Prometheus Exporter¶

Prometheus Install

The information below describes how to setup Harvest's Prometheus exporter. If you need help installing or setting up Prometheus, check out their documention.

Overview¶

The Prometheus exporter is responsible for:

formatting metrics into the Prometheus line protocol
creating a web-endpoint on http://<ADDR>:<PORT>/metrics for Prometheus to scrape

A web end-point is required because Prometheus scrapes Harvest by polling that end-point.

In addition to the /metrics end-point, the Prometheus exporter also serves an overview of all metrics and collectors available on its root address http://<ADDR>:<PORT>/.

Because Prometheus polls Harvest, don't forget to update your Prometheus configuration and tell Prometheus how to scrape each poller.

There are two ways to configure the Prometheus exporter: using a port range or individual ports.

The port range is more flexible and should be used when you want multiple pollers all exporting to the same instance of Prometheus. Both options are explained below.

Parameters¶

All parameters of the exporter are defined in the Exporters section of harvest.yml.

An overview of all parameters:

parameter	type	description	default
`port_range`	int-int (range), overrides `port` if specified	lower port to upper port (inclusive) of the HTTP end-point to create when a poller specifies this exporter. Starting at lower port, each free port will be tried sequentially up to the upper port.
`port`	int, required if port_range is not specified	port of the HTTP end-point
`local_http_addr`	string, optional	address of the HTTP server Harvest starts for Prometheus to scrape: use `localhost` to serve only on the local machine use `0.0.0.0` (default) if Prometheus is scrapping from another machine	`0.0.0.0`
`global_prefix`	string, optional	add a prefix to all metrics (e.g. `netapp_`)
`allow_addrs`	list of strings, optional	allow access only if host matches any of the provided addresses
`allow_addrs_regex`	list of strings, optional	allow access only if host address matches at least one of the regular expressions
`cache_max_keep`	string (Go duration format), optional	maximum amount of time metrics are cached (in case Prometheus does not timely collect the metrics)	`300s`
`add_meta_tags`	bool, optional	add `HELP` and `TYPE` metatags to metrics (currently no useful information, but required by some tools)	`false`
`sort_labels`	bool, optional	sort metric labels before exporting. Some open-metrics scrapers report stale metrics when labels are not sorted.	`false`

A few examples:

port_range¶

Exporters:
  prom-prod:
    exporter: Prometheus
    port_range: 2000-2030
Pollers:
  cluster-01:
    exporters:
      - prom-prod
  cluster-02:
    exporters:
      - prom-prod
  cluster-03:
    exporters:
      - prom-prod
  # ... more
  cluster-16:
    exporters:
      - prom-prod

Sixteen pollers will collect metrics from 16 clusters and make those metrics available to a single instance of Prometheus named prom-prod. Sixteen web end-points will be created on the first 16 available free ports between 2000 and 2030 (inclusive).

After staring the pollers in the example above, running bin/harvest status shows the following. Note that ports 2000 and 2003 were not available so the next free port in the range was selected. If no free port can be found an error will be logged.

Datacenter   Poller       PID     PromPort  Status              
++++++++++++ ++++++++++++ +++++++ +++++++++ ++++++++++++++++++++
DC-01        cluster-01   2339    2001      running         
DC-01        cluster-02   2343    2002      running         
DC-01        cluster-03   2351    2004      running         
...
DC-01        cluster-14   2405    2015      running         
DC-01        cluster-15   2502    2016      running         
DC-01        cluster-16   2514    2017      running

allow_addrs¶

Exporters:
  my_prom:
    allow_addrs:
      - 192.168.0.102
      - 192.168.0.103

will only allow access from exactly these two addresses.

allow_addrs_regex¶

Exporters:
  my_prom:
    allow_addrs_regex:
      - `^192.168.0.\d+$`

will only allow access from the IP4 range 192.168.0.0-192.168.0.255.

Configure Prometheus to scrape Harvest pollers¶

There are two ways to tell Prometheus how to scrape Harvest: using HTTP service discovery (SD) or listing each poller individually.

HTTP service discovery is the more flexible of the two. It is also less error-prone, and easier to manage. Combined with the port_range configuration described above, SD is the least effort to configure Prometheus and the easiest way to keep both Harvest and Prometheus in sync.

NOTE HTTP service discovery does not work with Docker yet. With Docker, you will need to list each poller individually or if possible, use the Docker Compose workflow that uses file service discovery to achieve a similar ease-of-use as HTTP service discovery.

See the example below for how to use HTTP SD and port_range together.

Prometheus HTTP Service Discovery¶

HTTP service discovery was introduced in Prometheus version 2.28.0. Make sure you're using that version or later.

The way service discovery works is:

shortly after a poller starts up, it registers with the SD node (if one exists)
the poller sends a heartbeat to the SD node, by default every 45s.
if a poller fails to send a heartbeat, the SD node removes the poller from the list of active targets after a minute
the SD end-point is reachable via SCHEMA:///api/v1/sd

To use HTTP service discovery you need to:

tell Harvest to start the HTTP service discovery process
tell Prometheus to use the HTTP service discovery endpoint

Enable HTTP service discovery in Harvest¶

Add the following to your harvest.yml

Admin:
  httpsd:
    listen: :8887

This tells Harvest to create an HTTP service discovery end-point on interface 0.0.0.0:8887. If you want to only listen on localhost, use 127.0.0.1:<port> instead. See net.Dial for details on the supported listen formats.

Start the SD process by running bin/harvest admin start. Once it is started, you can curl the end-point for the list of running Harvest pollers.

curl -s 'http://localhost:8887/api/v1/sd' | jq .
[
  {
    "targets": [
      "10.0.1.55:12990",
      "10.0.1.55:15037",
      "127.0.0.1:15511",
      "127.0.0.1:15008",
      "127.0.0.1:15191",
      "10.0.1.55:15343"
    ]
  }
]

Harvest HTTP Service Discovery options¶

HTTP service discovery (SD) is configured in the Admin > httpsd section of your harvest.yml.

parameter	type	description	default
`listen`	required	Interface and port to listen on, use localhost:PORT or :PORT for all interfaces
`auth_basic`	optional	If present, enables basic authentication on `/api/v1/sd` end-point
auth_basic `username`, `password`	required child of `auth_basic`
`tls`	optional	If present, enables TLS transport. If running in a container, see note
tls `cert_file`, `key_file`	required child of `tls`	Relative or absolute path to TLS certificate and key file. TLS 1.3 certificates required. FIPS complaint P-256 TLS 1.3 certificates can be created with `bin/harvest admin tls create server`
`ssl_cert`, `ssl_key`	optional if `auth_style` is `certificate_auth`	Absolute paths to SSL (client) certificate and key used to authenticate with the target system. If not provided, the poller will look for `<hostname>.key` and `<hostname>.pem` in `$HARVEST_HOME/cert/`. To create certificates for ONTAP systems, see using certificate authentication
`heart_beat`	optional, Go Duration format	How frequently each poller sends a heartbeat message to the SD node	45s
`expire_after`	optional, Go Duration format	If a poller fails to send a heartbeat, the SD node removes the poller after this duration	1m

Enable HTTP service discovery in Prometheus¶

Edit your prometheus.yml and add the following section

$ vim /etc/prometheus/prometheus.yml

scrape_configs:
  - job_name: harvest
    http_sd_configs:
      - url: http://localhost:8887/api/v1/sd

Harvest and Prometheus both support basic authentication for HTTP SD end-points. To enable basic auth, add the following to your Harvest config.

Admin:
  httpsd:
    listen: :8887
    # Basic auth protects GETs and publishes
    auth_basic:
      username: admin
      password: admin

Don't forget to also update your Prometheus config with the matching basic_auth credentials.

Prometheus HTTP Service Discovery and Port Range¶

HTTP SD combined with Harvest's port_range feature leads to significantly less configuration in your harvest.yml. For example, if your clusters all export to the same Prometheus instance, you can refactor the per-poller exporter into a single exporter shared by all clusters in Defaults as shown below:

Notice that none of the pollers specify an exporter. Instead, all the pollers share the single exporter named prometheus-r listed in Defaults. prometheus-r is the only exporter defined and as specified will manage up to 1,000 Harvest Prometheus exporters.

If you add or remove more clusters in the Pollers section, you do not have to change Prometheus since it dynamically pulls the targets from the Harvest admin node.

Admin:
  httpsd:
    listen: :8887

Exporters:
  prometheus-r:
    exporter: Prometheus
    port_range: 13000-13999

Defaults:
  collectors:
    - Zapi
    - ZapiPerf
  use_insecure_tls: false
  auth_style: password
  username: admin
  password: pass
  exporters:
    - prometheus-r

Pollers:
  umeng_aff300:
    datacenter: meg
    addr: 10.193.48.11

  F2240-127-26:
    datacenter: meg
    addr: 10.193.6.61

  # ... add more clusters

Static Scrape Targets¶

If we define four prometheus exporters at ports: 12990, 12991, 14567, and 14568 you need to add four sections to your prometheus.yml.

$ vim /etc/prometheus/prometheus.yml

Scroll down to near the end of file and add the following lines:

  - job_name: 'harvest'
    scrape_interval: 60s
    static_configs:
      - targets:
          - 'localhost:12990'
          - 'localhost:12991'
          - 'localhost:14567'
          - 'localhost:14568'

NOTE If Prometheus is not on the same machine as Harvest, then replace localhost with the IP address of your Harvest machine. Also note the scrape interval above is set to 60s. That matches the polling frequency of the default Harvest collectors. If you change the polling frequency of a Harvest collector to a lower value, you should also change the scrape interval.

Prometheus Alerts¶

Prometheus includes out-of-the-box support for simple alerting. Alert rules are configured in your prometheus.yml file. Setup and details can be found in the Prometheus guide on alerting.

Harvest also includes ems alerts and sample alerts for reference. Refer EMS Collector for more details about EMS events.

Alertmanager¶

Prometheus's builtin alerts are good for simple workflows. They do a nice job telling you what's happening at the moment. If you need a richer solution that includes summarization, notification, advanced delivery, deduplication, etc. checkout Alertmanager.