better service monitoring through histograms

33
Better service monitoring through histograms Fred Moyer - @phredmoyer San Francisco Perl Mongers, 07-26- 2016

Upload: fred-moyer

Post on 13-Apr-2017

455 views

Category:

Software


0 download

TRANSCRIPT

Page 1: Better service monitoring through histograms

Better service monitoring through histogramsFred Moyer - @phredmoyerSan Francisco Perl Mongers, 07-26-2016

Page 2: Better service monitoring through histograms

Systems break while we sleep

How often are you woken up for false alarms?

Welcome

Page 3: Better service monitoring through histograms

Synthetics

Easy to setup, but not a real user

Page 4: Better service monitoring through histograms

Synthetics

Stephen Falken: Uh, uh, General, what you see on these screens up here is a fantasy; a computer-enhanced hallucination. Those blips are not real missiles. They're phantoms. (War Games, 1983)

Page 5: Better service monitoring through histograms

Real Users

These are your users, right?

Page 6: Better service monitoring through histograms

Real data

Real Users

Page 7: Better service monitoring through histograms

500 ms is really 2,000 ms

Spike Erosion

Page 8: Better service monitoring through histograms

What threshold do you choose?

Threshold Alerting

Page 9: Better service monitoring through histograms

“Alert me if requests take longer than 200 ms”

10,10,10,10,10,10,10,10,10,5000

Alerts on one outlier in 10

Threshold Alerting

Page 10: Better service monitoring through histograms

“Alert if request average over one minute is longer than 200 ms”

avg(10,10,210,210,210,210) = 143 (860/6)

Does not alert on multiple high samples

Threshold Alerting

Page 11: Better service monitoring through histograms

‘average’ eq ‘arithmetic mean’A=S/N

A = averageN = the number of terms

S = the sum of the numbers in the set

Math Refresher

Page 12: Better service monitoring through histograms

median = midpoint of data set

The 50th percentile is 555 - q(0.5)

Value 111 222 333 444 555

666

777 888 999

Sample # 1 2 3 4 5 6 7 8 9

Math Refresher

Page 13: Better service monitoring through histograms

90th percentile - 90% of samples below it

The 90th percentile is 1,000 - q(0.9)

Value 111

222

333

444

555

666

777

888

999 1,00

01,111

Sample #

1 2 3 4 5 6 7 8 9 10 11

Math Refresher

Page 14: Better service monitoring through histograms

100th Percentile - the maximum value

The 100th percentile is 1,111 - q(1)

Value 111

222

333

444

555

666

777

888

999

1,000 1,11

1Sample #

1 2 3 4 5 6 7 8 9 10 11

Math Refresher

Page 15: Better service monitoring through histograms

Sample value

Number of samples

Histogram

Page 16: Better service monitoring through histograms

Sample value

Number of samples

Normal Distribution

Page 17: Better service monitoring through histograms

Sample value

Number of samples

Normal Distribution

34% within one sigma (σ)

Page 18: Better service monitoring through histograms

Sample value

Number of samples

Non-Normal Distribution

Page 19: Better service monitoring through histograms

Sample value

Number of samples

Non-Normal Distribution

Page 20: Better service monitoring through histograms

Non-Normal Distribution

Operations data groups at different points

Page 21: Better service monitoring through histograms

Non-Normal Distribution

Users to the right of the red line are gone

Page 22: Better service monitoring through histograms

Request latency“We keep hearing from people that the

website is slow. But it is fine when we test it, and the request latency graph is

constant”

You are only looking at part of the picture.

Page 23: Better service monitoring through histograms

Heat Map

Histograms over time windows

Page 24: Better service monitoring through histograms

Percentiles

Page 25: Better service monitoring through histograms

Practical PercentilesBandwidth usage is often billed at 95th percentile

usageRecord 5 minute data usage intervals

Sort samples by value of sampleThrow out the highest 5% of samples

Charge usage based on the remaining top sample, i.e. 300 MB transferred over 5 minutes = 1 MB/s rate

billing

Page 26: Better service monitoring through histograms

Practical Percentiles

If I measure 95th percentile per 5 minutes all month long,

I CANNOT calculate 95th percentile over the month.

Page 27: Better service monitoring through histograms

Angry users

How many users are you pissing off?

Page 28: Better service monitoring through histograms

Angry users

Page 29: Better service monitoring through histograms

“Alert me if request latency 90th percentile over one minute is

exceeded”

Percentile based alerting

q(0.9)[10,10,10,10,10,10,10,10,5000] == 10Alert IS NOT triggered

Do you want to be woken up for this? NO!

Page 30: Better service monitoring through histograms

“Alert me if request latency 90th percentile over one minute is exceeded”

Percentile based alerting

q(0.9)[10,10,10,10,10,10,250,300] = ~270Alert IS triggered

Do you want to be woken up for this? YES!

Page 31: Better service monitoring through histograms

Percentile based alerting

Page 32: Better service monitoring through histograms

Who’s using this approach?

Google.comCirconus.com

You?

Page 33: Better service monitoring through histograms

Questions?

Thanks to Circonus.com for the tools and help with the math

http://www.circonus.com/free-account/