Enabling load balancer metrics
This feature is available in select Cleura Cloud regions. Please check the feature support matrix for details.
Once you have created load balancers in Cleura Cloud, you can add a special listener that exposes load balancer metrics. This listener is meant to be used as a data source for an existing Prometheus monitoring system.
Prerequisites
The Cleura Cloud Management Panel does not support defining metrics endpoints, so you will have to work with the OpenStack CLI.
Enable it for the region you will be working in, and make sure you have the Python octaviaclient package installed.
For that, use either the package manager of your operating system, or pip:
apt install python3-octaviaclient
This particular Python module is unavailable via brew, but you
can install it via pip.
pip install python-octaviaclient
Assumptions and scenario
We assume you already have an HTTPS-terminated load balancer that forwards client requests to a back-end server pool.
In our test scenario, the load balancer was accepting HTTPS requests for whoogle.example.com and forwarding them to a two-member pool, with servers each running a Docker container for Whoogle Search.
Creating a metrics listener
Here is mylb, the load balancer we used during testing:
$ openstack loadbalancer list
+---------------+------+---------------+--------------+---------------------+------------------+----------+
| id | name | project_id | vip_address | provisioning_status | operating_status | provider |
+---------------+------+---------------+--------------+---------------------+------------------+----------+
| eaf6d4f3- | mylb | dfc7004673964 | 10.15.25.155 | ACTIVE | ONLINE | amphora |
| 8d73-4b77- | | 28bacba4376e7 | | | | |
| 8bc4- | | 2cc3e9 | | | | |
| 788a3b4e3916 | | | | | | |
+---------------+------+---------------+--------------+---------------------+------------------+----------+
You may now add mylb-listener-metrics, a new Prometheus-based listener for mylb.
Most likely, you will want to restrict access to this endpoint to just the IP address of your Prometheus server, which presumably lives on a different network and accesses the load balancer via its public (floating) IP address.
You can use the --allowed-cidr command-line option for that purpose.
The example below1 assumes that your Prometheus server’s outgoing IP address is 203.0.113.132:
$ openstack loadbalancer listener create \
--name mylb-listener-metrics \
--protocol PROMETHEUS \
--protocol-port 8088 \
--allowed-cidr "203.0.113.132/32" \
mylb
+-----------------------------+--------------------------------------+
| Field | Value |
+-----------------------------+--------------------------------------+
| admin_state_up | True |
| connection_limit | -1 |
| created_at | 2023-01-29T16:12:24 |
| default_pool_id | None |
| default_tls_container_ref | None |
| description | |
| id | c1bba40d-3b45-4b17-86fd-81d98909822e |
| insert_headers | None |
| l7policies | |
| loadbalancers | eaf6d4f3-8d73-4b77-8bc4-788a3b4e3916 |
| name | mylb-listener-metrics |
| operating_status | OFFLINE |
| project_id | dfc700467396428bacba4376e72cc3e9 |
| protocol | PROMETHEUS |
| protocol_port | 8088 |
| provisioning_status | PENDING_CREATE |
| sni_container_refs | [] |
| timeout_client_data | 50000 |
| timeout_member_connect | 5000 |
| timeout_member_data | 50000 |
| timeout_tcp_inspect | 0 |
| updated_at | None |
| client_ca_tls_container_ref | None |
| client_authentication | NONE |
| client_crl_container_ref | None |
| allowed_cidrs | ['203.0.113.132/32'] |
| tls_ciphers | None |
| tls_versions | None |
| alpn_protocols | None |
| tags | |
+-----------------------------+--------------------------------------+
To check the provisioning status of mylb-listener-metrics, type:
$ openstack loadbalancer listener show mylb-listener-metrics \
-c provisioning_status
+---------------------+--------+
| Field | Value |
+---------------------+--------+
| provisioning_status | ACTIVE |
+---------------------+--------+
Testing the listener
Once the listener reports as being ACTIVE, you should be able to check its endpoint.
From a client that matches your allowed_cidrs filter, you can do so using curl.
The example below assumes that your load balancer uses a public (floating) IP of 198.51.100.234:
$ curl -s http://198.51.100.234:8088/metrics | head
# HELP octavia_loadbalancer_cpu Load balancer CPU utilization (percentage).
# TYPE octavia_loadbalancer_cpu gauge
octavia_loadbalancer_cpu 0.0
# HELP octavia_loadbalancer_memory Load balancer memory utilization (percentage).
# TYPE octavia_loadbalancer_memory gauge
octavia_loadbalancer_memory 37.2
# HELP octavia_memory_pool_failures_total Total number of failed memory pool allocations.
# TYPE octavia_memory_pool_failures_total counter
octavia_memory_pool_failures_total 0
# HELP octavia_loadbalancer_max_connections Hard limit on the number of per-process connections (configured or imposed by Ulimit-n)
Overall, the metrics endpoint exports more than 140 different metrics of the gauge and counter types.
All exported metrics use the prefix octavia_.
Adding the listener to your Prometheus configuration
Once you have verified that your listener is functional, you can add it to your Prometheus configuration.
Again, this example assumes that your load balancer uses a public (floating) IP of 198.51.100.234:
[scrape_configs]
- job_name: 'mylb'
static_configs:
- targets:
- '198.51.100.234:8088'
-
If your
openstackCLI does not support the--protocol PROMETHEUSoption, you may have to upgrade your installedpython-octaviaclientpackage. ↩