worker_plugins 0.0.16 → 0.0.17

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 343e7440c6234a58f501c96d553f019c653c7eb32c8dc07847d4f02d9ee8020b
-  data.tar.gz: 5100e6c819ee4f818082e87ef515e7362523b7e1c5b0c25ca386741e07e9d8ec
+  metadata.gz: 0724aebdce87819ebafc787c475284e25e7a6ee31837040e8532b0c400c95db3
+  data.tar.gz: dc8cd4edddcdce80d22c2e3c10990f4fa966f112fe67f01c1bd58443c313079b
 SHA512:
-  metadata.gz: a658e1f90e8f9c33503e526b98a59ef578e9a366dc388d79abbf8efec602c60fe5c920afa4e0772a113de5e47f4d71eba4bd7346dfc38dcbefd31ac7808deca1
-  data.tar.gz: c73827b981b2d50abc9f5bd3c388645fa382f0d052188f3ff85f5f3e0e8ac1d47e62ba01a2fa73a2d1e6d4c78475870f36cbb520723154491bffe44245536db6
+  metadata.gz: f58a6c6a65122966a57b948f78758761825e2eed97bb13fb4277b627ec4db11bade6856b2baf23e1f1056fb7169d545bfeaf6d565bbbddbb027bfb61c9d16f26
+  data.tar.gz: a4fb5385013ab116c856bd5b5986ce04bcccbfb18b342edb41f463d652e664ecdd0b01567e75cf321f0299d64d8adfff5b37c9d2ef2ed75f81da31286975884a

data/README.md

@@ -67,6 +67,19 @@ end
 
 and schedule `Workplaces::DeleteOld` instead.
 
+## Scheduled cleanup of orphan links
+
+`WorkerPlugins::DeleteOrphanLinks` removes `worker_plugins_workplace_links` whose target row no longer exists — i.e. links that point at a resource that was destroyed without the link being cleaned up alongside. Run it periodically from a background job to keep the links table consistent and keep probes like `QueryLinksStatus#checked_count` honest.
+
+```ruby
+result = WorkerPlugins::DeleteOrphanLinks.execute!
+# => {deleted_count: <N>}
+```
+
+Links whose `resource_type` doesn't resolve to a Ruby class (e.g. a model was renamed or removed) are left alone — cleaning those up requires human judgement.
+
+Schedule the same way as `DeleteOldWorkplaces` — typically once a day off-hours.
+
 ## Release
 
 Run the release task from a clean worktree:

data/app/services/worker_plugins/delete_orphan_links.rb

@@ -0,0 +1,39 @@
+class WorkerPlugins::DeleteOrphanLinks < WorkerPlugins::ApplicationService
+  # Deletes `worker_plugins_workplace_links` whose target row no longer
+  # exists — i.e. links pointing at a resource that was destroyed without
+  # the link being cleaned up alongside. Intended to be scheduled by the
+  # consumer application from a background job.
+  #
+  # Links whose `resource_type` doesn't resolve to a Ruby class (e.g. a
+  # model was renamed or removed) are left alone — cleaning those up is
+  # a separate concern that requires human judgement.
+  def perform
+    deleted_count = distinct_resource_types.sum do |resource_type|
+      delete_orphans_for(resource_type)
+    end
+
+    succeed!(deleted_count:)
+  end
+
+  def distinct_resource_types
+    WorkerPlugins::WorkplaceLink.distinct.pluck(:resource_type)
+  end
+
+  def delete_orphans_for(resource_type)
+    model_class = resource_type.safe_constantize
+    return 0 unless model_class
+
+    WorkerPlugins::WorkplaceLink
+      .where(resource_type:)
+      .where.not(resource_id: live_ids_query(model_class))
+      .delete_all
+  end
+
+  def live_ids_query(model_class)
+    WorkerPlugins::SelectColumnWithTypeCast.execute!(
+      column_name_to_select: model_class.primary_key,
+      column_to_compare_with: WorkerPlugins::WorkplaceLink.column_for_attribute(:resource_id),
+      query: model_class.all
+    )
+  end
+end

data/app/services/worker_plugins/query_links_status.rb

@@ -3,7 +3,7 @@ class WorkerPlugins::QueryLinksStatus < WorkerPlugins::ApplicationService
 
   def perform
     query_count = query.count
-    checked_count = count_linked_rows
+    checked_count = count_linked_rows(query_count)
 
     succeed!(
       all_checked: query_count == checked_count,
@@ -13,46 +13,26 @@ class WorkerPlugins::QueryLinksStatus < WorkerPlugins::ApplicationService
     )
   end
 
-  def count_linked_rows
+  def count_linked_rows(query_count)
     base_scope = workplace.workplace_links.where(resource_type: query.klass.name)
 
-    # When the query applies no scoping, the original `resource_id IN (SELECT
-    # DISTINCT <target_table>.id FROM <target_table>)` subquery materialized
-    # every row of the target model just to count — 2+ seconds on a 340k-row
-    # target. We drop the DISTINCT subquery and instead `INNER JOIN` the
-    # target table on its primary key so the composite index on links drives
-    # the scan and each matching link does a cheap PK probe to confirm the
-    # target row still exists. Orphaned links (whose target has since been
-    # deleted) are correctly excluded from the count, so `checked_count`
-    # never exceeds `query_count`.
-    return base_scope.joins(unscoped_target_join_sql).count if relation_unscoped?(query)
-
-    base_scope.where(resource_id: query_with_selected_ids).count
-  end
-
-  def unscoped_target_join_sql
-    target_table = quote_table(query.klass.table_name)
-    target_pk = "#{target_table}.#{quote_column(query.klass.primary_key)}"
-    resource_id_column = "#{quote_table(WorkerPlugins::WorkplaceLink.table_name)}.#{quote_column(:resource_id)}"
-
-    "INNER JOIN #{target_table} ON #{target_pk} = #{resource_id_expression_for_join(resource_id_column)}"
-  end
-
-  # On MySQL / MariaDB and SQLite, implicit conversion handles comparing the
-  # target's primary key against the VARCHAR `resource_id` column. Postgres
-  # is strict about types and needs an explicit cast when they differ.
-  def resource_id_expression_for_join(resource_id_column)
-    return resource_id_column unless postgres?
-
-    target_pk_type = query.klass.column_for_attribute(query.klass.primary_key).type
-    resource_id_type = WorkerPlugins::WorkplaceLink.column_for_attribute(:resource_id).type
-
-    return resource_id_column if target_pk_type == resource_id_type
-
-    case target_pk_type
-    when :uuid then "CAST(#{resource_id_column} AS UUID)"
-    when :integer then "CAST(#{resource_id_column} AS BIGINT)"
-    else resource_id_column
+    # Fast path for unscoped queries: a plain index-only COUNT against the
+    # `(workplace_id, resource_type, resource_id)` composite index resolves in
+    # ~50 ms even for workplaces with hundreds of thousands of links. Joining
+    # back to the target table to exclude orphaned links (whose resource row
+    # has since been destroyed) would take 10+ seconds on the same data
+    # regardless of which shape we pick — the DB still has to cross-reference
+    # every link against the target's primary key. Instead we clamp the raw
+    # count to `query_count`; `WorkerPlugins::DeleteOrphanLinks` (scheduled
+    # daily by consumers) keeps the orphan count at zero so the raw count
+    # equals the live-linked count in practice. When orphans do briefly exist
+    # between cleanup runs, clamping bounds the over-count at the query's
+    # total — `all_checked` / `some_checked` stay correct because they're
+    # computed off the clamped value.
+    if relation_unscoped?(query)
+      [base_scope.count, query_count].min
+    else
+      base_scope.where(resource_id: query_with_selected_ids).count
     end
   end
 

data/lib/worker_plugins/version.rb

@@ -1,3 +1,3 @@
 module WorkerPlugins
-  VERSION = "0.0.16".freeze
+  VERSION = "0.0.17".freeze
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: worker_plugins
 version: !ruby/object:Gem::Version
-  version: 0.0.16
+  version: 0.0.17
 platform: ruby
 authors:
 - Kasper Stöckel
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-04-23 00:00:00.000000000 Z
+date: 2026-04-25 00:00:00.000000000 Z
 dependencies: []
 description: Rails framework for easily choosing and creating lists of objects and
   execute plugins against them.
@@ -27,6 +27,7 @@ files:
 - app/services/worker_plugins/add_query.rb
 - app/services/worker_plugins/application_service.rb
 - app/services/worker_plugins/delete_old_workplaces.rb
+- app/services/worker_plugins/delete_orphan_links.rb
 - app/services/worker_plugins/query_links_status.rb
 - app/services/worker_plugins/remove_query.rb
 - app/services/worker_plugins/select_column_with_type_cast.rb