Basic Load Balancing Cookbook¶

Introduction¶

This document contains several examples of using basic load balancing services as a tenant or “regular” cloud user.

For the purposes of this guide we assume that the neutron and barbican command-line interfaces are going to be used to configure all features of Neutron LBaaS with an Octavia back-end. In order to keep these examples short, we also assume that tasks not directly associated with deploying load balancing services have already been accomplished. This might include such things as deploying and configuring web servers, setting up Neutron networks, obtaining TLS certificates from a trusted provider, and so on. A description of the starting conditions is given in each example below.

Please also note that this guide assumes you are familiar with the specific load balancer terminology defined in the Octavia Glossary. For a description of load balancing itself and the Octavia project, please see: Introducing Octavia.

Examples¶

Deploy a basic HTTP load balancer¶

While this is technically the simplest complete load balancing solution that can be deployed, we recommend deploying HTTP load balancers with a health monitor to ensure back-end member availability. See Deploy a basic HTTP load balancer with a health monitor below.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an HTTP application on TCP port 80.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes web requests to the back-end servers.

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol HTTP --protocol-port 80 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a basic HTTP load balancer with a health monitor¶

This is the simplest recommended load balancing solution for HTTP applications. This solution is appropriate for operators with provider networks that are not compatible with Neutron floating-ip functionality (such as IPv6 networks). However, if you need to retain control of the external IP through which a load balancer is accessible, even if the load balancer needs to be destroyed or recreated, it may be more appropriate to deploy your basic load balancer using a floating IP. See Deploy a basic HTTP load balancer using a floating IP below.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an HTTP application on TCP port 80.
These back-end servers have been configured with a health check at the URL path “/healthcheck”. See HTTP health monitors below.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes web requests to the back-end servers, and which checks the “/healthcheck” path to ensure back-end member health.

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Create a health monitor on pool1 which tests the “/healthcheck” path.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol HTTP --protocol-port 80 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer healthmonitor create --delay 5 --max-retries 4 --timeout 10 --type HTTP --url-path /healthcheck pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a basic HTTP load balancer using a floating IP¶

It can be beneficial to use a floating IP when setting up a load balancer’s VIP in order to ensure you retain control of the IP that gets assigned as the floating IP in case the load balancer needs to be destroyed, moved, or recreated.

Note that this is not possible to do with IPv6 load balancers as floating IPs do not work with IPv6. Further, there is currently a bug in Neutron Distributed Virtual Routing (DVR) which prevents floating IPs from working correctly when DVR is in use. See: https://bugs.launchpad.net/neutron/+bug/1583694

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an HTTP application on TCP port 80.
These back-end servers have been configured with a health check at the URL path “/healthcheck”. See HTTP health monitors below.
Neutron network public is a shared external network created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes web requests to the back-end servers, and which checks the “/healthcheck” path to ensure back-end member health. Further, we want to do this using a floating IP.

Solution:

Create load balancer lb1 on subnet private-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Create a health monitor on pool1 which tests the “/healthcheck” path.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.
Create a floating IP address on public-subnet.
Associate this floating IP with the lb1’s VIP port.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id private-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol HTTP --protocol-port 80 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer healthmonitor create --delay 5 --max-retries 4 --timeout 10 --type HTTP --url-path /healthcheck pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1
openstack floating ip create public
# The following IDs should be visible in the output of previous commands
openstack floating ip set --port <load_balancer_vip_port_id> <floating_ip_id>

Deploy a basic HTTP load balancer with session persistence¶

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an HTTP application on TCP port 80.
The application is written such that web clients should always be directed to the same back-end server throughout their web session, based on an application cookie inserted by the web application named ‘PHPSESSIONID’.
These back-end servers have been configured with a health check at the URL path “/healthcheck”. See HTTP health monitors below.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes web requests to the back-end servers, persists sessions using the PHPSESSIONID as a key, and which checks the “/healthcheck” path to ensure back-end member health.

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool which defines session persistence on the ‘PHPSESSIONID’ cookie.
Create a health monitor on pool1 which tests the “/healthcheck” path.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol HTTP --protocol-port 80 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP --session-persistence type=APP_COOKIE,cookie_name=PHPSESSIONID
openstack loadbalancer healthmonitor create --delay 5 --max-retries 4 --timeout 10 --type HTTP --url-path /healthcheck pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a TCP load balancer¶

This is generally suitable when load balancing a non-HTTP TCP-based service.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an custom application on TCP port 23456
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes requests to the back-end servers.
We want to employ a TCP health check to ensure that the back-end servers are available.

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Create a health monitor on pool1 which probes pool1’s members’ TCP service port.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol TCP --protocol-port 23456 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol TCP
openstack loadbalancer healthmonitor create --delay 5 --max-retries 4 --timeout 10 --type TCP pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a QoS ruled load balancer¶

This solution limits the bandwidth available through the Load Balancer’s VIP by applying a Neutron Quality of Service(QoS) policy to the VIP, so Load Balancer can accept the QoS Policy from Neutron; Then limits the vip of Load Balancer incoming or outgoing traffic.

Note

Before using this feature, please make sure the Neutron QoS extension(qos) is enabled on running OpenStack environment by command

openstack extension list

Scenario description:

QoS-policy created from Neutron with bandwidth-limit-rules by us.
Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an HTTP application on TCP port 80.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer and want to limit the traffic bandwidth when web traffic reaches the vip.

Solution:

Create QoS policy qos-policy-bandwidth with bandwidth_limit in Neutron.
Create load balancer lb1 on subnet public-subnet with the id of qos-policy-bandwidth.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack network qos policy create qos-policy-bandwidth
openstack network qos rule create --type bandwidth_limit --max-kbps 1024 --max-burst-kbits 1024 qos-policy-bandwidth
openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet --vip-qos-policy-id qos-policy-bandwidth
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 lb1 --protocol HTTP --protocol-port 80
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer member create --subnet-id <private_subnet_id> --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id <private_subnet_id> --address 192.0.2.11 --protocol-port 80 pool1

Deploy a load balancer with access control list¶

This solution limits incoming traffic to a listener to a set of allowed source IP addresses. Any other incoming traffic will be rejected.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with an custom application on TCP port 23456
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes requests to the back-end servers.
The application on TCP port 23456 is accessible to a limited source IP addresses (192.0.2.0/24 and 198.51.100/24).

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1 with allowed CIDRs.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol TCP --protocol-port 23456 --allowed-cidr 192.0.2.0/24 --allowed-cidr 198.51.100/24 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol TCP
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a non-terminated HTTPS load balancer¶

A non-terminated HTTPS load balancer acts effectively like a generic TCP load balancer: The load balancer will forward the raw TCP traffic from the web client to the back-end servers without decrypting it. This means that the back-end servers themselves must be configured to terminate the HTTPS connection with the web clients, and in turn, the load balancer cannot insert headers into the HTTP session indicating the client IP address. (That is, to the back-end server, all web requests will appear to originate from the load balancer.) Also, advanced load balancer features (like Layer 7 functionality) cannot be used with non-terminated HTTPS.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with a TLS-encrypted web application on TCP port 443.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
We want to configure a basic load balancer that is accessible from the internet, which distributes requests to the back-end servers.
We want to employ a TCP health check to ensure that the back-end servers are available.

Solution:

Create load balancer lb1 on subnet public-subnet.
Create listener listener1.
Create pool pool1 as listener1’s default pool.
Create a health monitor on pool1 which probes pool1’s members’ TCP service port.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --name listener1 --protocol HTTPS --protocol-port 443 lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTPS
openstack loadbalancer healthmonitor create --delay 5 --max-retries 4 --timeout 10 --type HTTPS --url-path /healthcheck pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 443 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 443 pool1

Deploy a TLS-terminated HTTPS load balancer¶

With a TLS-terminated HTTPS load balancer, web clients communicate with the load balancer over TLS protocols. The load balancer terminates the TLS session and forwards the decrypted requests to the back-end servers. By terminating the TLS session on the load balancer, we offload the CPU-intensive encryption work to the load balancer, and enable the possibility of using advanced load balancer features, like Layer 7 features and header manipulation.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with regular HTTP application on TCP port 80.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
A TLS certificate, key, and intermediate certificate chain for www.example.com have been obtained from an external certificate authority. These now exist in the files server.crt, server.key, and ca-chain.crt in the current directory. The key and certificate are PEM-encoded, and the intermediate certificate chain is multiple PEM-encoded certs concatenated together. The key is not encrypted with a passphrase.
We want to configure a TLS-terminated HTTPS load balancer that is accessible from the internet using the key and certificate mentioned above, which distributes requests to the back-end servers over the non-encrypted HTTP protocol.
Octavia is configured to use barbican for key management.

Solution:

Combine the individual cert/key/intermediates to a single PKCS12 file.
Create a barbican secret resource for the PKCS12 file. We will call this tls_secret1.
Create load balancer lb1 on subnet public-subnet.
Create listener listener1 as a TERMINATED_HTTPS listener referencing tls_secret1 as its default TLS container.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openssl pkcs12 -export -inkey server.key -in server.crt -certfile ca-chain.crt -passout pass: -out server.p12
openstack secret store --name='tls_secret1' -t 'application/octet-stream' -e 'base64' --payload="$(base64 < server.p12)"
openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --protocol-port 443 --protocol TERMINATED_HTTPS --name listener1 --default-tls-container=$(openstack secret list | awk '/ tls_secret1 / {print $2}') lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy a TLS-terminated HTTPS load balancer with SNI¶

This example is exactly like Deploy a TLS-terminated HTTPS load balancer, except that we have multiple TLS certificates that we would like to use on the same listener using Server Name Indication (SNI) technology.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with regular HTTP application on TCP port 80.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
TLS certificates, keys, and intermediate certificate chains for www.example.com and www2.example.com have been obtained from an external certificate authority. These now exist in the files server.crt, server.key, ca-chain.crt, server2.crt, server2.key, and ca-chain2.crt in the current directory. The keys and certificates are PEM-encoded, and the intermediate certificate chains are multiple certs PEM-encoded and concatenated together. Neither key is encrypted with a passphrase.
We want to configure a TLS-terminated HTTPS load balancer that is accessible from the internet using the keys and certificates mentioned above, which distributes requests to the back-end servers over the non-encrypted HTTP protocol.
If a web client connects that is not SNI capable, we want the load balancer to respond with the certificate for www.example.com.

Solution:

Combine the individual cert/key/intermediates to single PKCS12 files.
Create barbican secret resources for the PKCS12 files. We will call them tls_secret1 and tls_secret2.
Create load balancer lb1 on subnet public-subnet.
Create listener listener1 as a TERMINATED_HTTPS listener referencing tls_secret1 as its default TLS container, and referencing both tls_secret1 and tls_secret2 using SNI.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.

CLI commands:

openssl pkcs12 -export -inkey server.key -in server.crt -certfile ca-chain.crt -passout pass: -out server.p12
openssl pkcs12 -export -inkey server2.key -in server2.crt -certfile ca-chain2.crt -passout pass: -out server2.p12
openstack secret store --name='tls_secret1' -t 'application/octet-stream' -e 'base64' --payload="$(base64 < server.p12)"
openstack secret store --name='tls_secret2' -t 'application/octet-stream' -e 'base64' --payload="$(base64 < server2.p12)"
openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --protocol-port 443 --protocol TERMINATED_HTTPS --name listener1 --default-tls-container=$(openstack secret list | awk '/ tls_secret1 / {print $2}') --sni-container-refs $(openstack secret list | awk '/ tls_secret1 / {print $2}') $(openstack secret list | awk '/ tls_secret2 / {print $2}') -- lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1

Deploy HTTP and TLS-terminated HTTPS load balancing on the same IP and backend¶

This example is exactly like Deploy a TLS-terminated HTTPS load balancer, except that we would like to have both an HTTP and TERMINATED_HTTPS listener that use the same back-end pool (and therefore, probably respond with the exact same content regardless of whether the web client uses the HTTP or HTTPS protocol to connect).

Please note that if you wish all HTTP requests to be redirected to HTTPS (so that requests are only served via HTTPS, and attempts to access content over HTTP just get redirected to the HTTPS listener), then please see the example in the Layer 7 Cookbook.

Scenario description:

Back-end servers 192.0.2.10 and 192.0.2.11 on subnet private-subnet have been configured with regular HTTP application on TCP port 80.
Subnet public-subnet is a shared external subnet created by the cloud operator which is reachable from the internet.
A TLS certificate, key, and intermediate certificate chain for www.example.com have been obtained from an external certificate authority. These now exist in the files server.crt, server.key, and ca-chain.crt in the current directory. The key and certificate are PEM-encoded, and the intermediate certificate chain is multiple PEM-encoded certs concatenated together. The key is not encrypted with a passphrase.
We want to configure a TLS-terminated HTTPS load balancer that is accessible from the internet using the key and certificate mentioned above, which distributes requests to the back-end servers over the non-encrypted HTTP protocol.
We also want to configure a HTTP load balancer on the same IP address as the above which serves the exact same content (ie. forwards to the same back-end pool) as the TERMINATED_HTTPS listener.

Solution:

Combine the individual cert/key/intermediates to a single PKCS12 file.
Create a barbican secret resource for the PKCS12 file. We will call this tls_secret1.
Create load balancer lb1 on subnet public-subnet.
Create listener listener1 as a TERMINATED_HTTPS listener referencing tls_secret1 as its default TLS container.
Create pool pool1 as listener1’s default pool.
Add members 192.0.2.10 and 192.0.2.11 on private-subnet to pool1.
Create listener listener2 as an HTTP listener with pool1 as its default pool.

CLI commands:

openssl pkcs12 -export -inkey server.key -in server.crt -certfile ca-chain.crt -passout pass: -out server.p12
openstack secret store --name='tls_secret1' -t 'application/octet-stream' -e 'base64' --payload="$(base64 < server.p12)"
openstack loadbalancer create --name lb1 --vip-subnet-id public-subnet
# Re-run the following until lb1 shows ACTIVE and ONLINE statuses:
openstack loadbalancer show lb1
openstack loadbalancer listener create --protocol-port 443 --protocol TERMINATED_HTTPS --name listener1 --default-tls-container=$(openstack secret list | awk '/ tls_secret1 / {print $2}') lb1
openstack loadbalancer pool create --name pool1 --lb-algorithm ROUND_ROBIN --listener listener1 --protocol HTTP
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.10 --protocol-port 80 pool1
openstack loadbalancer member create --subnet-id private-subnet --address 192.0.2.11 --protocol-port 80 pool1
openstack loadbalancer listener create --protocol-port 80 --protocol HTTP --name listener2 --default-pool pool1 lb1

Heath Monitor Best Practices¶

While it is possible to set up a listener without a health monitor, if a back-end pool member goes down, Octavia will not remove the failed server from the pool until a considerable time has passed. This can lead to service disruption for web clients. Because of this, we recommend always configuring production load balancers to use a health monitor.

The health monitor itself is a process that does periodic health checks on each back-end server to pre-emptively detect failed servers and temporarily pull them out of the pool. Since effective health monitors depend as much on back-end application server configuration as proper load balancer configuration, some additional discussion of best practices is warranted here.

Heath monitor options¶

All of the health monitors Octavia supports have the following configurable options:

delay: Number of seconds to wait between health checks.
timeout: Number of seconds to wait for any given health check to complete. timeout should always be smaller than delay.
max-retries: Number of subsequent health checks a given back-end server must fail before it is considered down, or that a failed back-end server must pass to be considered up again.

HTTP health monitors¶

In general, the application-side component of HTTP health checks are a part of the web application being load balanced. By default, Octavia will probe the “/” path on the application server. However, in many applications this is not appropriate because the “/” path ends up being a cached page, or causes the application server to do more work than is necessary for a basic health check.

In addition to the above options, HTTP health monitors also have the following options:

url_path: Path part of the URL that should be retrieved from the back-end server. By default this is “/”.
http_method: HTTP method that should be used to retrieve the url_path. By default this is “GET”.
expected_codes: List of HTTP status codes that indicate an OK health check. By default this is just “200”.

Please keep the following best practices in mind when writing the code that generates the health check in your web application:

The health monitor url_path should not require authentication to load.
By default the health monitor url_path should return a HTTP 200 OK status code to indicate a healthy server unless you specify alternate expected_codes.
The health check should do enough internal checks to ensure the application is healthy and no more. This may mean ensuring database or other external storage connections are up and running, server load is acceptable, the site is not in maintenance mode, and other tests specific to your application.
The page generated by the health check should be very light weight:
- It should return in a sub-second interval.
- It should not induce significant load on the application server.
The page generated by the health check should never be cached, though the code running the health check may reference cached data. For example, you may find it useful to run a more extensive health check via cron and store the results of this to disk. The code generating the page at the health monitor url_path would incorporate the results of this cron job in the tests it performs.
Since Octavia only cares about the HTTP status code returned, and since health checks are run so frequently, it may make sense to use the “HEAD” or “OPTIONS” HTTP methods to cut down on unnecessary processing of a whole page.

Other heath monitors¶

Other health monitor types include PING, TCP, HTTPS, TLS-HELLO, and UDP-CONNECT.

PING health monitors send periodic ICMP PING requests to the back-end servers. Obviously, your back-end servers must be configured to allow PINGs in order for these health checks to pass.

TCP health monitors open a TCP connection to the back-end server’s protocol port. Your custom TCP application should be written to respond OK to the load balancer connecting, opening a TCP connection, and closing it again after the TCP handshake without sending any data.

HTTPS health monitors operate exactly like HTTP health monitors, but with ssl back-end servers. Unfortunately, this causes problems if the servers are performing client certificate validation, as HAProxy won’t have a valid cert. In this case, using TLS-HELLO type monitoring is an alternative.

TLS-HELLO health monitors simply ensure the back-end server responds to SSLv3 client hello messages. It will not check any other health metrics, like status code or body contents.

UDP-CONNECT health monitors do a basic UDP port connect. Health monitors of this type may not work correctly if Destination Unreachable (ICMP type 3) is not enabled on the member server or is blocked by a security rule. A member server may be marked as operating status ONLINE when it is actually down.

Intermediate certificate chains¶

Some TLS certificates require you to install an intermediate certificate chain in order for web client browsers to trust the certificate. This chain can take several forms, and is a file provided by the organization from whom you obtained your TLS certificate.

PEM-encoded chains¶

The simplest form of the intermediate chain is a PEM-encoded text file that either contains a sequence of individually-encoded PEM certificates, or a PEM encoded PKCS7 block(s). If this is the type of intermediate chain you have been provided, the file will contain either -----BEGIN PKCS7----- or -----BEGIN CERTIFICATE----- near the top of the file, and one or more blocks of 64-character lines of ASCII text (that will look like gobbedlygook to a human). These files are also typically named with a .crt or .pem extension.

DER-encoded chains¶

If the intermediates chain provided to you is a file that contains what appears to be random binary data, it is likely that it is a PKCS7 chain in DER format. These files also may be named with a .p7b extension.

You may use the binary DER file as-is when building your PKCS12 bundle:

openssl pkcs12 -export -inkey server.key -in server.crt -certfile ca-chain.p7b -passout pass: -out server.p12

… or you can convert it to a series of PEM-encoded certificates:

openssl pkcs7 -in intermediates-chain.p7b -inform DER -print_certs -out intermediates-chain.crt

… or you can convert it to a PEM-encoded PKCS7 bundle:

openssl pkcs7 -in intermediates-chain.p7b -inform DER -outform PEM -out intermediates-chain.crt

If the file is not a PKCS7 DER bundle, either of the two openssl pkcs7 commands will fail.