Deduplicate Geo cronjobs

The default deduplication setting is `:until_executing`. For cronjobs
which are enqueued every minute, we can confidently take it a bit further
and use `:until_executed`.

This should be slightly more efficient, and it should help some edge
cases like https://gitlab.com/gitlab-org/gitlab/-/issues/328057.

The modified classes are inherited in the following class files, as shown in
grep output:

```
./ee/app/workers/geo/scheduler/primary/scheduler_worker.rb:      class SchedulerWorker < Geo::Scheduler::SchedulerWorker
./ee/app/workers/geo/scheduler/primary/per_shard_scheduler_worker.rb:      class PerShardSchedulerWorker < Geo::Scheduler::PerShardSchedulerWorker
./ee/app/workers/geo/scheduler/secondary/scheduler_worker.rb:      class SchedulerWorker < Geo::Scheduler::SchedulerWorker
./ee/app/workers/geo/scheduler/secondary/per_shard_scheduler_worker.rb:      class PerShardSchedulerWorker < Geo::Scheduler::PerShardSchedulerWorker
./ee/app/workers/geo/repository_sync_worker.rb:  class RepositorySyncWorker < Geo::Scheduler::Secondary::PerShardSchedulerWorker
./ee/app/workers/geo/registry_sync_worker.rb:  class RegistrySyncWorker < Geo::Scheduler::Secondary::SchedulerWorker
./ee/app/workers/geo/repository_verification/primary/shard_worker.rb:      class ShardWorker < Geo::Scheduler::Primary::SchedulerWorker
./ee/app/workers/geo/repository_verification/primary/batch_worker.rb:      class BatchWorker < Geo::Scheduler::Primary::PerShardSchedulerWorker
./ee/app/workers/geo/repository_verification/secondary/shard_worker.rb:      class ShardWorker < Geo::Scheduler::Secondary::SchedulerWorker
./ee/app/workers/geo/repository_verification/secondary/scheduler_worker.rb:      class SchedulerWorker < Geo::Scheduler::Secondary::PerShardSchedulerWorker
./ee/app/workers/geo/container_repository_sync_dispatch_worker.rb:  class ContainerRepositorySyncDispatchWorker < Geo::Scheduler::Secondary::SchedulerWorker
./ee/app/workers/geo/repository_shard_sync_worker.rb:  class RepositoryShardSyncWorker < Geo::Scheduler::Secondary::SchedulerWorker
./ee/app/workers/geo/file_download_dispatch_worker.rb:  class FileDownloadDispatchWorker < Geo::Scheduler::Secondary::SchedulerWorker
```

Resolves https://gitlab.com/gitlab-org/gitlab/-/issues/328364

Changelog: performance

ee/app/workers/geo/scheduler/per_shard_scheduler_worker.rb

@@ -13,6 +13,13 @@ module Geo
 
       feature_category :geo_replication
 
+      # These workers are enqueued every minute by sidekiq-cron. If one of them
+      # is already enqueued or running, then there isn't a strong case for
+      # enqueuing another. And there are edge cases where enqueuing another
+      # would exacerbate a problem. See
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/328057.
+      deduplicate :until_executed
+
       def perform
         each_eligible_shard { |shard_name| schedule_job(shard_name) }
       end

ee/app/workers/geo/scheduler/scheduler_worker.rb

@@ -10,6 +10,14 @@ module Geo
       include ::Gitlab::Utils::StrongMemoize
       include GeoBackoffDelay
 
+      # These workers are enqueued regularly by sidekiq-cron or by an per-shard
+      # worker which is enqueued by sidekiq-cron. If one of these workers is
+      # already enqueued or running, then there isn't a strong case for
+      # enqueuing another. And there are edge cases where enqueuing another
+      # would exacerbate a problem. See
+      # https://gitlab.com/gitlab-org/gitlab/-/issues/328057.
+      deduplicate :until_executed
+
       DB_RETRIEVE_BATCH_SIZE = 1000
       LEASE_TIMEOUT = 60.minutes
       RUN_TIME = 60.minutes.to_i

ee/changelogs/unreleased/mk-geo-deduplicate-until-executed.yml

+---
+title: Deduplicate Geo cronjobs
+merge_request: 59799
+author:
+type: performance