Changelog: other

Addressing: https://gitlab.com/gitlab-org/gitlab/-/issues/513075
Changelog: other

We will log response sizes over the GITLAB_LOG_DECOMPRESSED_RESPONSE_BYTESIZE if set

Addressing: https://gitlab.com/gitlab-org/gitlab/-/issues/513075
Changelog: other

Now that we've updated net-http and Ruby, we can drop this spec,
which wasn't running in Ruby 3.2 anyway.

In addition,
https://gitlab.com/gitlab-org/gitlab/-/merge_requests/154182 dropped
the equivalent spec in
`gems/gitlab-http/spec/gitlab/http_v2/net_http_patch_spec.rb`.

Patch Net/HTTP to log the decompressed size of compressed HTTP responses

Also removing unneeded helpers + Adding validation on Cells configuration

Addressing: https://gitlab.com/gitlab-org/gitlab/-/issues/510193
Changelog: other

It overrides the Rails method to use Set instead of an Array

- https://github.com/rails/rails/blob/main/activemodel/lib/active_model/attribute_methods.rb#L383

In this case, when `alias_attribute` is called multiple times for the same attribute, the number
doesn't grow.

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

This reverts merge request !168773

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

Till now, we enabled all workers when a node was connected as a geo
node; we now only enable the needed geo workers. With that, we can
in the future disable the offline cloud license provision worker for SaaS.

This features is behind a feature flag disable_bulk_sidekiq_job_activation

EE: true
Changelog: added

The `` 1_settings settings do not change after reload ``
test matches one of the following conditions:
1. has either ~"flakiness::1" or ~"flakiness::2" label set, which means the number of reported failures
is at or above 95 percentile, indicating unusually high failure count.

2. has ~"severity::1" label set, which means the number of reported failures
[spiked and exceeded its daily threshold](https://gitlab.com/gitlab-org/ruby/gems/gitlab_quality-test_tooling/-/blob/c9bc10536b1f8d2d4a03c3e0b6099a40fe67ad26/lib/gitlab_quality/test_tooling/report/concerns/issue_reports.rb#L51).

This MR quarantines the test. This is a discussion starting point to let the
responsible group know about the flakiness so that they can take action:

- accept the merge request and schedule the associated issue to improve the test
- close the merge request in favor of another merge request to delete the test

Please follow the
[Flaky tests management process](https://handbook.gitlab.com/handbook/engineering/infrastructure/engineering-productivity/flaky-tests-management-and-processes/#flaky-tests-management-process)
to help us increase `master` stability.

Please let us know your feedback in the
[Engineering Productivity issue tracker](https://gitlab.com/gitlab-org/quality/engineering-productivity/team/-/issues).

Related to https://gitlab.com/gitlab-org/gitlab/-/issues/501317.

This change was generated by
[gitlab-housekeeper](https://gitlab.com/gitlab-org/gitlab/-/tree/master/gems/gitlab-housekeeper)
using the Keeps::QuarantineFlakyTests keep.

To provide feedback on your experience with `gitlab-housekeeper` please create an issue with the
label ~"GitLab Housekeeper" and consider pinging the author of this keep.

Changelog: other

This commit introduces the `omit_aggregated_db_log_fields` ops feature flag.

The state leak can be reproduced without the fix with the following
command:

```
bundle exec rspec spec/initializers/secret_token_spec.rb spec/lib/gitlab/otp_key_rotator_spec.rb
```

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

This change adds a `overrides` keyword argument to the
`data_consistency` method in `WorkerAttributes`. In
`MODE_MULTIPLE_DATABASES`, data consistency for specified databases are
overrriden.

For example, `data_consistency :always, overrides: { ci: :sticky }`
would stick main db connections to primary while ci db connections will
prefer the read replicas.

Changelog: other

We found that this setting can interfere with DNS
requests made by UrlBlocker when web proxies are
used.

Also extracted non-EE tests into the proper file.

Changelog: fixed

This reverts commit 550c01a2.

We've seen in development that secrets.yml quietly changed for some
reason. Back up this file in case the old values need to be restored.

Changelog: changed

This reverts merge request !166986

DB Rake tasks would establish a db connection with `name: primary`. As
seen in
https://github.com/rails/rails/blob/v7.0.8.4/activerecord/lib/active_record/tasks/postgresql_database_tasks.rb#L97

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

The categories product_analytics_data_management and
product_analytics_visualization have been renamed to product_analytics,
according to
https://gitlab.com/gitlab-com/www-gitlab-com/-/commit/9baeca3c4516740258e448654df9b6bcd21c85b4.

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

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

This fixes the problem of crons lingering after the worker class is
removed from Gitlab::SidekiqConfig.cron_jobs.

Changelog: fixed

When transactional tests are enabled, Rails ensures that all threads
use the same connection for accessing the database. However this may
cause a false positive with the Sidekiq worker open transaction
check because:

1. A browser may have multiple requests in-flight.
2. A transaction may start in the middle of another request. For example,
an update to a merge request column initiates a transaction via SAVEPOINT.

To avoid false positives, disable this check for JavaScript feature
specs run. The alternative is to use the deletion strategy, but that
would significantly slow tests.

Relates to https://gitlab.com/gitlab-org/gitlab/-/issues/478601

It removes newly introduced check that doesn't allow defining
an enum without backed column.

A migration job fails because it:

- Performs database migration on a particular version
- The version does not contain email_confirmation_setting column
- Performs another migration that initializes Rails env
- Initializers call Gitlab::CurrentSetting
- ApplicationSetting tries to define email_confirmation_setting enum
- Fails because the enum is not backed by a column