14484: Fixes bug in pdh transaction grouping

[arvados.git] / services / api / app / models / container.rb

diff --git a/services/api/app/models/container.rb b/services/api/app/models/container.rb
index 0682676c5ccc2d7f7f738c2f99971447f1456be9..e46ef6fd378ce225c2fb8d8d75f66735b37ee3bf 100644 (file)
--- a/services/api/app/models/container.rb
+++ b/services/api/app/models/container.rb
@@ -346,7 +346,7 @@ class Container < ArvadosModel
     transaction do
       reload
       check_lock_fail
-      update_attributes!(state: Locked)
+      update_attributes!(state: Locked, lock_count: self.lock_count+1)
     end
   end
 
@@ -364,7 +364,14 @@ class Container < ArvadosModel
     transaction do
       reload(lock: 'FOR UPDATE')
       check_unlock_fail
-      update_attributes!(state: Queued)
+      if self.lock_count < Rails.configuration.max_container_dispatch_attempts
+        update_attributes!(state: Queued)
+      else
+        update_attributes!(state: Cancelled,
+                           runtime_status: {
+                             error: "Container exceeded 'max_container_dispatch_attempts' (lock_count=#{self.lock_count}."
+                           })
+      end
     end
   end
 
@@ -398,6 +405,65 @@ class Container < ArvadosModel
     end
   end
 
+  # NOTE: Migration 20190322174136_add_file_info_to_collection.rb relies on this function.
+  #
+  # Change with caution!
+  #
+  # Correctly groups pdhs to use for batch database updates. Helps avoid
+  # updating too many database rows in a single transaction.
+  def self.group_pdhs_for_multiple_transactions(log_prefix)
+    batch_size_max = 1 << 28 # 256 MiB
+    last_pdh = '0'
+    done = 0
+    any = true
+
+    total = ActiveRecord::Base.connection.exec_query(
+      'SELECT DISTINCT portable_data_hash FROM collections'
+    ).rows.count
+
+    while any
+      any = false
+      pdhs = ActiveRecord::Base.connection.exec_query(
+        'SELECT DISTINCT portable_data_hash FROM collections '\
+        "WHERE portable_data_hash > '#{last_pdh}' "\
+        'GROUP BY portable_data_hash LIMIT 1000'
+      )
+      break if pdhs.rows.count.zero?
+
+      Container.group_pdhs_by_manifest_size(pdhs.rows, batch_size_max) do |grouped_pdhs|
+        any = true
+        yield grouped_pdhs
+        done += grouped_pdhs.size
+        last_pdh = pdhs[-1]
+        Rails.logger.info(log_prefix + ": #{done}/#{total}")
+      end
+    end
+    Rails.logger.info(log_prefix + ': finished')
+  end
+
+  # NOTE: Migration 20190322174136_add_file_info_to_collection.rb relies on this function.
+  #
+  # Change with caution!
+  #
+  # Given an array of pdhs, yield a subset array of pdhs when the total
+  # size of all manifest_texts is no more than batch_size_max. Pdhs whose manifest_text 
+  # is bigger than batch_size_max are yielded by themselves
+  def self.group_pdhs_by_manifest_size(pdhs, batch_size_max)
+    batch_size = 0
+    batch_pdhs = {}
+    pdhs.each do |pdh|
+      manifest_size = pdh.split('+')[1].to_i
+      if batch_size > 0 && batch_size + manifest_size > batch_size_max
+        yield batch_pdhs.keys
+        batch_pdhs = {}
+        batch_size = 0
+      end
+      batch_pdhs[pdh] = true
+      batch_size += manifest_size
+    end
+    yield batch_pdhs.keys
+  end
+
   protected
 
   def fill_field_defaults
@@ -457,7 +523,7 @@ class Container < ArvadosModel
 
     case self.state
     when Locked
-      permitted.push :priority, :runtime_status, :log
+      permitted.push :priority, :runtime_status, :log, :lock_count
 
     when Queued
       permitted.push :priority
@@ -478,7 +544,7 @@ class Container < ArvadosModel
       when Running
         permitted.push :finished_at, *progress_attrs
       when Queued, Locked
-        permitted.push :finished_at, :log
+        permitted.push :finished_at, :log, :runtime_status
       end
 
     else