Language

English

中文

Purpose

Test the connection draining in aws elastic beanstalk.

Method

Call the route "/test" before and after rolling update, and observe the behavior.

Control variable

• 1 - Connection draining : disabled (Classic Load Balancer)
• 2 - Connection draining : enabled (Draining timeout: 40 seoncds) (Classic Load Balancer)
• 3 - Application Load Balancer

Setting

EB

	Platform: Node.js 10 running on 64bit Amazon Linux 2/5.3.0
	Proxy: nginx
	EC2: t3.micro * 2
	Health check path: /ok
	Deployment policy: Rolling (batch size: fixed 1)

server

	server ready delay: 60 seconds
	server "/test" send ok delay: 10 seconds

client

	call "/test" interval: 5 seconds
	requset timeout: 40 seconds

Result

Case 1. Connection draining : disabled (Classic Load Balancer)

Time:

	start calling "/test":    14:42:00
	1st start deployment:     14:43:28
	1st complete deployment:  14:47:23
	2nd start deployment:     14:48:28
	2nd complete deployment:  14:52:19
	stop calling "/test":     14:53:20
	check result:             14:54:10

Avergage deployment time: 233 seconds

Client result:

	sent request count: 134
	received response count: 126
		- ok: 126
		- shutting down & not ready: 0
		- not ready: 0
		- shutting down: 0
		- not found: 0
		- other: 0
	error count: 8
		- "Error: Network Error" * 8
	unhandled request count: 0

EC2 logs

• instance 1

• instance 2

Case 2. Connection draining : enabled (Draining timeout: 40 seconds) (Classic Load Balancer)

Time:

	start calling "/test":    15:13:00
	1st start deployment:     15:13:54
	1st complete deployment:  15:18:02
	2nd start deployment:     15:18:58
	2nd complete deployment:  15:23:03
	stop calling "/test":     15:24:30
	check result:             15:25:10

Avergage deployment time: 246.5 seconds

Client result:

	sent request count: 138
	received response count: 138
		- ok: 138
		- shutting down & not ready: 0
		- not ready: 0
		- shutting down: 0
		- not found: 0
		- other: 0
	error count: 0
	unhandled request count: 0

EC2 logs

• instance 1

• instance 2

Case 3. Application Load Balancer

Time:

	start calling "/test":    10:02:20
	1st start deployment:     10:03:15
	1st complete deployment:  10:07:12
	2nd start deployment:     10:08:28
	2nd complete deployment:  10:12:25
	stop calling "/test":     10:13:45
	check result:             10:14:30

Avergage deployment time: 237 seconds

Client result:

	sent request count: 137
	received response count: 137
		- ok: 137
		- shutting down & not ready: 0
		- not ready: 0
		- shutting down: 0
		- not found: 0
		- other: 0
	error count: 0
	unhandled request count: 0

EC2 logs

• instance 1

• instance 2

Conclusion

1. When you enable Connection Draining on a load balancer, any back-end instances that you deregister will complete requests that are in progress before deregistration. - AWS what's new 2014-03-20

2. Application Load Balancer enable connection draining in default.

Pros of connection draining

• Avoid network error

Cons of connection draining

• Slightly increase the time of deployment