Data-Driven Operations - Practice realtime data analyse

Data-Driven OperationsPractice realtime data analyse

@khsing

Who am I

• Currently, I am a operations architect in SINA.

• Focus on automation tools and devops method

What kind of data is for operations?

Before we talk data

How is one day of ops?

• Check the Dashboard and looks good.

• Start work, write scripts or configurations

• Suddenly, Receiving alert SMS/Email or problem reported by CS.

• Start work with event/problem/outage

You are the Fireman http://www.flickr.com/photos/40699207@N05/3838012090/

Find the problem

• take a look at Dashboard, Nagios, and monitor

• grep logs from hundreds of host.

• watch the network diagram

• guess what is going wrong

Driven by problem

Passive

Be Active

Let’s talk data

datas

• Logs

• Access log, error log, exception log, step log

• Configuration Change log, Release log

• Performance Measurement

• Product operations data.

Logs

• Success is useless.

• Error is useful.

Process logs

• Realtime or near realtime take big benefit

• You can’t waste 1 hour when problem really happen

• You have to feel problem before too many users blame.

Process Logs

• Automatically category.

Normal logs

Categorised logs

Performance Measurement

• How fast when end-user visit our website?

• Where are they come from?

• Which datacenter are they visited?

• What the slow/fast user ratio?

Product Operations Data

• like DAU

• Drop, Spike, Increase are event, need take action.

Change/Release log

• Many problem come with Change or Release

• You have to watch those data after you did a change or release.

• Change/Release log have to visible on dashboard.

Change/Release log

Be active

Don’t defensive

–Olbrich Desouza

Attack is the best form of defence

Tools

• Splunk - commercial

• Logstash, ElasticSearch, Kibana

• Graphite

• StatsD

Q&A