InfluxDB 101 – Concepts and Architecture by Michael DeSa, Software Engineer | InfluxData

Michael Desa / Software Engineer
InfluxDB 101

© 2018 InfluxData. All rights reserved.2
Agenda
By the end of this session, users should be able to...
● Define time-series data
● Describe what InfluxDB is and its relationship to
InfluxData
● Explain the InfluxDB data model
● Reason about the impact of the schema in an instance

What is time-series data?

Regular vs Irregular Time-Series

Metrics vs Events

Question: Time-Series?

What is a time-series database (tsdb)?

Time-series Database
● A database where you manage and store time-series
data
● Efficiently handles time-series data
● Supports time based queries

Why couldn’t I just use [insert db]?

But, time-series is not just a database problem

Time-series problems
● Visualizing your data
● Alerting you data
● Processing your data
● Taking action based on your data

What is InfluxDB/InfluxData?

InfluxData History
● 2012 - Errplane
● 2014 - InfluxDB is born
● 2015 - Transition to InfluxData
○ A platform for time-series data
● 2018 - 2.0 of InfluxData

Why InfluxData
● Easy to get started with
● Aims to solve the entire time-series problem
● Scales well
○ Both horizontally and vertically

Canonical Time-Series Line Graph

The Label (measurement)

The Legend (tags)
ticker=A
ticker=AA
ticker=AAPL
market=NASDAQ
market=NYSE

Y-Axis Values
price=177.03
price=32.10
price=35.52

X-Axis Values

Series
stock_price,ticker=A,market=NASDAQ

Data Model
● Measurement
○ High level grouping of data
● Tags
○ Indexed key-value pairs
● Fields
○ Key-value pairs of actual data
● Timestamp
○ Time of the data
● Series
○ A unique combination of measurement+tags

Line Protocol
cpu,host=serverA,num=1,region=west idle=1.667,system=2342.2 1492214400000000000
Measurement

Line Protocol
Tag
s

Line Protocol
Fields

Line Protocol
timestamp

InfluxData Languages
● InfluxQL
○ SQL-like query language
● TICKscript
○ Time-series data processing language
● Flux
○ Next generation functional data scripting language

InfluxQL
> SELECT index, id FROM h2o_quality WHERE time > now() - 1w GROUP BY location
name: h2o_quality
tags: location = coyote_creek
time index id
---- ----- ---
2015-08-18T00:00:00Z 41 1
2015-08-18T00:00:00Z 41 1
name: h2o_quality
tags: location = santa_monica
time index id
---- ----- ---
2015-08-18T00:00:00Z 99 2
2015-08-18T00:06:00Z 56 2

TICKscript
var measurement = 'requests'
var data = stream
|from()
.measurement(measurement)
|where(lambda: "is_up" == TRUE)
|where(lambda: "my_field" > 10)
|window()
.period(5m)
.every(5m)
// Count number of points in window
data
|count('value')
.as('the_count')
// Compute mean of data window
data
|mean('value')
.as('the_average')

Flux
// get all data from the telegraf db
from(bucket:”telegraf/autogen”)
// filter that by the last hour
|> range(start:-1h)
// filter further by series with a specific measurement and field
|> filter(fn: (r) => r._measurement == "cpu" and r._field == "usage_system")

Why Flux?
● Composabile
○ Users should be able to take pieces of different scripts and
combine them into a single one to solve their own problem
● Extensible
○ Adding new functions and capabilities to flux should be easy
● Shareable
○ Users should be able to create libraries and packages to solve
specific problems
● Flexible
○ Users should be able to use the language for arbitrary data
processing

What not to do

Don’t Encode Data into Measurement/Tags
Bad:
cpu.server-5.us-west value=2 1444234982000000000
cpu.server-6.us-west value=4 1444234982000000000
mem-free.server-6.us-west value=2500 1444234982000000000
Good:
cpu,host=server-5,region=us-west value=2 1444234982000000000
cpu,host=server-6,region=us-west value=4 1444234982000000000
mem-free,host=server-6,region=us-west value=2500 1444234982000000

Don’t Encode Data into Measurement/Tags
Bad:
cpu,server=localhost.us-west value=2 1444234982000000000
cpu,server=localhost.us-east value=3 1444234982000000000
Good:
cpu,host=localhost,region=us-west value=2 1444234982000000000
cpu,host=localhost,region=us-east value=3 1444234982000000000

Don’t Use Tags with Very High Variability
Bad:
response_time,session_id=33254331,request_id=3424347 value=340 14442349820000
Good-ish:
response_time session_id=33254331,request_id=3424347,value=340 14442349820000

Don’t Use Too Few Tags
Bad:
cpu,region=us-west host="server1",value=0.5 1444234986000
cpu,region=us-west host="server2",value=4 1444234982000
cpu,region=us-west host="server2",value=1 1444234982000
Good-ish:
cpu,region=us-west,host=server1 value=0.5 1444234986000
cpu,region=us-west,host=server2 value=4 1444234982000
cpu,region=us-west,host=server2 value=1 1444234982000

What should I do then?

Designing a Schema
● What dashboards do I need?
● What alerts do I need?
● What kind of reports do I want to generate?
● What type of information do I need readily available when there’s
an incident?

Schema Example
● I operate a SaaS application
● There are ~1000 different services
● I want to know the request and error rates for each service
● I want to trigger an alert if the error rate for each service
● I want to see the services with the highest average request duration

Data Available
● app Service name, e.g. user_service, auth_service…
● container_id Container ID of the container running the service
● path HTTP request path
● method HTTP method, e.g. GET, POST, DELETE…
● src Hostname of client making request
● dest Hostname of server being contacted
● status HTTP status code associated with the request
● request_id Unique request identifier
● duration Duration of request
● bytes_tx Number of bytes transmitted
● bytes_rx Number of bytes received

Question
Why would it be a bad idea to make container_id or
request_id a tag?

Answer
Why would it be a bad idea to make container_id or
request_id a tag?
request_id and container_id both have a high cardinality and could result in an large number of
series, which impacts memory utilization.
request_id is substantially worse than container_id. In the next few releases we hope to allow for
indexing on container_id.

Question
How should we organize our data?

Data Available
● app Service name, e.g. user_service, auth_service…
● container_id Container ID of the container running the service
● path HTTP request path
● method HTTP method, e.g. GET, POST, DELETE…
● src Hostname of client making request
● dest Hostname of server being contacted
● status HTTP status code associated with the request
● request_id Unique request identifier
● duration Duration of request
● bytes_tx Number of bytes transmitted
● bytes_rx Number of bytes received

Schema
measurement:
latency
tags:
app container_id path method src dst status
fields:
request_id duration bytes_tx bytes_rx

Request/Error Rate per Service
Top 10 average request duration
> SELECT top(avg_duration, app, 10) FROM (
SELECT mean(duration) AS avg_dur
FROM latency
WHERE time > now() - 1h
GROUP BY time(1m), *
)
Request Rate Per Service
> SELECT count(duration)
FROM latency
WHERE time > now() - 10m
GROUP BY app, time(1s) fill(none)
Error Rate Per Service
> SELECT count(duration)
FROM latency
WHERE time > now() - 10m AND status != ‘200’
GROUP BY app, time(1s) fill(none)

InfluxDB 101 – Concepts and Architecture by Michael DeSa, Software Engineer | InfluxData

InfluxDB 101 – Concepts and Architecture by Michael DeSa, Software Engineer | InfluxData

More Related Content

What's hot (20)

Similar to InfluxDB 101 – Concepts and Architecture by Michael DeSa, Software Engineer | InfluxData (20)

More from InfluxData (20)

Recently uploaded (20)

InfluxDB 101 – Concepts and Architecture by Michael DeSa, Software Engineer | InfluxData