switchman-inst-jobs 3.2.2 → 3.2.7

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 98c61abb5cfc89f0f020fdaf985b7dd97620813192052fa6859504d06b9d8394
-  data.tar.gz: d57c6382a723425eb15abbcc7bcd62869b3a31defc15df1339e2d81ca656d8cd
+  metadata.gz: ac66ce458b7356068a5f576593f9692b5aa58d608584033a3aca941b31dc4a2d
+  data.tar.gz: c790f410b55117257f9b447987f68617b61222b10bf621385736b096affa25f8
 SHA512:
-  metadata.gz: 31f03b8c1ccb6d95e7824671728e78ddb362b326fb007509db9b543621b17f22179d097f383e43c2b233fe78a563eec09a4fddc6d4af872cc4554fa1f68a6e88
-  data.tar.gz: 5a84c259b4dfe96df5d8cad336b0ba4555fb0f89e110d172f59fdbfeca0f090fed9ad9e1028d361cbf281997c6415dd6ae2e828de47d23b519706e624f580f46
+  metadata.gz: 5d21bcfec5e444aeaf4bf839c13ab40453ea7b27c780e9153aeea23d5692800ff0e36dac604c7b7dec7c7723aea6a1baba675748a543240a69b37a3486665110
+  data.tar.gz: ea0797fc7294cf1be7b641ad874e4560a69a6a7bc04228ee85f7c284f5fcc1288139b6824c72ae5db1c521b19fa3aec55d36c7831dc819046a8af7f2d623886b

data/db/migrate/20210929204903_update_conflicting_singleton_function_to_use_index.rb

@@ -0,0 +1,27 @@
+# frozen_string_literal: true
+
+class UpdateConflictingSingletonFunctionToUseIndex < ActiveRecord::Migration[5.2]
+  def up
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_before_unlock_delete_conflicting_singletons_row_fn')} () RETURNS trigger AS $$
+      BEGIN
+        DELETE FROM delayed_jobs WHERE id<>OLD.id AND singleton=OLD.singleton AND locked_by IS NULL;
+        RETURN NEW;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+
+  def down
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_before_unlock_delete_conflicting_singletons_row_fn')} () RETURNS trigger AS $$
+      BEGIN
+        IF EXISTS (SELECT 1 FROM delayed_jobs j2 WHERE j2.singleton=OLD.singleton) THEN
+          DELETE FROM delayed_jobs WHERE id<>OLD.id AND singleton=OLD.singleton;
+        END IF;
+        RETURN NEW;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+end

data/db/migrate/20211101190934_update_after_delete_trigger_for_singleton_index.rb

@@ -0,0 +1,137 @@
+# frozen_string_literal: true
+
+class UpdateAfterDeleteTriggerForSingletonIndex < ActiveRecord::Migration[5.2]
+  def up
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id AND (j3.locked_by IS NULL OR j3.locked_by IS NOT NULL)))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        ELSIF OLD.singleton IS NOT NULL THEN
+          UPDATE delayed_jobs SET next_in_strand = 't' WHERE singleton=OLD.singleton AND next_in_strand=false AND locked_by IS NULL;
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+
+  def down
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        ELSIF OLD.singleton IS NOT NULL THEN
+          UPDATE delayed_jobs SET next_in_strand = 't' WHERE singleton=OLD.singleton AND next_in_strand=false;
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+end

data/db/migrate/20211207094200_update_after_delete_trigger_for_singleton_transition_cases.rb

@@ -0,0 +1,171 @@
+# frozen_string_literal: true
+
+class UpdateAfterDeleteTriggerForSingletonTransitionCases < ActiveRecord::Migration[5.2]
+  def up
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        next_strand varchar;
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+        transition boolean;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id AND (j3.locked_by IS NULL OR j3.locked_by IS NOT NULL)))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        END IF;
+
+        IF OLD.singleton IS NOT NULL THEN
+          transition := EXISTS (SELECT 1 FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL);
+
+          IF transition THEN
+            next_strand := (SELECT j1.strand FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL AND j1.strand IS NOT NULL LIMIT 1);
+
+            IF next_strand IS NOT NULL THEN
+              -- if the singleton has a new strand defined, we need to lock it to ensure we obey n_strand constraints --
+              IF NOT pg_try_advisory_xact_lock(half_md5_as_bigint(next_strand)) THEN
+                -- a failure to acquire the lock means that another process already has it and will thus handle this singleton --
+                RETURN OLD;
+              END IF;
+            END IF;
+          ELSIF OLD.strand IS NOT NULL THEN
+            -- if there is no transition and there is a strand then we have already handled this singleton in the case above --
+            RETURN OLD;
+          END IF;
+
+          -- handles transitioning a singleton from stranded to not stranded --
+          -- handles transitioning a singleton from unstranded to stranded --
+          -- handles transitioning a singleton from strand A to strand B --
+          -- these transitions are a relatively rare case, so we take a shortcut and --
+          -- only start the next singleton if its strand does not currently have any running jobs --
+          -- if it does, the next stranded job that finishes will start this singleton if it can --
+          UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.singleton=OLD.singleton AND
+                j2.locked_by IS NULL AND
+                (j2.strand IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.strand=j2.strand AND j3.id<>j2.id))
+              FOR UPDATE
+            );
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+
+  def down
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id AND (j3.locked_by IS NULL OR j3.locked_by IS NOT NULL)))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        ELSIF OLD.singleton IS NOT NULL THEN
+          UPDATE delayed_jobs SET next_in_strand = 't' WHERE singleton=OLD.singleton AND next_in_strand=false AND locked_by IS NULL;
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+end

data/db/migrate/20211220112800_fix_singleton_race_condition_insert.rb

@@ -0,0 +1,59 @@
+# frozen_string_literal: true
+
+class FixSingletonRaceConditionInsert < ActiveRecord::Migration[5.2]
+  def change
+    reversible do |direction|
+      direction.up do
+        execute(<<~SQL)
+          CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_before_insert_row_tr_fn')} () RETURNS trigger AS $$
+          BEGIN
+            IF NEW.strand IS NOT NULL THEN
+              PERFORM pg_advisory_xact_lock(half_md5_as_bigint(NEW.strand));
+              IF (SELECT COUNT(*) FROM (
+                  SELECT 1 FROM delayed_jobs WHERE strand = NEW.strand AND next_in_strand=true LIMIT NEW.max_concurrent
+                ) s) = NEW.max_concurrent THEN
+                NEW.next_in_strand := false;
+              END IF;
+            END IF;
+            IF NEW.singleton IS NOT NULL THEN
+              PERFORM pg_advisory_xact_lock(half_md5_as_bigint(CONCAT('singleton:', NEW.singleton)));
+              -- this condition seems silly, but it forces postgres to use the two partial indexes on singleton,
+              -- rather than doing a seq scan
+              PERFORM 1 FROM delayed_jobs WHERE singleton = NEW.singleton AND (locked_by IS NULL OR locked_by IS NOT NULL);
+              IF FOUND THEN
+                NEW.next_in_strand := false;
+              END IF;
+            END IF;
+            RETURN NEW;
+          END;
+          $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+        SQL
+      end
+      direction.down do
+        execute(<<~SQL)
+          CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_before_insert_row_tr_fn')} () RETURNS trigger AS $$
+          BEGIN
+            IF NEW.strand IS NOT NULL THEN
+              PERFORM pg_advisory_xact_lock(half_md5_as_bigint(NEW.strand));
+              IF (SELECT COUNT(*) FROM (
+                  SELECT 1 FROM delayed_jobs WHERE strand = NEW.strand AND next_in_strand=true LIMIT NEW.max_concurrent
+                ) s) = NEW.max_concurrent THEN
+                NEW.next_in_strand := false;
+              END IF;
+            END IF;
+            IF NEW.singleton IS NOT NULL THEN
+              -- this condition seems silly, but it forces postgres to use the two partial indexes on singleton,
+              -- rather than doing a seq scan
+              PERFORM 1 FROM delayed_jobs WHERE singleton = NEW.singleton AND (locked_by IS NULL OR locked_by IS NOT NULL);
+              IF FOUND THEN
+                NEW.next_in_strand := false;
+              END IF;
+            END IF;
+            RETURN NEW;
+          END;
+          $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+        SQL
+      end
+    end
+  end
+end

data/db/migrate/20211220113000_fix_singleton_race_condition_delete.rb

@@ -0,0 +1,207 @@
+# frozen_string_literal: true
+
+class FixSingletonRaceConditionDelete < ActiveRecord::Migration[5.2]
+  def up
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        next_strand varchar;
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+        transition boolean;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id AND (j3.locked_by IS NULL OR j3.locked_by IS NOT NULL)))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        END IF;
+
+        IF OLD.singleton IS NOT NULL THEN
+          PERFORM pg_advisory_xact_lock(half_md5_as_bigint(CONCAT('singleton:', OLD.singleton)));
+
+          transition := EXISTS (SELECT 1 FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL);
+
+          IF transition THEN
+            next_strand := (SELECT j1.strand FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL AND j1.strand IS NOT NULL LIMIT 1);
+
+            IF next_strand IS NOT NULL THEN
+              -- if the singleton has a new strand defined, we need to lock it to ensure we obey n_strand constraints --
+              IF NOT pg_try_advisory_xact_lock(half_md5_as_bigint(next_strand)) THEN
+                -- a failure to acquire the lock means that another process already has it and will thus handle this singleton --
+                RETURN OLD;
+              END IF;
+            END IF;
+          ELSIF OLD.strand IS NOT NULL THEN
+            -- if there is no transition and there is a strand then we have already handled this singleton in the case above --
+            RETURN OLD;
+          END IF;
+
+          -- handles transitioning a singleton from stranded to not stranded --
+          -- handles transitioning a singleton from unstranded to stranded --
+          -- handles transitioning a singleton from strand A to strand B --
+          -- these transitions are a relatively rare case, so we take a shortcut and --
+          -- only start the next singleton if its strand does not currently have any running jobs --
+          -- if it does, the next stranded job that finishes will start this singleton if it can --
+          UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.singleton=OLD.singleton AND
+                j2.locked_by IS NULL AND
+                (j2.strand IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.strand=j2.strand AND j3.id<>j2.id))
+              FOR UPDATE
+            );
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+
+  def down
+    execute(<<~SQL)
+      CREATE OR REPLACE FUNCTION #{connection.quote_table_name('delayed_jobs_after_delete_row_tr_fn')} () RETURNS trigger AS $$
+      DECLARE
+        next_strand varchar;
+        running_count integer;
+        should_lock boolean;
+        should_be_precise boolean;
+        update_query varchar;
+        skip_locked varchar;
+        transition boolean;
+      BEGIN
+        IF OLD.strand IS NOT NULL THEN
+          should_lock := true;
+          should_be_precise := OLD.id % (OLD.max_concurrent * 4) = 0;
+
+          IF NOT should_be_precise AND OLD.max_concurrent > 16 THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 as one FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) subquery_for_count);
+            should_lock := running_count < OLD.max_concurrent;
+          END IF;
+
+          IF should_lock THEN
+            PERFORM pg_advisory_xact_lock(half_md5_as_bigint(OLD.strand));
+          END IF;
+
+          -- note that we don't really care if the row we're deleting has a singleton, or if it even
+          -- matches the row(s) we're going to update. we just need to make sure that whatever
+          -- singleton we grab isn't already running (which is a simple existence check, since
+          -- the unique indexes ensure there is at most one singleton running, and one queued)
+          update_query := 'UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.strand=$1.strand AND
+                (j2.singleton IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.singleton=j2.singleton AND j3.id<>j2.id AND (j3.locked_by IS NULL OR j3.locked_by IS NOT NULL)))
+              ORDER BY j2.strand_order_override ASC, j2.id ASC
+              LIMIT ';
+
+          IF should_be_precise THEN
+            running_count := (SELECT COUNT(*) FROM (
+              SELECT 1 FROM delayed_jobs WHERE strand = OLD.strand AND next_in_strand = 't' LIMIT OLD.max_concurrent
+            ) s);
+            IF running_count < OLD.max_concurrent THEN
+              update_query := update_query || '($1.max_concurrent - $2)';
+            ELSE
+              -- we have too many running already; just bail
+              RETURN OLD;
+            END IF;
+          ELSE
+            update_query := update_query || '1';
+
+            -- n-strands don't require precise ordering; we can make this query more performant
+            IF OLD.max_concurrent > 1 THEN
+              skip_locked := ' SKIP LOCKED';
+            END IF;
+          END IF;
+
+          update_query := update_query || ' FOR UPDATE' || COALESCE(skip_locked, '') || ')';
+          EXECUTE update_query USING OLD, running_count;
+        END IF;
+
+        IF OLD.singleton IS NOT NULL THEN
+          transition := EXISTS (SELECT 1 FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL);
+
+          IF transition THEN
+            next_strand := (SELECT j1.strand FROM delayed_jobs AS j1 WHERE j1.singleton = OLD.singleton AND j1.strand IS DISTINCT FROM OLD.strand AND locked_by IS NULL AND j1.strand IS NOT NULL LIMIT 1);
+
+            IF next_strand IS NOT NULL THEN
+              -- if the singleton has a new strand defined, we need to lock it to ensure we obey n_strand constraints --
+              IF NOT pg_try_advisory_xact_lock(half_md5_as_bigint(next_strand)) THEN
+                -- a failure to acquire the lock means that another process already has it and will thus handle this singleton --
+                RETURN OLD;
+              END IF;
+            END IF;
+          ELSIF OLD.strand IS NOT NULL THEN
+            -- if there is no transition and there is a strand then we have already handled this singleton in the case above --
+            RETURN OLD;
+          END IF;
+
+          -- handles transitioning a singleton from stranded to not stranded --
+          -- handles transitioning a singleton from unstranded to stranded --
+          -- handles transitioning a singleton from strand A to strand B --
+          -- these transitions are a relatively rare case, so we take a shortcut and --
+          -- only start the next singleton if its strand does not currently have any running jobs --
+          -- if it does, the next stranded job that finishes will start this singleton if it can --
+          UPDATE delayed_jobs SET next_in_strand=true WHERE id IN (
+            SELECT id FROM delayed_jobs j2
+              WHERE next_in_strand=false AND
+                j2.singleton=OLD.singleton AND
+                j2.locked_by IS NULL AND
+                (j2.strand IS NULL OR NOT EXISTS (SELECT 1 FROM delayed_jobs j3 WHERE j3.strand=j2.strand AND j3.id<>j2.id))
+              FOR UPDATE
+            );
+        END IF;
+        RETURN OLD;
+      END;
+      $$ LANGUAGE plpgsql SET search_path TO #{::Switchman::Shard.current.name};
+    SQL
+  end
+end

data/lib/switchman_inst_jobs/jobs_migrator.rb

@@ -277,7 +277,10 @@ module SwitchmanInstJobs
         connection = ::Delayed::Job.connection
         quoted_keys = keys.map { |k| connection.quote_column_name(k) }.join(', ')
 
-        connection.execute "COPY #{::Delayed::Job.quoted_table_name} (#{quoted_keys}) FROM STDIN"
+        connection.execute 'DROP TABLE IF EXISTS delayed_jobs_bulk_copy'
+        connection.execute "CREATE TEMPORARY TABLE delayed_jobs_bulk_copy
+          (LIKE #{::Delayed::Job.quoted_table_name} INCLUDING DEFAULTS)"
+        connection.execute "COPY delayed_jobs_bulk_copy (#{quoted_keys}) FROM STDIN"
         records.map do |record|
           connection.raw_connection.put_copy_data("#{keys.map { |k| quote_text(record[k]) }.join("\t")}\n")
         end
@@ -289,6 +292,9 @@ module SwitchmanInstJobs
         rescue StandardError => e
           raise connection.send(:translate_exception, e, 'COPY FROM STDIN')
         end
+        connection.execute "INSERT INTO #{::Delayed::Job.quoted_table_name} (#{quoted_keys})
+          SELECT #{quoted_keys} FROM delayed_jobs_bulk_copy
+          ON CONFLICT (singleton) WHERE singleton IS NOT NULL AND locked_by IS NULL DO NOTHING"
         result.cmd_tuples
       end
 

data/lib/switchman_inst_jobs/version.rb

@@ -1,3 +1,3 @@
 module SwitchmanInstJobs
-  VERSION = '3.2.2'.freeze
+  VERSION = '3.2.7'.freeze
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: switchman-inst-jobs
 version: !ruby/object:Gem::Version
-  version: 3.2.2
+  version: 3.2.7
 platform: ruby
 authors:
 - Bryan Petty
-autorequire:
+autorequire: 
 bindir: bin
 cert_chain: []
-date: 2021-09-28 00:00:00.000000000 Z
+date: 2021-12-20 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: inst-jobs
@@ -19,7 +19,7 @@ dependencies:
         version: 2.4.9
     - - "<"
       - !ruby/object:Gem::Version
-        version: '3.0'
+        version: '4.0'
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
@@ -29,7 +29,7 @@ dependencies:
         version: 2.4.9
     - - "<"
       - !ruby/object:Gem::Version
-        version: '3.0'
+        version: '4.0'
 - !ruby/object:Gem::Dependency
   name: parallel
   requirement: !ruby/object:Gem::Requirement
@@ -260,7 +260,7 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: '1.4'
-description:
+description: 
 email:
 - bpetty@instructure.com
 executables: []
@@ -300,6 +300,11 @@ files:
 - db/migrate/20210812210128_add_singleton_column.rb
 - db/migrate/20210917232626_add_delete_conflicting_singletons_before_unlock_trigger.rb
 - db/migrate/20210928174754_fix_singleton_condition_in_before_insert.rb
+- db/migrate/20210929204903_update_conflicting_singleton_function_to_use_index.rb
+- db/migrate/20211101190934_update_after_delete_trigger_for_singleton_index.rb
+- db/migrate/20211207094200_update_after_delete_trigger_for_singleton_transition_cases.rb
+- db/migrate/20211220112800_fix_singleton_race_condition_insert.rb
+- db/migrate/20211220113000_fix_singleton_race_condition_delete.rb
 - lib/switchman-inst-jobs.rb
 - lib/switchman_inst_jobs.rb
 - lib/switchman_inst_jobs/active_record/connection_adapters/postgresql_adapter.rb
@@ -323,8 +328,10 @@ files:
 homepage: https://github.com/instructure/switchman-inst-jobs
 licenses:
 - MIT
-metadata: {}
-post_install_message:
+metadata:
+  allowed_push_host: https://rubygems.org
+  rubygems_mfa_required: 'true'
+post_install_message: 
 rdoc_options: []
 require_paths:
 - lib
@@ -339,8 +346,8 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubygems_version: 3.2.24
-signing_key:
+rubygems_version: 3.1.4
+signing_key: 
 specification_version: 4
 summary: Switchman and Instructure Jobs compatibility gem.
 test_files: []