can be done. The core idea is to capture the lines which will be written out to SpectatorD.

• Replace uses of PercentileDistributionSummary with direct use of the Registry pct_distribution_summary method.

  # before
  from spectator import GlobalRegistry
  from spectator.histogram import PercentileDistributionSummary
  
  d = PercentileDistributionSummary(GlobalRegistry, "server.requestSize")
  d.record(10)
  

  # after
  from spectator import GlobalRegistry
  
  GlobalRegistry.pct_distribution_summary("server.requestSize").record(10)
  

• Replace uses of PercentileTimer with direct use of the Registry pct_timer method.

  # before
  from spectator import GlobalRegistry
  from spectator.histogram import PercentileTimer
  
  t = PercentileTimer(GlobalRegistry, "server.requestSize")
  t.record(0.01)
  

  # after
  from spectator import GlobalRegistry
  
  GlobalRegistry.pct_timer("server.requestSize").record(0.1)
  

• Implemented new meter types supported by SpectatorD: age_gauge, max_gauge and monotonic_counter. See the SpectatorD documentation or the class docstrings for more details.