switchman-inst-jobs 3.2.0 → 3.2.10

data/lib/switchman_inst_jobs/jobs_migrator.rb

@@ -89,7 +89,9 @@ module SwitchmanInstJobs
         migrate_everything
       end
 
-      def migrate_strands
+      def migrate_strands(batch_size: 1_000)
+        source_shard = ::Switchman::Shard.current(:delayed_jobs)
+
         # there are 4 scenarios to deal with here
         # 1) no running job, no jobs moved: do nothing
         # 2) running job, no jobs moved; create blocker with next_in_strand=false
@@ -98,60 +100,89 @@ module SwitchmanInstJobs
         #    those (= do nothing since it should already be false)
         # 4) no running job, jobs moved: set next_in_strand=true on the first of
         #    those (= do nothing since it should already be true)
+        handler = lambda { |scope, column, blocker_job_kwargs = {}, advisory_lock_cb = nil|
+          shard_map = build_shard_map(scope, source_shard)
+          shard_map.each do |(target_shard, source_shard_ids)|
+            shard_scope = scope.where(shard_id: source_shard_ids)
 
-        source_shard = ::Switchman::Shard.current(:delayed_jobs)
-        strand_scope = ::Delayed::Job.shard(source_shard).where.not(strand: nil)
-        shard_map = build_shard_map(strand_scope, source_shard)
-        shard_map.each do |(target_shard, source_shard_ids)|
-          shard_scope = strand_scope.where(shard_id: source_shard_ids)
-
-          # 1) is taken care of because it should not show up here in strands
-          strands = shard_scope.distinct.order(:strand).pluck(:strand)
-
-          target_shard.activate(:delayed_jobs) do
-            strands.each do |strand|
-              transaction_on([source_shard, target_shard]) do
-                this_strand_scope = shard_scope.where(strand: strand)
-                # we want to copy all the jobs except the one that is still running.
-                jobs_scope = this_strand_scope.where(locked_by: nil)
-
-                # 2) and part of 3) are taken care of here by creating a blocker
-                # job with next_in_strand = false. as soon as the current
-                # running job is finished it should set next_in_strand
-                # We lock it to ensure that the jobs worker can't delete it until we are done moving the strand
-                # Since we only unlock it on the new jobs queue *after* deleting from the original
-                # the lock ensures the blocker always gets unlocked
-                first = this_strand_scope.where.not(locked_by: nil).next_in_strand_order.lock.first
-                if first
-                  first_job = ::Delayed::Job.create!(strand: strand, next_in_strand: false)
-                  first_job.payload_object = ::Delayed::PerformableMethod.new(Kernel, :sleep, args: [0])
-                  first_job.queue = first.queue
-                  first_job.tag = 'Kernel.sleep'
-                  first_job.source = 'JobsMigrator::StrandBlocker'
-                  first_job.max_attempts = 1
-                  # If we ever have jobs left over from 9999 jobs moves of a single shard,
-                  # something has gone terribly wrong
-                  first_job.strand_order_override = -9999
-                  first_job.save!
-                  # the rest of 3) is taken care of here
-                  # make sure that all the jobs moved over are NOT next in strand
-                  ::Delayed::Job.where(next_in_strand: true, strand: strand, locked_by: nil).
-                    update_all(next_in_strand: false)
-                end
+            # 1) is taken care of because it should not show up here in strands
+            values = shard_scope.distinct.order(column).pluck(column)
 
-                # 4) is taken care of here, by leaving next_in_strand alone and
-                # it should execute on the new shard
-                batch_move_jobs(
-                  target_shard: target_shard,
-                  source_shard: source_shard,
-                  scope: jobs_scope
-                ) do |job, new_job|
-                  # This ensures jobs enqueued on the old jobs shard run before jobs on the new jobs queue
-                  new_job.strand_order_override = job.strand_order_override - 1
+            target_shard.activate(:delayed_jobs) do
+              values.each do |value|
+                transaction_on([source_shard, target_shard]) do
+                  source_shard.activate(::Delayed::Backend::ActiveRecord::AbstractJob) do
+                    advisory_lock_cb&.call(value)
+                  end
+
+                  value_scope = shard_scope.where(**{ column => value })
+                  # we want to copy all the jobs except the one that is still running.
+                  jobs_scope = value_scope.where(locked_by: nil)
+
+                  # 2) and part of 3) are taken care of here by creating a blocker
+                  # job with next_in_strand = false. as soon as the current
+                  # running job is finished it should set next_in_strand
+                  # We lock it to ensure that the jobs worker can't delete it until we are done moving the strand
+                  # Since we only unlock it on the new jobs queue *after* deleting from the original
+                  # the lock ensures the blocker always gets unlocked
+                  first = value_scope.where.not(locked_by: nil).next_in_strand_order.lock.first
+                  if first
+                    create_blocker_job(
+                      queue: first.queue,
+                      shard_id: first.shard_id,
+                      **{ column => value },
+                      **blocker_job_kwargs
+                    )
+
+                    # the rest of 3) is taken care of here
+                    # make sure that all the jobs moved over are NOT next in strand
+                    ::Delayed::Job.where(next_in_strand: true, locked_by: nil, **{ column => value }).
+                      update_all(next_in_strand: false)
+                  end
+
+                  # 4) is taken care of here, by leaving next_in_strand alone and
+                  # it should execute on the new shard
+                  batch_move_jobs(
+                    target_shard: target_shard,
+                    source_shard: source_shard,
+                    scope: jobs_scope,
+                    batch_size: batch_size
+                  ) do |job, new_job|
+                    # This ensures jobs enqueued on the old jobs shard run before jobs on the new jobs queue
+                    new_job.strand_order_override = job.strand_order_override - 1
+                  end
                 end
               end
             end
+          end
+        }
+
+        strand_scope = ::Delayed::Job.shard(source_shard).where.not(strand: nil)
+        singleton_scope = ::Delayed::Job.shard(source_shard).where('strand IS NULL AND singleton IS NOT NULL')
+        all_scope = ::Delayed::Job.shard(source_shard).where('strand IS NOT NULL OR singleton IS NOT NULL')
 
+        singleton_blocker_additional_kwargs = {
+          locked_at: DateTime.now,
+          locked_by: ::Delayed::Backend::Base::ON_HOLD_BLOCKER
+        }
+
+        quoted_function_name = ::Delayed::Job.connection.quote_table_name('half_md5_as_bigint')
+        strand_advisory_lock_fn = lambda do |value|
+          ::Delayed::Job.connection.execute("SELECT pg_advisory_xact_lock(#{quoted_function_name}('#{value}'))")
+        end
+
+        singleton_advisory_lock_fn = lambda do |value|
+          ::Delayed::Job.connection.execute(
+            "SELECT pg_advisory_xact_lock(#{quoted_function_name}('singleton:#{value}'))"
+          )
+        end
+
+        handler.call(strand_scope, :strand, {}, strand_advisory_lock_fn)
+        handler.call(singleton_scope, :singleton, singleton_blocker_additional_kwargs, singleton_advisory_lock_fn)
+
+        shard_map = build_shard_map(all_scope, source_shard)
+        shard_map.each do |(target_shard, source_shard_ids)|
+          target_shard.activate(:delayed_jobs) do
             updated = ::Switchman::Shard.where(id: source_shard_ids, block_stranded: true).
               update_all(block_stranded: false)
             # If this is being manually re-run for some reason to clean something up, don't wait for nothing to happen
@@ -166,26 +197,40 @@ module SwitchmanInstJobs
         end
       end
 
-      def unblock_strands(target_shard)
-        target_shard.activate(:delayed_jobs) do
-          loop do
-            # We only want to unlock stranded jobs where they don't belong to a blocked shard (if they *do* belong)
-            # to a blocked shard, they must be part of a concurrent jobs migration from a different source shard to
-            # this target shard, so we shouldn't unlock them yet.  We only ever unlock one job here to keep the
-            # logic cleaner; if the job is n-stranded, after the first one runs, the trigger will unlock larger
-            # batches
-            break if ::Delayed::Job.where(id: ::Delayed::Job.select('DISTINCT ON (strand) id').
-              where.not(strand: nil).
-              where.not(shard_id: ::Switchman::Shard.where(block_stranded: true).pluck(:id)).where(
+      def unblock_strands(target_shard, batch_size: 10_000)
+        block_stranded_ids = ::Switchman::Shard.where(block_stranded: true).pluck(:id)
+        query = lambda { |column, scope|
+          ::Delayed::Job.
+            where(id: ::Delayed::Job.select("DISTINCT ON (#{column}) id").
+              where(scope).
+              where.not(shard_id: block_stranded_ids).
+              where(
                 ::Delayed::Job.select(1).from("#{::Delayed::Job.quoted_table_name} dj2").
                 where("dj2.next_in_strand = true OR dj2.source = 'JobsMigrator::StrandBlocker'").
-                where('dj2.strand = delayed_jobs.strand').arel.exists.not
-              ).order(:strand, :strand_order_override, :id)).limit(500).update_all(next_in_strand: true).zero?
+                where("dj2.#{column} = delayed_jobs.#{column}").arel.exists.not
+              ).
+              order(column, :strand_order_override, :id)).limit(batch_size)
+        }
+
+        target_shard.activate(:delayed_jobs) do
+          # We only want to unlock stranded jobs where they don't belong to a blocked shard (if they *do* belong)
+          # to a blocked shard, they must be part of a concurrent jobs migration from a different source shard to
+          # this target shard, so we shouldn't unlock them yet.  We only ever unlock one job here to keep the
+          # logic cleaner; if the job is n-stranded, after the first one runs, the trigger will unlock larger
+          # batches
+
+          loop do
+            break if query.call(:strand, 'strand IS NOT NULL').update_all(next_in_strand: true).zero?
+          end
+
+          loop do
+            break if query.call(:singleton,
+                                'strand IS NULL AND singleton IS NOT NULL').update_all(next_in_strand: true).zero?
           end
         end
       end
 
-      def migrate_everything
+      def migrate_everything(batch_size: 1_000)
         source_shard = ::Switchman::Shard.current(:delayed_jobs)
         scope = ::Delayed::Job.shard(source_shard).where('strand IS NULL')
 
@@ -194,13 +239,26 @@ module SwitchmanInstJobs
           batch_move_jobs(
             target_shard: target_shard,
             source_shard: source_shard,
-            scope: scope.where(shard_id: source_shard_ids).where(locked_by: nil)
+            scope: scope.where(shard_id: source_shard_ids).where(locked_by: nil),
+            batch_size: batch_size
           )
         end
       end
 
       private
 
+      def create_blocker_job(**kwargs)
+        first_job = ::Delayed::Job.create!(**kwargs, next_in_strand: false)
+        first_job.payload_object = ::Delayed::PerformableMethod.new(Kernel, :sleep, args: [0])
+        first_job.tag = 'Kernel.sleep'
+        first_job.source = 'JobsMigrator::StrandBlocker'
+        first_job.max_attempts = 1
+        # If we ever have jobs left over from 9999 jobs moves of a single shard,
+        # something has gone terribly wrong
+        first_job.strand_order_override = -9999
+        first_job.save!
+      end
+
       def build_shard_map(scope, source_shard)
         shard_ids = scope.distinct.pluck(:shard_id)
 
@@ -215,10 +273,10 @@ module SwitchmanInstJobs
         shard_map
       end
 
-      def batch_move_jobs(target_shard:, source_shard:, scope:)
+      def batch_move_jobs(target_shard:, source_shard:, scope:, batch_size:)
         while scope.exists?
           # Adapted from get_and_lock_next_available in delayed/backend/active_record.rb
-          target_jobs = scope.limit(1000).lock('FOR UPDATE SKIP LOCKED')
+          target_jobs = scope.limit(batch_size).lock('FOR UPDATE SKIP LOCKED')
 
           query = source_shard.activate(:delayed_jobs) do
             "WITH limited_jobs AS (#{target_jobs.to_sql}) " \
@@ -277,7 +335,10 @@ 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 +350,9 @@ 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 = '3.2.0'.freeze
+  VERSION = '3.2.10'.freeze
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: switchman-inst-jobs
 version: !ruby/object:Gem::Version
-  version: 3.2.0
+  version: 3.2.10
 platform: ruby
 authors:
 - Bryan Petty
-autorequire:
+autorequire: 
 bindir: bin
 cert_chain: []
-date: 2021-09-09 00:00:00.000000000 Z
+date: 2022-02-23 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: inst-jobs
@@ -16,20 +16,20 @@ dependencies:
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 2.4.0
+        version: 2.4.9
     - - "<"
       - !ruby/object:Gem::Version
-        version: '3.0'
+        version: '4.0'
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 2.4.0
+        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: []
@@ -298,6 +298,17 @@ files:
 - db/migrate/20200825011002_add_strand_order_override.rb
 - db/migrate/20210809145804_add_n_strand_index.rb
 - 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
+- db/migrate/20220127091200_fix_singleton_unique_constraint.rb
+- db/migrate/20220128084800_update_insert_trigger_for_singleton_unique_constraint_change.rb
+- db/migrate/20220128084900_update_delete_trigger_for_singleton_unique_constraint_change.rb
+- db/migrate/20220203063200_remove_old_singleton_index.rb
 - lib/switchman-inst-jobs.rb
 - lib/switchman_inst_jobs.rb
 - lib/switchman_inst_jobs/active_record/connection_adapters/postgresql_adapter.rb
@@ -321,8 +332,10 @@ 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
@@ -337,8 +350,8 @@ 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: []