metrics

An ansible role which configures performance analysis services for the managed host. This (optionally) includes a list of remote systems to be monitored by the managed host.

Requirements

Performance Co-Pilot (PCP) v5+. All of the packages are available from the standard repositories on Fedora, CentOS 8, and RHEL 8. On RHEL 7 and RHEL 6, you will need to enable the Optional repository/channel on the managed host.

The role can optionally use Grafana v6+ (metrics_graph_service) and Redis v5+ (metrics_query_service) on Fedora, CentOS 8, RHEL 8 and later.

Collection requirements

The role requires the firewall role and the selinux role from the fedora.linux_system_roles collection, if metrics_manage_firewall and metrics_manage_selinux is set to true, respectively. (Please see also the variables in the Role Variables section.)

If the metrics is a role from the fedora.linux_system_roles collection or from the Fedora RPM package, the requirement is already satisfied.

The role requires additional collections to manage rpm-ostree systems. If you need to manage rpm-ostree systems, run the below command to install the collections.

ansible-galaxy collection install -r meta/collection-requirements.yml

Role Variables

metrics_monitored_hosts: []

List of remote hosts to be analysed by the managed host. These hosts will have metrics recorded on the managed host, so care should be taken to ensure sufficient disk space exists below /var/log for each host.

Example:

metrics_monitored_hosts: ["webserver.example.com", "database.example.com"]

metrics_webhook_endpoint: ''

Webhook endpoint (URL) where notification about any automatically detected performance issues are to be sent. By default, these events are logged to the local system log only.

metrics_retention_days: 14

Retain historical performance data for the specified number of days; after this time it will be removed (day by day).

metrics_graph_service: false

Boolean flag allowing host to be setup with graphing services. Enabling this starts PCP and Grafana servers for visualizing PCP metrics. This option requires Grafana v6+ which is available on Fedora, CentOS 8, RHEL 8, or later versions of these platforms.

metrics_query_service: false

Boolean flag allowing host to be setup with time series query services. Enabling this starts PCP and Redis servers for querying recorded PCP metrics. This option requires Redis v5+ which is available on Fedora, CentOS 8, RHEL 8, or later versions of these platforms.

metrics_into_elasticsearch: false

Boolean flag allowing metric values to be exported into Elasticsearch.

metrics_from_elasticsearch: false

Boolean flag allowing metrics from Elasticsearch to be made available.

metrics_from_postfix: false

Boolean flag allowing metrics from Postfix to be made available.

metrics_from_mssql: false

Boolean flag allowing metrics from SQL Server to be made available. Enabling this flag requires a 'trusted' connection to SQL Server.

metrics_from_bpftrace: false

Boolean flag allowing metrics from bpftrace to be made available.

metrics_username: metrics

An account to establish authenticated access to remote metrics via the PCP pmcd daemon. For more information, see https://pcp.readthedocs.io/en/latest/QG/AuthenticatedConnections.html.

Additionally, if the bpftrace metrics are configured, this user account will be able to register bpftrace scripts.

metrics_password: metrics

Do not use a clear text metrics_password. Use Ansible Vault to encrypt the password.

Mandatory authentication for executing dynamic bpftrace scripts.

metrics_provider: pcp

The metrics collector to use to provide metrics.

Currently Performance Co-Pilot is the only supported metrics provider. When using the PCP provider these TCP ports will be used - 44321 (pmcd, live metric value sampling), 44322 (pmproxy, with metrics_query_service or metrics_graph_service), 6379 (redis-server for metrics_query_service) and 3000 (grafana-server for metrics_graph_service).

metrics_manage_firewall: false

Boolean flag allowing to configure firewall using the firewall role. Manage the pmcd port, the pmproxy port, the Grafana port and the Redis port depending upon the configuration parameters. If the variable is set to false, the metrics role does not manage the firewall.

NOTE: metrics_manage_firewall is limited to adding ports. It cannot be used for removing ports. If you want to remove ports, you will need to use the firewall system role directly.

NOTE: the firewall management is not supported on RHEL 6.

metrics_manage_selinux: false

Boolean flag allowing to configure selinux using the selinux role. Assign the pmcd port, the pmproxy port, the Grafana port and the Redis port depending upon the configuration parameters. If the variable is set to false, the metrics role does not manage the selinux.

Please note that the pmcd and pmproxy services are in the "ephemeral" range requiring no special setup and the Grafana port is "unregistered". The Redis port is gated by the redis_port_t SELinux type and may need to be further configured, if you require direct access (not required if you are accessing it from metrics role tools like Grafana and PCP). Use the selinux system role to manage port access, for SELinux contexts.

NOTE: metrics_manage_selinux is limited to adding policy. It cannot be used for removing policy. If you want to remove policy, you will need to use the selinux system role directly.

Example Playbook

Basic metric recording setup for each managed host only, with one weeks worth of data retained before culling.

---
- name: Manage metrics service
  hosts: all
  vars:
    metrics_retention_days: 7
  roles:
    - linux-system-roles.metrics

Scalable metric recording, analysis and visualization setup for the managed hosts, providing a REST API server with an OpenMetrics endpoint, graphs and scalable querying.

---
- name: Manage metrics with graph and query services
  hosts: all
  vars:
    metrics_graph_service: true
    metrics_query_service: true
  roles:
    - linux-system-roles.metrics

Centralized metric recording setup for several remote hosts and scalable metric recording, analysis and visualization setup for the local host, providing a REST API server with an OpenMetrics endpoint, graphs and scalable querying.

---
- name: Manage centralized metrics gathering
  hosts: monitors
  vars:
    metrics_monitored_hosts: [app.example.com, db.example.com, nas.example.com]
    metrics_graph_service: true
    metrics_query_service: true
  roles:
    - linux-system-roles.metrics

rpm-ostree

See README-ostree.md

License

MIT

kurik/metrics