A troposphere-inspired library
for programmatic, declarative definition and management of SignalFx Charts,
Dashboards, and Detectors.
This library assumes a basic familiarity with resources in SignalFx. For a good overview of the SignalFx API consult the upstream documentation.
- Features
- Installation
- Usage
- Building Charts
- Building Dashboards
- Updating Dashboards
- Dashboard Filters
- Dashboard Event Overlays
- Creating Detectors
- Using Flow and Combinator Functions In Formulas
- Building Dashboard Groups
- Updating Dashboard Group
- Talking to the SignalFlow API Directly
- General
ResourceGuidelines - Creating a CLI for your resources
- Contributing
- Provides bindings for the SignalFlow DSL
- Provides abstractions for:
- Charts
- Dashboards, DashboardGroups
- Detectors
- A CLI builder to wrap resource definitions (useful for automation)
Add signal_analog to the requirements file in your project:
# requirements.txt
# ... your other dependencies
signal_analog
Then run the following command to update your environment:
pip install -r requirements.txt
signal_analog provides two kinds of abstractions, one for building resources
in the SignalFx API and the other for describing metric timeseries through the
Signal Flow DSL.
The following sections describe how to use Resource abstractions in
conjunction with the Signal Flow DSL.
signal_analog provides constructs for building charts in the
signal_analog.charts module.
Consult the upstream documentation for more information Charts.
Let's consider an example where we would like to build a chart to monitor memory utilization for a single applicaton in a single environment.
This assumes a service reports metrics for application name as app and
environment as env with memory utilization reporting via the
memory.utilization metric name.
In a timeseries chart, all data displayed on the screen comes from at least one
data definition in the SignalFlow language. Let's begin by defining our
timeseries:
from signal_analog.flow import Data
ts = Data('memory.utilization')In SignalFlow parlance a timeseries is only displayed on a chart if it has been
"published". All stream functions in SignalFlow have a publish method that
may be called at the end of all timeseries transformations.
ts = Data('memory.utilization').publish()As a convenience, all transformations on stream functions return the callee,
so in the above example ts remains bound to an instance of Data.
Now, this timeseries isn't very useful by itself; if we attached this program to a chart we would see all timeseries for all Riposte applications reporting to SignalFx!
We can restrict our view of the data by adding a filter on application name:
from signal_analog.flow import Data, Filter
app_filter = Filter('app', 'foo')
ts = Data('memory.utilization', filter=app_filter).publish()Now if we created a chart with this program we would only be looking at metrics
that relate to the foo application. Much better, but we're still
looking at instance of foo regardless of the environment it
lives in.
What we'll want to do is combine our app_filter with another filter for the
environment. The signal_analog.combinators module provides some helpful
constructs for achieving this goal:
from signal_analog.combinators import And
env_filter = Filter('env', 'prod')
all_filters = And(app_filter, env_filter)
ts = Data('memory.utilization', filter=all_filters).publish()Excellent! We're now ready to create our chart.
First, let's give our chart a name:
from signal_analog.charts import TimeSeriesChart
memory_chart = TimeSeriesChart().with_name('Memory Used %')Like it's flow counterparts, charts adhere to the builder pattern for
constructing objects that interact with the SignalFx API.
With our name in place, let's go ahead and add our program:
memory_chart = TimeSeriesChart().with_name('Memory Used %').with_program(ts)Each Chart understands how to serialize our SignalFlow programs appropriately, so it is sufficient to simply pass in our reference here.
Finally, let's change the plot type on our chart so that we see solid areas instead of flimsy lines:
from signal_analog.charts import PlotType
memory_chart = TimeSeriesChart()\
.with_name('Memory Used %')\
.with_program(ts)
.with_default_plot_type(PlotType.area_chart)Terrific; there's only a few more details before we have a complete chart.
In the following sections we'll see how we can create dashboards from collections of charts.
signal_analog provides constructs for building charts in the
signal_analog.dashboards module.
Consult the upstream documentation for more information on the Dashboard API.
Building on the examples described in the previous section, we'd now like to build a dashboard containing our memory chart.
We start with the humble Dashboard object:
from signal_analog.dashboards import Dashboard
dash = Dashboard()Many of the same methods for charts are available on dashboards as well, so let's give our dashboard a memorable name and configure it's API token:
dash.with_name('My Little Dashboard: Metrics are Magic')\
.with_api_token('my-api-token')Our final task will be to add charts to our dashboard and create it in the API!
response = dash\
.with_charts(memory_chart)\
.with_api_token('my-api-token')\
.create()At this point one of two things will happen:
- We receive some sort of error from the SignalFx API and an exception is thrown
- We successfully created the dashboard, in which case the JSON response is returned as a dictionary.
Now, storing API keys in source isn't ideal, so if you'd like to see how you can pass in your API keys at runtime check the documentation below to see how you can dynamically build a CLI for your resources.
Once you have created a dashboard you can update properties like name and description:
dash.update(
name='updated_dashboard_name',
description='updated_dashboard_description'
)Dashboard updates will also update any Chart configurations it owns.
Note: If the given dashboard does not already exist, `update` will create a new dashboard for you
Dashboards can be configured to provide various filters that affect the behavior of all configured charts (overriding any conflicting filters at the chart level). You may wish to do this in order to quickly change the environment that you're observing for a given set of charts.
from signal_analog.filters import DashboardFilters, FilterVariable, FilterSource, FilterTime
app_var = FilterVariable().with_alias('app')\
.with_property('app')\
.with_is_required(True)\
.with_value('foo')
env_var = FilterVariable().with_alias('env')\
.with_property('env')\
.with_is_required(True)\
.with_value('prod')
aws_src = FilterSource().with_property("aws_region").with_value('us-west-2')
time = FilterTime().with_start("-1h").with_end("Now")
app_filter = DashboardFilters() \
.with_variables(app_var, env_var) \
.with_sources(aws_src) \
.with_time(time)So, here we are creating a few filters "app=foo" and "env=prod", a source filter "aws_region=us-west-2" and a time filter "-1h till Now" Now we can pass this config to a dashboard object:
response = dash\
.with_charts(memory_chart)\
.with_api_token('my-api-token')\
.with_filters(app_filter)\
.create()If you are updating an existing dashboard:
response = dash\
.with_filters(app_filter)\
.update()To view events overlayed on your charts within a dashboard requires an event to be viewed, a chart with showEventLines enabled, and a dashboard with the correct eventOverlays settings (and selectedEventOverlays to show events by default).
Assuming that the events you would like to see exist; you would make a chart with showEventLines like so:
from signal_analog.flow import Data
from signal_analog.charts import TimeSeriesChart
program = Data('cpu.utilization').publish()
chart = TimeSeriesChart().with_name('Chart With Event Overlays')\
.with_program(program).show_event_lines(True)With our chart defined, we are ready to prepare our event overlays and selected event overlays for the dashboard. First we define the event signals we would like to match. In this case, we will look for an event named "test" (include leading and/or trailing asterisks as wildcards if you need partial matching). Next we use those event signals to create our eventOverlays, making sure to include a color index for our event's symbol, and setting event line to True. We also pass our event signals along to the selectedEventOverlays, which will tell the dashboard to display matching events by default.
from signal_analog.eventoverlays import EventSignals, EventOverlays, SelectedEventOverlays
events = EventSignals().with_event_search_text("*test*")\
.with_event_type("eventTimeSeries")
eventoverlay = EventOverlays().with_event_signals(events)\
.with_event_color_index(1)\
.with_event_line(True)
selectedeventoverlay = SelectedEventOverlays()\
.with_event_signals(events)Next we combine our chart, our event overlay, and our selected event overlay into a dashboard object:
from signal_analog.dashboards import Dashboard
dashboard_with_event_overlays = Dashboard().with_name('Dashboard With Overlays')\
.with_charts(chart)\
.with_event_overlay(eventoverlay)\
.with_selected_event_overlay(selectedeventoverlay)Finally we build our resources in SignalFX with the cli builder:
if __name__ == '__main__':
from signal_analog.cli import CliBuilder
cli = CliBuilder().with_resources(dashboard_with_event_overlays)\
.build()
cli()signal_analog provides a means of managing the lifecycle of Detectors in
the signal_analog.detectors module. As of v0.21.0 only a subset of
the full Detector API is supported.
Consult the upstream documentation for more information about Detectors.
Detectors are comprised of a few key elements:
- A name
- A SignalFlow Program
- A set of rules for alerting
We start by building a Detector object and giving it a name:
from signal_analog.detectors import Detector
detector = Detector().with_name('My Super Serious Detector')We'll now need to give it a program to alert on:
from signal_analog.flow import Program, Detect, Filter, Data
from signal_analog.combinators import GT
# This program fires an alert if memory utilization is above 90% for the
# 'bar' application.
data = Data('memory.utilization', filter=Filter('app', 'bar')).publish(label='A')
alert_label = 'Memory Utilization Above 90'
detect = Detect(GT(data, 90)).publish(label=alert_label)
detector.with_program(Program(detect))With our name and program in hand, it's time to build up an alert rule that we can use to notify our teammates:
# We provide a number of notification strategies in the detectors module.
from signal_analog.detectors import EmailNotification, Rule, Severity
info_rule = Rule()\
# From our detector defined above.
.for_label(alert_label)\
.with_severity(Severity.Info)\
.with_notifications(EmailNotification('[email protected]'))
detector.with_rules(info_rule)
# We can now create this resource in SignalFx:
detector.with_api_token('foo').create()
# For a more robust solution consult the "Creating a CLI for your Resources"
# section below.To add multiple alerting rules we would need to use different detect
statements with distinct labels to differentiate them from one another.
More complex detectors, like those created as a function of two other data streams, require a more complex setup including data stream assignments. If we wanted to create a detector that watched for an average above a certain threshold, we may want to use the quotient of the sum() of the data and the count() of the datapoints over a given period of time.
from signal_analog.flow import \
Assign, \
Data, \
Detect, \
Ref, \
When
from signal_analog.combinators import \
Div, \
GT
program = Program( \
Assign('my_var', Data('cpu.utilization')) \
Assign('my_other_var', Data('cpu.utilization').count()) \
Assign('mean', Div(Ref('my_var'), Ref('my_other_var'))) \
Detect(When(GT(Ref('mean'), 2000))) \
)
print(program)The above code generates the following program:
my_var = data('cpu.utilization')
my_other_var = data('cpu.utilization').count()
mean = (my_var / my_other_var)
when(detect(mean > 2000))
We can also build up Detectors from an existing chart, which allows us to reuse our SignalFlow program and ensure consistency between what we're monitoring and what we're alerting on.
Let's assume that we already have a chart defined for our use:
from signal_analog.flow import Program, Data
from signal_analog.charts import TimeSeriesChart
program = Program(Data('cpu.utilization').publish(label='A'))
cpu_chart = TimeSeriesChart().with_name('Disk Utilization').with_program(program)In order to alert on this chart we'll use the from_chart builder for
detectors:
from signal_analog.combinators import GT
from signal_analog.detectors import Detector
from signal_analog.flow import Detect
# Alert when CPU utilization rises above 95%
detector = Detector()\
.with_name('CPU Detector')\
.from_chart(
cpu_chart,
# `p` is the Program object from the cpu_chart we passed in.
lambda p: Detect(GT(p.find_label('A'), 95).publish(label='Info Alert'))
)The above example won't actually alert on anything until we add a Rule, which
you can find examples for in the previous section.
signal_analog also provides functions for combining SignalFlow statements
into more complex SignalFlow Formulas. These sorts of Formulas can be useful
when creating more complex detectors and charts. For instance, if you would like
to multiply one data stream by another and receive the sum of that Formula,
it can be accomplished using Op and Mul like so:
from signal_analog.flow import Op, Program, Data
from signal_analog.combinators import Mul
# Multiply stream A by stream B and sum the result
A = Data('request.mean')
B = Data('request.count')
C = Op(Mul(A,B)).sum()Print(C) in the above example would produce the following output:
(data("request.mean") * data("request.count")).sum()
signal_analog provides abstractions for building dashboard groups in the
signal_analog.dashboards module.
Consult the upstream documentation for more information on the Dashboard Groups API.
Building on the examples described in the previous section, we'd now like to build a dashboard group containing our dashboards.
First, lets build a couple of Dashboard objects similar to how we did it in
the Building Dashboards example:
from signal_analog.dashboards import Dashboard, DashboardGroup
dg = DashboardGroup()
dash1 = Dashboard().with_name('My Little Dashboard1: Metrics are Magic')\
.with_charts(memory_chart)
dash2 = Dashboard().with_name('My Little Dashboard2: Metrics are Magic')\
.with_charts(memory_chart)Note: we do not create Dashboard objects ourselves, the DashboardGroup object is responsible for creating all child resources.
Many of the same methods for dashboards are available on dashboard groups as well, so let's give our dashboard group a memorable name and configure it's API token:
dg.with_name('My Dashboard Group')\
.with_api_token('my-api-token')Our final task will be to add dashboard to our dashboard group and create it in the API!
response = dg\
.with_dashboards(dash1)\
.with_api_token('my-api-token')\
.create()Now, storing API keys in source isn't ideal, so if you'd like to see how you can pass in your API keys at runtime check the documentation below to see how you can dynamically build a CLI for your resources.
Once you have created a dashboard group, you can update properties like name and description of a dashboard group or add/remove dashboards in a group.
Example 1:
dg.with_api_token('my-api-token')\
.update(name='updated_dashboard_group_name',
description='updated_dashboard_group_description')Example 2:
dg.with_api_token('my-api-token').with_dashboards(dash1, dash2).update()If you need to process SignalFx data outside the confince of the API it may be useful to call the SignalFlow API directly. Note that you may incur time penalties when pulling data out depending on the source of the data (e.g. AWS/CloudWatch).
SignalFlow constructs are contained in the flow module. The following is an
example SignalFlow program that monitors an API services (like Riposte)
RPS metrics for the foo application in the test environment.
from signal_analog.flow import Data, Filter
from signal_analog.combinators import And
all_filters = And(Filter('env', 'prod'), Filter('app', 'foo'))
program = Data('requests.count', filter=all_filters)).publish()You now have an object representation of the SignalFlow program. To take it for a test ride you can use the official SignalFx client like so:
# Original example found here:
# https://github.com/signalfx/signalfx-python#executing-signalflow-computations
import signalfx
from signal_analog.flow import Data, Filter
from signal_analog.combinators import And
app_filter = Filter('app', 'foo')
env_filter = Filter('env', 'prod')
program = Data('requests.count', filter=And(app_filter, env_filter)).publish()
with signalfx.SignalFx().signalflow('MY_TOKEN') as flow:
print('Executing {0} ...'.format(program))
computation = flow.execute(str(program))
for msg in computation.stream():
if isinstance(msg, signalfx.signalflow.messages.DataMessage):
print('{0}: {1}'.format(msg.logical_timestamp_ms, msg.data))
if isinstance(msg, signalfx.signalflow.messages.EventMessage):
print('{0}: {1}'.format(msg.timestamp_ms, msg.properties))It is always assumed that a Chart belongs to an existing Dashboard. This makes it easier for the library to manage the state of the world.
In a signal_analog world it is assumed that all resource names are unique.
That is, if we have two dashboards 'Foo Dashboard', when we attempt to update
either dashboard via signal_analog we expect to see errors.
Resource names are assumed to be unique in order to simplify state management by the library itself. In practice we have not found this to be a major inconvenience.
When conflicts arise between the state of a resource in your configuration and what SignalFx thinks that state should be, this library always prefers the local configuration.
Resource objects contain a number of builder methods to enable a "fluent" API
when describing your project's dashboards in SignalFx. It is assumed that these
methods do not perform state-affecting actions in the SignalFx API.
Only "CCRUD" (Create, Clone, Read, Update, and Delete) methods will affect the state of your resources in SignalFx.
signal_analog provides builders for fully featured command line clients that
can manage the lifecycle of sets of resources.
Integrating with the CLI is as simple as importing the builder and passing it your resources. Let's consider an example where we want to update two existing dashboards:
#!/usr/bin/env python
# ^ It's always good to include a "hashbang" so that your terminal knows
# how to run your script.
from signal_analog.dashboards import Dashboard
from signal_analog.cli import CliBuilder
ingest_dashboard = Dashboard().with_name('my-ingest-service')
service_dashboard = Dashboard().with_name('my-service')
if __name__ == '__main__':
cli = CliBuilder()\
.with_resources(ingest_dashboard, service_dashboard)\
.build()
cli()Assuming we called this dashboards.py we could run it in one of two ways:
- Give the script execution rights and run it directly
(typically
chmod +x dashboards.py)./dashboards.py --api-key mykey update
- Pass the script in to the Python executor
python dashboards.py --api-key mykey update
If you want to know about the available actions you can take with your new
CLI you can always the --help command.
./dashboards.py --helpThis gives you the following features:
- Consistent resource management
- All resources passed to the CLI builder can be updated with one
updateinvocation, rather than calling theupdate()method on each resource indvidually
- All resources passed to the CLI builder can be updated with one
- API key handling for all resources
- Rather than duplicating your API key for each resource, you can instead invoke the CLI with an API key
- This also provides a way to supply keys for users who don't want to store them in source control (that's you! don't store your keys in source control)
Please read our docs here for more info about contributing.