14484: Adds functionality and test for pdh grouping in the container model
[arvados.git] / services / api / app / models / container.rb
index cd763a8e7e1eb0d851f08517b730ddf9f230a113..b3328d9c76942949f8153449ec1368e4eb8418ba 100644 (file)
@@ -223,7 +223,11 @@ class Container < ArvadosModel
       if mount['kind'] != 'collection'
         next
       end
-      if (uuid = mount.delete 'uuid')
+
+      uuid = mount.delete 'uuid'
+
+      if mount['portable_data_hash'].nil? and !uuid.nil?
+        # PDH not supplied, try by UUID
         c = Collection.
           readable_by(current_user).
           where(uuid: uuid).
@@ -232,13 +236,7 @@ class Container < ArvadosModel
         if !c
           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
         end
-        if mount['portable_data_hash'].nil?
-          # PDH not supplied by client
-          mount['portable_data_hash'] = c.portable_data_hash
-        elsif mount['portable_data_hash'] != c.portable_data_hash
-          # UUID and PDH supplied by client, but they don't agree
-          raise ArgumentError.new "cannot mount collection #{uuid.inspect}: current portable_data_hash #{c.portable_data_hash.inspect} does not match #{c['portable_data_hash'].inspect} in request"
-        end
+        mount['portable_data_hash'] = c.portable_data_hash
       end
     end
     return c_mounts
@@ -281,14 +279,6 @@ class Container < ArvadosModel
     candidates = candidates.where_serialized(:runtime_constraints, resolve_runtime_constraints(attrs[:runtime_constraints]), md5: true)
     log_reuse_info(candidates) { "after filtering on runtime_constraints #{attrs[:runtime_constraints].inspect}" }
 
-    candidates = candidates.where('runtime_user_uuid = ? or (runtime_user_uuid is NULL and runtime_auth_scopes is NULL)',
-                                  attrs[:runtime_user_uuid])
-    log_reuse_info(candidates) { "after filtering on runtime_user_uuid #{attrs[:runtime_user_uuid].inspect}" }
-
-    candidates = candidates.where('runtime_auth_scopes = ? or (runtime_user_uuid is NULL and runtime_auth_scopes is NULL)',
-                                  SafeJSON.dump(attrs[:runtime_auth_scopes].sort))
-    log_reuse_info(candidates) { "after filtering on runtime_auth_scopes #{attrs[:runtime_auth_scopes].inspect}" }
-
     log_reuse_info { "checking for state=Complete with readable output and log..." }
 
     select_readable_pdh = Collection.
@@ -356,7 +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
 
@@ -374,7 +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
 
@@ -385,6 +382,9 @@ class Container < ArvadosModel
     else
       kwargs = {}
     end
+    if users_list.select { |u| u.is_admin }.any?
+      return super
+    end
     Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").exists)
   end
 
@@ -405,6 +405,67 @@ 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'
+      )
+      if pdhs.rows.count.zero?
+        break
+      end
+
+      Container.group_pdhs_by_manifest_size(pdhs, 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
@@ -464,7 +525,7 @@ 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
@@ -485,7 +546,7 @@ 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