This removes influxdb from our Gitlab::Metrics module. It also removes
influxdb sampling.

The following tasks now accept an `index_name` parameter to specify
the name of the index to be created or acted upon:

  - gitlab:elastic:create_empty_index
  - gitlab:elastic:delete_index
  - gitlab:elastic:recreate_index

The following obsolete task has been removed:

  - gitlab:elastic:reindex_to_another_cluster

Because otherwise helpers won't be verified, and since we're using
prepend to override direct methods mostly, it should work anyway.

We also skip instrumentation when we're doing static verification.

Also we need to move some methods from geo_git_access to git_access,
because that's where it's trying to override methods, not in geo itself.

This also assumes GitAccessWiki does not use those methods, of course.
It won't work anyway, because it's not getting methods from GitAccess!

The indexer now triggers indexing on all indices.
Since it is outside the common multi-indices architecture, we expose
elastic_write_targets as public so indexer can operate on indices.

Remove unused encoder

Usage removed when Ruby elasticsearch indexer was removed

There should be no cases where we need to inherit=true.

This changes any mention of gitlab-ce to gitlab-foss, and any mention of
gitlab-ee to just gitlab.

Doc for multi-indices archtecture

See https://gitlab.com/gitlab-org/gitlab-ee/merge_requests/14428

When there a large number of discussions on diffs, the instrumentation
has a significant impact on performance since it's often run in a tight
loop for line position calculation etc. We should disable this since
there are other ways to benchmark performance.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/52898

This commit fixes the following error in development:

    ArgumentError: A copy of Gitlab::Metrics::Transaction has been
    removed from the module tree but is still active!

Because this could potentially load a model and we shouldn't
load models before all the patches we have in places.

Normally, this is what's supposed to happen when you invoke `rails g
generate migration`:

1. `Rails.application.load_generators` is called here:
https://github.com/rails/rails/blob/v4.2.10/railties/lib/rails/commands/commands_tasks.rb#L129
2. This loads the local `generator_templates` path in
https://github.com/rails/rails/blob/v4.2.10/railties/lib/rails/generators.rb#L69.

However, when metrics are enabled, the application is eager loaded here
in the initializer. This short-circuits the `load_generators` and loads
`ActiveRecord::Generators::MigrationGenerator` automatically, so the
custom path is never loaded properly.

With this change, we skip the initializing of metrics and eager loading
if generators are in use.

Closes https://gitlab.com/gitlab-org/gitlab-ce/issues/52651

This allows us (and others) to test drive Puma without it affecting all
users. Puma can be enabled by setting the environment variable
"EXPERIMENTAL_PUMA" to a non empty value.

On a previous commit it was enabled but after the CE Upstream merge it
needed to be disabled again

Cleanup code, and refactor tests that still use Rugged. After this, there should
be no Rugged code that access the instance's repositories on non-test
environments. There is still some rugged code for other tasks like the
repository import task, but since it doesn't access any repository storage path
it can stay.

Cleanup code, and refactor tests that still use Rugged. After this, there should
be no Rugged code that access the instance's repositories on non-test
environments. There is still some rugged code for other tasks like the
repository import task, but since it doesn't access any repository storage path
it can stay.

The removal in 4d376c66 had an unintended
side effect in that it got rid of the `require 'rouge/plugins/redcarpet'`.

For one README.md file, this instrumentation was adding several seconds and
significant RAM and CPU overhead since these filters often iterate through
all HTML nodes in tight loops.

Relates to #49409

Signed-off-by: Rémy Coutable <remy@rymai.me>

also, I refactored the MergeRequest#fetch_ref method to express
the side-effect that this method has.

  MergeRequest#fetch_ref -> MergeRequest#fetch_ref!
  Repository#fetch_source_branch -> Repository#fetch_source_branch!

This reduces some polling overhead and reduces the need to aggregate
empty metrics from the Sidekiq worker.