Merge branch '21600-banner-tests'
[arvados.git] / services / api / app / models / collection.rb
index ddb85862a26c6af6751680a8358147ce4fe841f6..16e85c0dd9af575d93fc1aeda8279e798b60e56c 100644 (file)
@@ -3,7 +3,6 @@
 # SPDX-License-Identifier: AGPL-3.0
 
 require 'arvados/keep'
-require 'sweep_trashed_objects'
 require 'trashable'
 
 class Collection < ArvadosModel
@@ -14,28 +13,38 @@ class Collection < ArvadosModel
   include CommonApiTemplate
   include Trashable
 
-  serialize :properties, Hash
-  serialize :storage_classes_desired, Array
-  serialize :storage_classes_confirmed, Array
+  # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
+  # already know how to properly treat them.
+  attribute :properties, :jsonbHash, default: {}
+  attribute :storage_classes_desired, :jsonbArray, default: lambda { Rails.configuration.DefaultStorageClasses }
+  attribute :storage_classes_confirmed, :jsonbArray, default: []
 
   before_validation :default_empty_manifest
   before_validation :default_storage_classes, on: :create
+  before_validation :managed_properties, on: :create
   before_validation :check_encoding
   before_validation :check_manifest_validity
   before_validation :check_signatures
   before_validation :strip_signatures_and_update_replication_confirmed
+  before_validation :name_null_if_empty
+  validate :ensure_filesystem_compatible_name
   validate :ensure_pdh_matches_manifest_text
   validate :ensure_storage_classes_desired_is_not_empty
   validate :ensure_storage_classes_contain_non_empty_strings
+  validate :versioning_metadata_updates, on: :update
+  validate :past_versions_cannot_be_updated, on: :update
+  validate :protected_managed_properties_updates, on: :update
+  after_validation :set_file_count_and_total_size
   before_save :set_file_names
+  around_update :manage_versioning, unless: :is_past_version?
 
   api_accessible :user, extend: :common do |t|
-    t.add :name
+    t.add lambda { |x| x.name || "" }, as: :name
     t.add :description
     t.add :properties
     t.add :portable_data_hash
-    t.add :signed_manifest_text, as: :manifest_text
     t.add :manifest_text, as: :unsigned_manifest_text
+    t.add :manifest_text, as: :manifest_text
     t.add :replication_desired
     t.add :replication_confirmed
     t.add :replication_confirmed_at
@@ -45,8 +54,15 @@ class Collection < ArvadosModel
     t.add :delete_at
     t.add :trash_at
     t.add :is_trashed
+    t.add :version
+    t.add :current_version_uuid
+    t.add :preserve_version
+    t.add :file_count
+    t.add :file_size_total
   end
 
+  UNLOGGED_CHANGES = ['preserve_version', 'updated_at']
+
   after_initialize do
     @signatures_checked = false
     @computed_pdh_for_manifest_text = false
@@ -54,16 +70,13 @@ class Collection < ArvadosModel
 
   def self.attributes_required_columns
     super.merge(
-                # If we don't list manifest_text explicitly, the
-                # params[:select] code gets confused by the way we
-                # expose signed_manifest_text as manifest_text in the
-                # API response, and never let clients select the
-                # manifest_text column.
-                #
-                # We need trash_at and is_trashed to determine the
-                # correct timestamp in signed_manifest_text.
-                'manifest_text' => ['manifest_text', 'trash_at', 'is_trashed'],
+                # If we don't list unsigned_manifest_text explicitly,
+                # the params[:select] code gets confused by the way we
+                # expose manifest_text as unsigned_manifest_text in
+                # the API response, and never let clients select the
+                # unsigned_manifest_text column.
                 'unsigned_manifest_text' => ['manifest_text'],
+                'name' => ['name'],
                 )
   end
 
@@ -77,7 +90,7 @@ class Collection < ArvadosModel
 
   FILE_TOKEN = /^[[:digit:]]+:[[:digit:]]+:/
   def check_signatures
-    return false if self.manifest_text.nil?
+    throw(:abort) if self.manifest_text.nil?
 
     return true if current_user.andand.is_admin
 
@@ -114,7 +127,7 @@ class Collection < ArvadosModel
             # Signature provided, but verify_signature did not like it.
             logger.warn "Invalid signature on locator #{tok}"
             raise ArvadosModel::PermissionDeniedError
-          elsif Rails.configuration.permit_create_collection_with_unsigned_manifest
+          elsif !Rails.configuration.Collections.BlobSigning
             # No signature provided, but we are running in insecure mode.
             logger.debug "Missing signature on locator #{tok} ignored"
           elsif Blob.new(tok).empty?
@@ -132,7 +145,9 @@ class Collection < ArvadosModel
   def strip_signatures_and_update_replication_confirmed
     if self.manifest_text_changed?
       in_old_manifest = {}
-      if not self.replication_confirmed.nil?
+      # manifest_text_was could be nil when dealing with a freshly created snapshot,
+      # so we skip this case because there was no real manifest change. (Bug #18005)
+      if (not self.replication_confirmed.nil?) and (not self.manifest_text_was.nil?)
         self.class.each_manifest_locator(manifest_text_was) do |match|
           in_old_manifest[match[1]] = true
         end
@@ -182,6 +197,12 @@ class Collection < ArvadosModel
     end
   end
 
+  def name_null_if_empty
+    if name == ""
+      self.name = nil
+    end
+  end
+
   def set_file_names
     if self.manifest_text_changed?
       self.file_names = manifest_files
@@ -189,14 +210,33 @@ class Collection < ArvadosModel
     true
   end
 
+  def set_file_count_and_total_size
+    # Only update the file stats if the manifest changed
+    if self.manifest_text_changed?
+      m = Keep::Manifest.new(self.manifest_text)
+      self.file_size_total = m.files_size
+      self.file_count = m.files_count
+    # If the manifest didn't change but the attributes did, ignore the changes
+    elsif self.file_count_changed? || self.file_size_total_changed?
+      self.file_count = self.file_count_was
+      self.file_size_total = self.file_size_total_was
+    end
+    true
+  end
+
   def manifest_files
     return '' if !self.manifest_text
 
+    done = {}
     names = ''
     self.manifest_text.scan(/ \d+:\d+:(\S+)/) do |name|
+      next if done[name]
+      done[name] = true
       names << name.first.gsub('\040',' ') + "\n"
     end
     self.manifest_text.scan(/^\.\/(\S+)/m) do |stream_name|
+      next if done[stream_name]
+      done[stream_name] = true
       names << stream_name.first.gsub('\040',' ') + "\n"
     end
     names
@@ -206,10 +246,123 @@ class Collection < ArvadosModel
     self.manifest_text ||= ''
   end
 
-  def check_encoding
-    if manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?
-      true
+  def skip_uuid_existence_check
+    # Avoid checking the existence of current_version_uuid, as it's
+    # assigned on creation of a new 'current version' collection, so
+    # the collection's UUID only lives on memory when the validation check
+    # is performed.
+    ['current_version_uuid']
+  end
+
+  def manage_versioning
+    should_preserve_version = should_preserve_version? # Time sensitive, cache value
+    return(yield) unless (should_preserve_version || syncable_updates.any?)
+
+    # Put aside the changes because with_lock does a record reload
+    changes = self.changes
+    snapshot = nil
+    restore_attributes
+    with_lock do
+      # Copy the original state to save it as old version
+      if should_preserve_version
+        snapshot = self.dup
+        snapshot.uuid = nil # Reset UUID so it's created as a new record
+        snapshot.created_at = self.created_at
+        snapshot.modified_at = self.modified_at_was
+      end
+
+      # Restore requested changes on the current version
+      changes.keys.each do |attr|
+        if attr == 'preserve_version' && changes[attr].last == false && !should_preserve_version
+          next # Ignore false assignment, once true it'll be true until next version
+        end
+        self.attributes = {attr => changes[attr].last}
+        if attr == 'uuid'
+          # Also update the current version reference
+          self.attributes = {'current_version_uuid' => changes[attr].last}
+        end
+      end
+
+      if should_preserve_version
+        self.version += 1
+      end
+
+      yield
+
+      sync_past_versions if syncable_updates.any?
+      if snapshot
+        snapshot.attributes = self.syncable_updates
+        leave_modified_by_user_alone do
+          leave_modified_at_alone do
+            act_as_system_user do
+              snapshot.save
+            end
+          end
+        end
+      end
+    end
+  end
+
+  def maybe_update_modified_by_fields
+    if !(self.changes.keys - ['updated_at', 'preserve_version']).empty?
+      super
+    end
+  end
+
+  def syncable_updates
+    updates = {}
+    if self.changes.any?
+      changes = self.changes
     else
+      # If called after save...
+      changes = self.saved_changes
+    end
+    (syncable_attrs & changes.keys).each do |attr|
+      if attr == 'uuid'
+        # Point old versions to current version's new UUID
+        updates['current_version_uuid'] = changes[attr].last
+      else
+        updates[attr] = changes[attr].last
+      end
+    end
+    return updates
+  end
+
+  def sync_past_versions
+    Collection.where('current_version_uuid = ? AND uuid != ?', self.uuid_before_last_save, self.uuid_before_last_save).update_all self.syncable_updates
+  end
+
+  def versionable_updates?(attrs)
+    (['manifest_text', 'description', 'properties', 'name'] & attrs).any?
+  end
+
+  def syncable_attrs
+    ['uuid', 'owner_uuid', 'delete_at', 'trash_at', 'is_trashed', 'replication_desired', 'storage_classes_desired']
+  end
+
+  def is_past_version?
+    # Check for the '_was' values just in case the update operation
+    # includes a change on current_version_uuid or uuid.
+    !(new_record? || self.current_version_uuid_was == self.uuid_was)
+  end
+
+  def should_preserve_version?
+    return false unless (Rails.configuration.Collections.CollectionVersioning && versionable_updates?(self.changes.keys))
+
+    return false if self.is_trashed
+
+    idle_threshold = Rails.configuration.Collections.PreserveVersionIfIdle
+    if !self.preserve_version_was &&
+      !self.preserve_version &&
+      (idle_threshold < 0 ||
+        (idle_threshold > 0 && self.modified_at_was > db_current_time-idle_threshold.seconds))
+      return false
+    end
+    return true
+  end
+
+  def check_encoding
+    if !(manifest_text.encoding.name == 'UTF-8' and manifest_text.valid_encoding?)
       begin
         # If Ruby thinks the encoding is something else, like 7-bit
         # ASCII, but its stored bytes are equal to the (valid) UTF-8
@@ -224,7 +377,7 @@ class Collection < ArvadosModel
       rescue
       end
       errors.add :manifest_text, "must use UTF-8 encoding"
-      false
+      throw(:abort)
     end
   end
 
@@ -234,26 +387,26 @@ class Collection < ArvadosModel
       true
     rescue ArgumentError => e
       errors.add :manifest_text, e.message
-      false
+      throw(:abort)
     end
   end
 
-  def signed_manifest_text
+  def signed_manifest_text_only_for_tests
     if !has_attribute? :manifest_text
       return nil
     elsif is_trashed
       return manifest_text
     else
       token = Thread.current[:token]
-      exp = [db_current_time.to_i + Rails.configuration.blob_signature_ttl,
+      exp = [db_current_time.to_i + Rails.configuration.Collections.BlobSigningTTL.to_i,
              trash_at].compact.map(&:to_i).min
-      self.class.sign_manifest manifest_text, token, exp
+      self.class.sign_manifest_only_for_tests manifest_text, token, exp
     end
   end
 
-  def self.sign_manifest manifest, token, exp=nil
+  def self.sign_manifest_only_for_tests manifest, token, exp=nil
     if exp.nil?
-      exp = db_current_time.to_i + Rails.configuration.blob_signature_ttl
+      exp = db_current_time.to_i + Rails.configuration.Collections.BlobSigningTTL.to_i
     end
     signing_opts = {
       api_token: token,
@@ -363,7 +516,7 @@ class Collection < ArvadosModel
   #
   # If filter_compatible_format is true (the default), only return image
   # collections which are support by the installation as indicated by
-  # Rails.configuration.docker_image_formats.  Will follow
+  # Rails.configuration.Containers.SupportedDockerImageFormats.  Will follow
   # 'docker_image_migration' links if search_term resolves to an incompatible
   # image, but an equivalent compatible image is available.
   def self.find_all_for_docker_image(search_term, search_tag=nil, readers=nil, filter_compatible_format: true)
@@ -374,15 +527,17 @@ class Collection < ArvadosModel
       joins("JOIN collections ON links.head_uuid = collections.uuid").
       order("links.created_at DESC")
 
-    if (Rails.configuration.docker_image_formats.include? 'v1' and
-        Rails.configuration.docker_image_formats.include? 'v2') or filter_compatible_format == false
+    docker_image_formats = Rails.configuration.Containers.SupportedDockerImageFormats.keys.map(&:to_s)
+
+    if (docker_image_formats.include? 'v1' and
+        docker_image_formats.include? 'v2') or filter_compatible_format == false
       pattern = /^(sha256:)?[0-9A-Fa-f]{64}\.tar$/
-    elsif Rails.configuration.docker_image_formats.include? 'v2'
+    elsif docker_image_formats.include? 'v2'
       pattern = /^(sha256:)[0-9A-Fa-f]{64}\.tar$/
-    elsif Rails.configuration.docker_image_formats.include? 'v1'
+    elsif docker_image_formats.include? 'v1'
       pattern = /^[0-9A-Fa-f]{64}\.tar$/
     else
-      raise "Unrecognized configuration for docker_image_formats #{Rails.configuration.docker_image_formats}"
+      raise "Unrecognized configuration for docker_image_formats #{docker_image_formats}"
     end
 
     # If the search term is a Collection locator that contains one file
@@ -390,7 +545,16 @@ class Collection < ArvadosModel
     if loc = Keep::Locator.parse(search_term)
       loc.strip_hints!
       coll_match = readable_by(*readers).where(portable_data_hash: loc.to_s).limit(1)
-      return get_compatible_images(readers, pattern, coll_match)
+      rc = Rails.configuration.RemoteClusters.select{ |k|
+        k != :"*" && k != Rails.configuration.ClusterID}
+      if coll_match.any? or rc.length == 0
+        return get_compatible_images(readers, pattern, coll_match)
+      else
+        # Allow bare pdh that doesn't exist in the local database so
+        # that federated container requests which refer to remotely
+        # stored containers will validate.
+        return [Collection.new(portable_data_hash: loc.to_s)]
+      end
     end
 
     if search_tag.nil? and (n = search_term.index(":"))
@@ -438,12 +602,7 @@ class Collection < ArvadosModel
   end
 
   def self.full_text_searchable_columns
-    super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed"]
-  end
-
-  def self.where *args
-    SweepTrashedObjects.sweep_if_stale
-    super
+    super - ["manifest_text", "storage_classes_desired", "storage_classes_confirmed", "current_version_uuid"]
   end
 
   protected
@@ -453,11 +612,28 @@ class Collection < ArvadosModel
   # validation on empty desired storage classes return an error.
   def default_storage_classes
     if self.storage_classes_desired.nil? || self.storage_classes_desired.empty?
-      self.storage_classes_desired = ["default"]
+      self.storage_classes_desired = Rails.configuration.DefaultStorageClasses
     end
     self.storage_classes_confirmed ||= []
   end
 
+  # Sets managed properties at creation time
+  def managed_properties
+    managed_props = Rails.configuration.Collections.ManagedProperties.with_indifferent_access
+    if managed_props.empty?
+      return
+    end
+    (managed_props.keys - self.properties.keys).each do |key|
+      if managed_props[key]['Function'] == 'original_owner'
+        self.properties[key] = self.user_owner_uuid
+      elsif managed_props[key]['Value']
+        self.properties[key] = managed_props[key]['Value']
+      else
+        logger.warn "Unidentified default property definition '#{key}': #{managed_props[key].inspect}"
+      end
+    end
+  end
+
   def portable_manifest_text
     self.class.munge_manifest_locators(manifest_text) do |match|
       if match[2] # size
@@ -511,4 +687,53 @@ class Collection < ArvadosModel
       end
     end
   end
+
+  def past_versions_cannot_be_updated
+    if is_past_version?
+      errors.add(:base, "past versions cannot be updated")
+      false
+    end
+  end
+
+  def protected_managed_properties_updates
+    managed_properties = Rails.configuration.Collections.ManagedProperties.with_indifferent_access
+    if managed_properties.empty? || !properties_changed? || current_user.is_admin
+      return true
+    end
+    protected_props = managed_properties.keys.select do |p|
+      Rails.configuration.Collections.ManagedProperties[p]['Protected']
+    end
+    # Pre-existent protected properties can't be updated
+    invalid_updates = properties_was.keys.select{|p| properties_was[p] != properties[p]} & protected_props
+    if !invalid_updates.empty?
+      invalid_updates.each do |p|
+        errors.add("protected property cannot be updated:", p)
+      end
+      raise PermissionDeniedError.new
+    end
+    true
+  end
+
+  def versioning_metadata_updates
+    valid = true
+    if !is_past_version? && current_version_uuid_changed?
+      errors.add(:current_version_uuid, "cannot be updated")
+      valid = false
+    end
+    if version_changed?
+      errors.add(:version, "cannot be updated")
+      valid = false
+    end
+    valid
+  end
+
+  def assign_uuid
+    super
+    self.current_version_uuid ||= self.uuid
+    true
+  end
+
+  def log_update
+    super unless (saved_changes.keys - UNLOGGED_CHANGES).empty?
+  end
 end