arc/docs/monitoring_metrics/faq.rst

Frequently Asked Questions
==========================

Here are a list of questions and answers that should help you get start with monitoring
with zabbix and prometheus.

How do I access zabbix?
-----------------------

1. First obtain Kerberos ticket with kinit:

.. code-block::

    $ kinit myusername@FEDORAPROJECT.ORG
    Password for myusername@FEDORAPROJECT.ORG:

2. Login to https://zabbix.stg.fedoraproject.org/zabbix.php?action=dashboard.view to see
   dashboard
3. If you need to be added in special privilege group (to see specific systems metrics),
   Open a PR in <path-to-inventory> with your FAS id in the list under the group and ask
   sysadmin of that groups to +1.

How do I access zabbix when I'm a community member?
---------------------------------------------------

1. First obtain Kerberos ticket with kinit:

.. code-block::

    $ kinit myusername@FEDORAPROJECT.ORG
    Password for myusername@FEDORAPROJECT.ORG:

2. Login to https://zabbix.stg.fedoraproject.org/zabbix.php?action=dashboard.view to see
   guest/public dashboard

How do I access Prometheus?
---------------------------

Prometheus is running in the application monitoring namespace, standard routing applies,
i.e.: https://prometheus-route-application-monitoring.app.os.stg.fedoraproject.org/graph

To access it you need to have account in the openshift it is running in.

How do I access Prometheus when I'm a community member?
-------------------------------------------------------

You shouldn't access prometheus directly, unless you are maintaining an application in
openshift.

Data from prometheus can be exported and viewed in Grafana or Zabbix, meaning we can
give access to a more limited public view through dashboards in one of these.

Do you have a 5 minutes guide on how to use prometheus?
-------------------------------------------------------

In other words, do you have some how-tos/links I should read to understand/get started
with prometheus?

- quick introduction to the stack we are running:
  https://www.youtube.com/watch?v=-37OPXXhrTw
- to get idea on how to use it, look at sample queries:
  https://prometheus.io/docs/prometheus/latest/querying/examples/
- for instrumentation, look at the libraries in https://github.com/prometheus/

How do I get basic HW (disk, cpu, memory, network...) monitoring for a host?
----------------------------------------------------------------------------

There are out of the box template for most of basic monitoring requirement that can be
seen on the web UI once you run the zabbix-agent-role against the node. if you want to
send any custom metrics, we recommend zabbix-sender. Zabbix sender is a command line
utility that may be used to send performance data to zabbix server for processing.
Adding the zabbix sender command in crontab is one way of continuously sending data to
server that can processed on server side (in your web UI). See
https://www.zabbix.com/documentation/current/manpages/zabbix_sender

How do I monitor a list of services?
------------------------------------

- pagure.io and src.fp.o have two different list of services to monitor
      they partly overlap but aren't exactly the same, how can I monitor them?
- For prometheus, metrics exported are usually done by instrumentation, meaning if i.e.
  pagure was instrumented to export /metrics endpoint, you just need to make sure you
  are collecting them, either because they run in openshift, and you configured
  appropriate ServiceMonitor or PodMonitor objects, or if outside of openshift, it is in
  additional scrape configuration of prometheus. Because collected metrics are labeled,
  it is simple to distinguish which belong where.
- For Zabbix, if you want to send any custom metrics, we recommend zabbix-sender. Zabbix
  sender is a command line utility that may be used to send performance data to zabbix
  server for processing. Adding the zabbix sender command in crontab is one way of
  continuously sending data to server that can processed on server side (in your web
  UI). See https://www.zabbix.com/documentation/current/manpages/zabbix_sender

How do I get alerted for a service not running?
-----------------------------------------------

- Prometheus supports configuring rules for alert-manager that can then notify through
  various services. You can learn about the configuration here:
  https://prometheus.io/docs/alerting/latest/configuration/#configuration-file The rules
  specifying when to alert are done in prometheus itself :
  https://prometheus.io/docs/prometheus/latest/configuration/alerting_rules/ You can
  specify them in CRDs in your project in simmilar fashion as with Service Monitor To
  use IRC, there needs to be a separate gateway installed in a sidecar:
  https://github.com/google/alertmanager-irc-relay
- In Zabbix, you can set custom alerting for yourself (or for groups through web UI).
  Follow https://www.zabbix.com/documentation/5.0/manual/config/triggers/trigger

How can I tune the alerts?
--------------------------

As in, who gets alerted? When? How?

- In Zabbix, we will have different groups with different configurations. When you are
  added in that group, you will receive notifications relevant to that group (you can
  change what alerting you want for the group once you have access to that). You can
  filter down the alerting even more for yourself in web UI. Follow this tutorial:
  https://www.zabbix.com/documentation/5.0/manual/config/triggers/trigger If you want to
  tweak how you receive your alerts, follow
  https://www.zabbix.com/documentation/5.0/manual/config/notifications/media

How do I ask for the service to be restarted <X> times before being alerted?
----------------------------------------------------------------------------

- In prometheus you can't. It is assumed you are using kubernetes that would manage
  something like this for you.
- In zabbix, <TODO>, you can do events based on triggers and there are event correlation
  options but yet to figure out this customization

How do I monitor rabbitmq queues?
---------------------------------

- In prometheus, according to
  https://www.rabbitmq.com/prometheus.html#overview-prometheus you just need to make
  sure you are collecting the exported metrics.
- In Zabbix, according to https://www.zabbix.com/integrations/rabbitmq, there is a way
  to build push data to zabbix that can be processed on server side

How do we alert about checks not passing to people outside of our teams?
------------------------------------------------------------------------

    -> the OSCI team is interesting in having notifications/monitoring for the CI
        queues in rabbitmq

How can we chain a prometheus instance to ours?
-----------------------------------------------

This allows to consolidate in a single instance monitoring coming from different
instances. This can be done with configuring federation in additional scrape configs:
https://prometheus.io/docs/prometheus/latest/federation/

How can I monitor the performances of my application?
-----------------------------------------------------

Number of requests served? Number of 500 errors? Number of DB connections?

With prometheus, you need to instrument your application and configure prometheus t
collect its metrics.

How do I ack an alert so it stops alerting?
-------------------------------------------

With prometheus and Alertmanager, there is no way to just ACK an alert, it is assumed
that something more high-level like opsgenie would take care of actually interacting
with regular human ops people.

For small enough teams, just using silence on alert in alertmanager could be enough.

There is a sidecar that serves to provide a little bit more features to the barebones
alerting. like https://github.com/prymitive/kthxbye.

- In Zabbix, you can acknowledge the problem and it will stop alerting. Follow
  https://www.zabbix.com/documentation/current/manual/acknowledges

How do I pre-emptively stop a check before I start working on an outage?
------------------------------------------------------------------------

In other words: I know that I'll cause an outage while working on <service>, how do I
turn off the checks for this service to avoid notifying admins while I'm working on it?

In Prometheus and Alertmanager there are Silences, where you can set a time when certain
alerts wouldn't be firing. You are able to create and remove these through rest api,

- In Zabbix, simplest way is to stop zabbix agent (or custom sender) on the system and
  ack on server side that it's not reachable.