Remove unit from alerts

[dstathis]

No description provided.

[balbirthomas]

@dstathis It would help in review and for posterity if your commit message included

• A description of the bug
• An explanation why this commit is the right fix for the problem.

[balbirthomas]

For whatever it is worth. Here is my analysis of the impacts of the changes introduced by this commit. I am still going through to double check these so take it with a pinch of salt.

1. JujuTopology.as_promql_label_dict() removes (pop("juju_unit") juju unit
   topology information. as_promql_label_dict(). The impacts of this change are

• JujuTopology.identifer() uses as_promql_label_dict() - for these impacts
  see point 2 below.

• JujuTopology.promql_labels() uses as_promql_label_dict() - for these impacts
  see point 3 below.

2. The JujuTopology.identifer() method previously used as_dict(). Now it
   uses it uses as_promql_label_dict(). as_dict() included juju_unit but
   as_promql_label_dict() does not. The impact of this change is as follows

• ProviderTopology.scrape_identifier() uses JujuTopology.identifer().
  Previously scrape_identifer() included juju unit now it does not.
  scrape_identifier() is used by MetricsEndpointConsumer to set the
  prefix of the Prometheus scrape job name only for jobs forwarded by
  MetricsEndpointProvider (but not for jobs sent by MetricsEndpointAggregator)

• The AlertRules._group_name() uses JujuTopology.identier()
  to construct a name for an alert rule group. AlertRules can be instantiated
  with any subclass of JujuTopology. However AlertRules is only used by
  MetricsEndpointProvider and PrometheusRulesProvider but not by
  MetricsEndpointAggregator.

  The alert rule group names constructed by AlertRules will never container
  juju unit regardless of which kind of topology (ProviderTopology or
  AggregatorTopology is used)

• MetricsEndpointConsumer.alerts() uses JujuTopology.identifer() to build
  an indexed list of alerts, which are indexed using the "identifier". This index
  will no longer contain juju unit topology information, for alert rules from both
  MetricsEndpointProvider and MetricsEndpointAggregator.

• MetricsEndpointConsumer._static_scrape_config() uses JujuTopology.identifier()
  for constructing the scrape job name on for jobs provided by MetricsEndpointProvider.
  After this commit these job names will not include juju unit topology information.

3. JujuTopology.promql_labels() previously previously used
   JujuTopology.as_dict() but now it uses JujuTopology.as_promql_label_dict().
   The impact of this change is as follows.

• JujuTopology.promql_labels() is only used in JujuTopology.render() method
  which is used to inject topology information into an alert rule expression by
  the AlertRules object. Hence AlertRules object will never inject juju unit
  topology into any alert rule read by it.

4. ProviderTopology.scrape_identifer() now uses JujuTopology.identier() Since
   ProviderTopology previously constructed the identifer without juju unit information
   this change may not have any impacts.

[simskij]

once tests pass and the conversations have been properly resolved, I'm happy to approve.

All objections resolved

[rbarry82]

Just quibbling about the alert name. Is the change to tox.ini intentional?