NGINX: High Performance Load Balancing

NGINX: High Performance
Load Balancing

MORE INFORMATION AT NGINX.COM
Who Are We?
Floyd Smith
Director, Content Marketing, NGINX
Formerly:
• Sr. Technical Writer, Apple
• Group Channel Manager, Altavista
• Author of best-selling technology books
Faisal Memon
Product Marketer,NGINX
Formerly:
• Sr. Technical Marketing Engineer,
Riverbed
• Technical Marketing Engineer, Cisco
• Software Engineer, Cisco

1. Introducing NGINX
2. Load Balancing History
3. Basic NGINX HTTP load balancing
4. Essentials: health checks, persistence,
SSL termination, etc.
5. Improving performance
6. TCP/UDP load balancing
7. DNS service discovery
Agenda
3

4
50%of the top 100,000
busiest websites
Source: W3Techs Web Technology Survey
>

Where NGINX Plus fits
5
Internet
Web Server
Serve content from disk
Application Gateway
FastCGI, uWSGI, Passenger…
Reverse Proxy
Caching, load balancing…
HTTP traffic

NGINX and Load Balancing
6
• Survey says: Large companies interested in global load balancing,
smaller ones in CDNs and public cloud
• Load balancing is a hot topic on our website
• Gartner report shows NGINX as an ADC/load balancing leader
• Ebook on 5 Reasons to Switch to Load Balancing
• Load balancing training
• NGINX Professional Services – architectural experts
…and much more; contact Sales for free evaluation

Agenda
7

NGINX Plus works in all environments
8
Public/Private/Hybrid CloudBare Metal Containers

Agenda
9

Load Balancing Overview
Pool
Virtual
Server
Why a load balancer?
• Availability
• Performance
• Security
• Control

Basic Load Balancing Configuration
upstream my_upstream {
server server1.example.com;
}
server {
listen 80;
location / {
proxy_set_header Host $host;
proxy_pass http://my_upstream;
}
}
• upstream defines the pool
• server defines the virtual server
• listen defines the IPand port the virtual
server listens to. Default is to bind to port 80
on all IPs on the system.
• proxy_pass tells the virtual server what pool
to use
• proxy_set_header passes through original
client Host header. Default is to rewrite Host
header to name and port of proxied server.
• location defines what URI enclosed
configuration applies to

Key Files and Directories
• /etc/nginx/ -- Where all NGINX configuration is stored
• /etc/nginx/nginx.conf -- Top-level NGINX configuration, should not
require much modification
• /etc/nginx/conf.d/*.conf -- Where your configuration for virtual servers
and upstreams goes, i.e. www.example.com.conf

Agenda
13

Multiplexing Multiple Sites on One IP
server {
listen 80 default_server;
server_name www.example.com;
# ...
}
server {
listen 80;
server_name www.example2.com;
# ...
}
server {
listen 80;
server_name www.example3.com;
# ...
}
• server_name defines the hostname the
virtual server is responsible for.
• default_server defines the virtual server
to use if Host header is empty.

Layer 7 Request Routing
server {
# ...
location /service1 {
proxy_pass http://upstream1;
}
}
}
}
• location blocks are used to do Layer 7
routing based on URL

d
Active Health Checks
zone my_upstream 64k;
server server1.example.com slow_start=30s;
}
server {
# ...
location /health {
internal;
health_check interval=5s uri=/test.php
match=statusok;
proxy_set_header HOST www.example.com;
}
match statusok {
# Used for /test.php health check
status 200;
header Content-Type = text/html;
body ~ "Server[0-9]+ is alive";
}
• Polls /test.php every 5 seconds
• If response is not 200, server marked
as failed
• If response body does not contain
“ServerN is alive”, server marked as
failed
• Recovered/new servers will slowly
ramp up traffic over 30 seconds
• Exclusive to NGINX Plus

Hash Session Persistence
hash $binary_remote_addr consistent;
}
• Always pins same client IP address to same
server
Potential pitfalls:
• Uneven distribution - Could be 100s of
users sharing one IP address behind NAT
• Mobile – If client changes IP, session is lost

Sticky Cookie Session Persistence
sticky cookie name expires=1h
domain=.example.com path=/;
}
• NGINX will insert a cookie using the specified
name
• expires defines how long the cookie is valid for.
The default is for the cookie to expire at the end of
the browser session.
• domain specifies the domain the cookie is valid
for. If not specified, domain field of cookie is left
blank
• path specifies the path the cookie is set for. If not
specified, path field of cookie is left blank

Basic SSL Configuration
server {
listen 80 default_server;
return 301 https://$server_name$request_uri;
}
server {
listen 443 ssl default_server;
ssl_certificate cert.crt;
ssl_certificate_key cert.key;
# ...
}
• return 301 will force all traffic to
SSL, which is good for SEO.
• ssl parameter needs to be added to
listen directive
• ssl_certificate specifies where
the public certificate is located.
• ssl_certificate_key specifies
where the private key is located.

Using SSL to Upstream Servers
}
server {
listen 443 ssl;
# ...
location / {
proxy_pass https://my_upstream;
}
}
• Use https instead of http in the
proxy_pass directive

Agenda
21

Modifications to main nginx.conf
user nginx;
worker_processes auto;
# ...
http {
# ...
keepalive_timeout 300s;
keepalive_requests 100000;
}
• Set in main nginx.conf file
• Default value for worker_processes varies on
system and installation source
• auto means to create one worker process per core.
This is recommended for most deployments.
• keepalive_timeout controls how long to keep idle
connections to clients open. Default: 75s
• keeplive_requests Max requests on a single client
connection before its closed
• keepalive_* can also be set per virtual server

HTTP/1.1 Keepalive to Upstreams
keepalive 32;
}
server {
location / {
proxy_http_version 1.1;
proxy_set_header Connection "";
}
}
• keepalive enables TCP connection cache
• By default NGINX uses HTTP/1.0 with
Connection: Close
• proxy_http_version upgrades connection
to HTTP/1.1
• proxy_set_header enables keepalive by
clearing Connection: Close HTTP header

Dual-stack RSA/ECC SSL Configuration
server {
listen 443 ssl default_server;
ssl_certificate cert_rsa.crt;
ssl_certificate_key cert_rsa.key;
ssl_certificate cert_ecdsa.crt;
ssl_certificate_key cert_ecdsa.key;
# ...
}
• Specify two sets of
ssl_certificate and
ssl_certificate_key directives
• NGINX will serve ECDSA cert for
clients that support it and RSA for
those that don’t
• Most modern browsers and OS’s
support ECC
• ECC has 2-3x better performance
than RSA based on our testing.

SSL Session Caching and HTTP/2
server {
listen 443 ssl http2 default_server;
ssl_certificate cert.crt
ssl_certificate_key cert.key
ssl_session_cache shared:SSL:10m;
ssl_session_timeout 10m;
}
• Improves SSL/TLS performance
• 1 MB session cache can store about 4,000
sessions
• Cache shared across all NGINX workers
• HTTP/2 improves performance
• Note: HTTP/2 requires OpenSSL 1.0.2 or
greater to work properly

Agenda
26

Basic TCP/UDP Load Balancing Configuration
stream {
server server1.example.com:1234;
server server2.example.com:2345;
}
server {
listen 1123 [udp];
proxy_pass my_upstream;
}
}
• All TCP/UDP load balancing configuration
goes within the stream block.
• The port number is mandatory when
configuring upstream TCP/UDP servers.
• udp parameter to listen directive enables
UDP load balancing.
Note: stream configuration cannot go into the
conf.d/ folder. It is recommended to put a
stream block in the main nginx.conf file and use
the include directive to include a separate folder
such as stream.conf.d/ which contains virtual
server and pool configuration.

MySQL Load Balancing Configuration
stream {
upstream galera {
server db1.example.com:3306;
server db2.example.com:3306 backup;
server db3.example.com:3306 down;
}
server {
listen 3306;
proxy_pass galera;
proxy_connect_timeout 1s;
}
}
• TCP load balancing across a MySQL
Galera cluster.
• Uses a single master for all reads/writes
• Having multiple write masters can lead to
collisions and potentially erroneous data

MySQL Load Balancing with Split Read/Writes
stream {
upstream galera_write {
server db2.example.com:3306 backup;
server db3.example.com:3306 down;
}
upstream galera_read {
}
server {
listen 3308;
proxy_pass galera_read;
}
server {
listen 3309;
proxy_pass galera_write;
}
}
• Use separate virtual servers with different
ports to separate reads and writes.
• All writes go to designated master, with
other servers as backups
• Reads load balanced across other servers
• Requires application code to be more
database aware

DNS Load Balancing Configuration
stream {
upstream dns_servers {
server 192.168.136.130:53;
server 192.168.136.131:53;
}
server {
listen 53 udp;
listen 53; #tcp
proxy_pass dns_servers;
proxy_responses 1;
proxy_timeout 1s;
error_log /var/log/nginx/dns.log info;
}
}
• Multiple listen directives as DNS uses
TCP for responses greater than 512
bytes
• proxy_responses let’s NGINX know
there will only be a single response from
the upstream server
• error_log instructs NGINX to output
proxy events into the specified log file.
There is no access log for TCP/UDP
traffic as NGINX does not inspect the
payload.

TCP/UDP Health Checks
stream {
server {
listen 12345;
proxy_pass tcp;
health_check;
}
server {
listen 53 udp;
proxy_pass dns_upstream;
health_check udp;
}
}
• For TCP applications NGINX establishes a
TCP connection every 5 seconds. Standard
“TCP Connect” health check.
• For UDP it will send “nginx health check”
every 5 seconds and expect the absence of
ICMP “Destination Unreachable” in the
response.
1. Improving performance1. Improving performance

Agenda
32

Service Discovery with Consul
• Special “registrator” container
which for other containers
starting or stopping
• When container goes up/down,
registrator updates service
registry
• NGINX Plus polls service
registry DNS interface to get
updated listed of container
IP/port

DNS Service Discovery with Consul
resolver consul:53 valid=10s;
upstream service1 {
zone service1 64k;
server service1.service.consul service=http
resolve;
}
• NGINX Plus will look up consul
in /etc/hosts/ file if using links or
using Docker embedded DNS
server.
• By default Consul uses this
format for services:
[tag.]<service>.service[.
datacenter].<domain>

DNS Service Discovery with Docker
resolver 127.0.0.11 valid=10s;
upstream service1 {
zone service1 64k;
server service1 service=http resolve;
}
• resolver is the IP Address of the
DNS server. For the Docker
embedded DNS server this is
always 127.0.0.11.
• The optional valid parameter
overrides the DNS TTL value.
• service=http tell NGINX Plus to
look for DNS SRV records which
contain port number.
• The resolve parameter tell
NGINX Plus to re-resolve this
hostname

Summary
• The server directive defines a virtual server
• The upstream directive defines the pool
• The proxy_pass directive links the virtual server to the pool
• location blocks are used to do layer 7 request routing based on URL
• Multiple ssl_certificate and ssl_certificate_key directives can be used to do dual-
stack RSA/ECC
• The stream directive is used for TCP/UDP load balancing
• NGINX Plus can integrate with the DNS interface of Docker and Consul

Q & ATry NGINX Plus free for 30 days: nginx.com/free-trial-
request

nginx.conf 2017
Sep 6-8, 2017 | Portland,
OR
nginx.com/nginxconf
Sign up now,
get 50% off
Use code: WEBINAR
Learn from industry veterans at the world’s top companies
Special guest: Jimmy Yang from HBO’s Silicon Valley
Architect The Future

NGINX: High Performance Load Balancing

More Related Content

What's hot (20)

Similar to NGINX: High Performance Load Balancing (20)

More from NGINX, Inc. (20)

Recently uploaded (20)

NGINX: High Performance Load Balancing

Editor's Notes