* This is a bit tricky, each consul node will report that it knows about
the primary node
* So this alert assumes that there will be quorum
* We check to see if there's 0 reporting `passing`
* I linked to the main postgresql document due to troubleshooting this
being quite tricky
* Closes: https://gitlab.com/gitlab-com/gl-infra/infrastructure/issues/5358

useful.

This MR also removes the concept of `info` criticality alerts which don't
make much sense.

The `postgres:up` recording relied on `info` alerts but was the only thing
to do so.

* Drop obsoletely `node_cpu` metric recordings.
* Drop `CPUOutlierDetectionOnPrd` that doesn't work due to missing
recordings.
* Add new 1m rate recordings, with a 1m interval.
* Move CPU alerts to new metrics.
* Drop environment filter from CPU alerts.
* Drop 80% CPU threshold for "High CPU" to avoid alert noise.
* Move old 5m alerting to separate rule group.

* Move haproxy recording rules into haproxy rule group.
* Move remaining 'recording.yml' into rule group for generic process
metrics.

In future we want to pagerduty on critical alerts, not on alerts that
are set to `pager=pagerduty`. This is a first step in that direction.
Pagerduty alerts are critical alerts. Critical alerts are pagerduty
alerts.

As a next step, we can simply start routing alerts to pagerduty when
severity is critical and drop the pager attribute altogether.

Split the Gitaly rule groups to reduce execution time per group.

Ben Kochie authored Oct 18, 2018 and John Skarbek committed Oct 18, 2018

* Remove obsolete Prometheus 1.x local storage alerts.
* Simplify queries.
* Add an alert for rule group evaluation taking longer than 70% of the
interval.
* Update the slow rule documentation.

John T Skarbek authored Oct 18, 2018 and Alex Hanselka committed Oct 18, 2018

* This node is no longer serving as an archive replica
* Now he's participating as a node in the cluster
* Removes an alert that was crafted for him as an archive replica
* Removes him from the exception list of the rest of our alerts