Reports about the battle of CI and CD, orchestration and secrets of OpenStack
On September 27, we conducted the second Orcs here ’mitap - about orchestration, automation and field application of CI / CD. In this post full video and timecode with important places from three reports.
Topics are:
- Environment as a Service - about the operation and configuration secrets of OpenStack
- Pod, Cloud and Two Smoking Hubs - about scaling Selenium farms
- CI vs CD: the arms race - about how CI and CD “fought” Yandex.Money
Environment as a Service
Alexander Prizov and Maxim Ogryzkov from Yandex.Money told about the features of operation and orchestration of high-loaded OpenStack installations for different types of test environment.
About our approach
0:20 - What is Environment as as Service?
1:00 - How to prepare the infrastructure for change?
1:30 - What to do with the old test benches?
3:55 - Requirements for new test stands
4:30 - Why did we choose OpenStack? A little about Murano
5:50 - About our approach to automating the creation of test patterns
7:30 - Example of a declarative description of the calculations
8:20 - About Jenkinks, Grafana, kibana and update tracking
9:10 - What numbers did we get?
OpenStack and highload
9:40 - What happened when we picked up 50 virtual machines at the same time?
11:25 - You need to post cron tasks. About the work of the scheduler
Neutron and highload
13:45 - We use the dual stack. What is it and why doesn’t NAT need there?
15:33 - It happens that the functionality of the tool is not enough. What can be done?
18:00 - Many different services that affect the work of each other. How to be?
22:00 - Write the code as documentation. How we found a setting that is not found anywhere
23:35 - Solution “no response to inactivity of the probe after 10 seconds, disconnecting”
25:31 - How to adjust the size of the mac-table
26:00 - How to enable rx-flow-hash
26:37 - About global and personal safety rules
Pod, Cloud and Two Smoking Hubs
Alexey Boyko from Wrike told how they saved time and money on scaling a Selenium farm by using Google Cloud Engine and Kubernetes.
Prehistory
0:15 - What is Wrike and where is Selenium
2:20 - What did Wrike have when scaling came up?
3:20 - Browser queues, Selenium Hub failures and slow regression packs
4:50 - Scaling requirements and how to implement them
7:30 - Why did you choose Google Cloud: kubernetes, short-lived nodes and savings?
8:50 - Why chose kubernetes?
How to mix technology cocktail
9:50 - About autoscaling and why?
12:00 - How autoscaling works
14:40 - Preemptible nodes - low price and dangerous restrictions
18:00 - Why do I need a load balancer?
19:00 - How it all works together. TeamCity.
21:00 - TeamCity role in the process
21:30 - Detailed process description
23:40 - What was the result?
CI vs CD: arms race
Vasily Sozykin (ex-Yandex.Money), Dmitry Zinin (Yandex.Money)
We present the description of the report from the authors without changes.
War. The essence of the war never changes, but the release cycle, automation and microservices - as much as you like.
Start. It used to be quietly living with releases once a month, but then this fragile world reeled.
Chronicle. When it became forty releases a day, the relationship of development and operation began to heat up. This has resulted in arming the parties, new automation and optimization tools, and symmetrical responses.
')
Reboot. The parties entered into a peace agreement, and the confrontation is in the past.
Hostage of the war. The civilian population from neighboring departments during the confrontation had not sweet.
At some moments, the microphone could not withstand the charisma of Basil, so a few seconds during the report became quiet.
1:40 - On the role of regulators in automation of fintech
3:10 - “Release”, when there are few components and developers
4:30 - Distribution of roles for releases
6:40 - Nexus, Gradle, jacoco and a wet loot
8:50 - Jenkins and two releases per day
9:30 - Automation of integration and load testing. Zone 21.
10:15 - Release attendant. Why is it needed?
11:10 - Ansible Playbook for Releases
12:20 - On the abolition of the "release committee". Notification Automation
13:30 - New distribution of roles at releases. Two releases per week.
14:30 - And two release masters. And this is why?
16:00 - 20 releases per day and our first bot.
18:00 - Release on commit. Every 30 minutes.
19:00 - Multi-streaming from the CD
20:30 - About CI and CD without people
21:50 - About the future of releases even more without people
24:00 - Conclusions - from the release once a month to the release every hour.
That's all for today - come to our events to hear everything live, and subscribe to the blog so as not to miss other reports on important topics.
Source: https://habr.com/ru/post/425695/
All Articles