X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/3f92c7068e94ee3a8f6bbe1907f2dd369c62cd7c..2130de30acf0a3b89e06494f957aacb350c15067:/services/api/app/models/container.rb

diff --git a/services/api/app/models/container.rb b/services/api/app/models/container.rb
index 1dbdb57105..079ac4c299 100644
--- a/services/api/app/models/container.rb
+++ b/services/api/app/models/container.rb
@@ -5,6 +5,7 @@
 require 'log_reuse_info'
 require 'whitelist_update'
 require 'safe_json'
+require 'update_priority'
 
 class Container < ArvadosModel
   include ArvadosModelUpdates
@@ -22,11 +23,13 @@ class Container < ArvadosModel
   serialize :command, Array
   serialize :scheduling_parameters, Hash
   serialize :secret_mounts, Hash
+  serialize :runtime_status, Hash
 
   before_validation :fill_field_defaults, :if => :new_record?
   before_validation :set_timestamps
   validates :command, :container_image, :output_path, :cwd, :priority, { presence: true }
   validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
+  validate :validate_runtime_status
   validate :validate_state_change
   validate :validate_change
   validate :validate_lock
@@ -35,8 +38,11 @@ class Container < ArvadosModel
   before_save :sort_serialized_attrs
   before_save :update_secret_mounts_md5
   before_save :scrub_secret_mounts
+  before_save :clear_runtime_status_when_queued
+  after_save :update_cr_logs
   after_save :handle_completed
   after_save :propagate_priority
+  after_commit { UpdatePriority.run_update_thread }
 
   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
   belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
@@ -56,6 +62,7 @@ class Container < ArvadosModel
     t.add :priority
     t.add :progress
     t.add :runtime_constraints
+    t.add :runtime_status
     t.add :started_at
     t.add :state
     t.add :auth_uuid
@@ -126,7 +133,6 @@ class Container < ArvadosModel
       # Update the priority of child container requests to match new
       # priority of the parent container (ignoring requests with no
       # container assigned, because their priority doesn't matter).
-      ActiveRecord::Base.connection.execute('LOCK container_requests, containers IN EXCLUSIVE MODE')
       ContainerRequest.
         where(requesting_container_uuid: self.uuid,
               state: ContainerRequest::Committed).
@@ -227,13 +233,13 @@ class Container < ArvadosModel
 
   def self.find_reusable(attrs)
     log_reuse_info { "starting with #{Container.all.count} container records in database" }
-    candidates = Container.where_serialized(:command, attrs[:command])
+    candidates = Container.where_serialized(:command, attrs[:command], md5: true)
     log_reuse_info(candidates) { "after filtering on command #{attrs[:command].inspect}" }
 
     candidates = candidates.where('cwd = ?', attrs[:cwd])
     log_reuse_info(candidates) { "after filtering on cwd #{attrs[:cwd].inspect}" }
 
-    candidates = candidates.where_serialized(:environment, attrs[:environment])
+    candidates = candidates.where_serialized(:environment, attrs[:environment], md5: true)
     log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
 
     candidates = candidates.where('output_path = ?', attrs[:output_path])
@@ -243,13 +249,14 @@ class Container < ArvadosModel
     candidates = candidates.where('container_image = ?', image)
     log_reuse_info(candidates) { "after filtering on container_image #{image.inspect} (resolved from #{attrs[:container_image].inspect})" }
 
-    candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]))
+    candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]), md5: true)
     log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
 
-    candidates = candidates.where('secret_mounts_md5 = ?', Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts]))))
-    log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
+    secret_mounts_md5 = Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts])))
+    candidates = candidates.where('secret_mounts_md5 = ?', secret_mounts_md5)
+    log_reuse_info(candidates) { "after filtering on secret_mounts_md5 #{secret_mounts_md5.inspect}" }
 
-    candidates = candidates.where_serialized(:runtime_constraints, resolve_runtime_constraints(attrs[:runtime_constraints]))
+    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}" }
 
     log_reuse_info { "checking for state=Complete with readable output and log..." }
@@ -274,9 +281,10 @@ class Container < ArvadosModel
       return usable
     end
 
-    # Check for Running candidates and return the most likely to finish sooner.
+    # Check for non-failing Running candidates and return the most likely to finish sooner.
     log_reuse_info { "checking for state=Running..." }
     running = candidates.where(state: Running).
+              where("(runtime_status->'error') is null").
               order('progress desc, started_at asc').
               limit(1).first
     if running
@@ -316,10 +324,6 @@ class Container < ArvadosModel
     # (because state might have changed while acquiring the lock).
     check_lock_fail
     transaction do
-      # Locking involves assigning auth_uuid, which involves looking
-      # up container requests, so we must lock both tables in the
-      # proper order to avoid deadlock.
-      ActiveRecord::Base.connection.execute('LOCK container_requests, containers IN EXCLUSIVE MODE')
       reload
       check_lock_fail
       update_attributes!(state: Locked)
@@ -374,24 +378,11 @@ class Container < ArvadosModel
     current_user.andand.is_admin
   end
 
-  def permission_to_update
-    # Override base permission check to allow auth_uuid to set progress and
-    # output (only).  Whether it is legal to set progress and output in the current
-    # state has already been checked in validate_change.
-    current_user.andand.is_admin ||
-      (!current_api_client_authorization.nil? and
-       [self.auth_uuid, self.locked_by_uuid].include? current_api_client_authorization.uuid)
-  end
-
   def ensure_owner_uuid_is_permitted
-    # Override base permission check to allow auth_uuid to set progress and
-    # output (only).  Whether it is legal to set progress and output in the current
-    # state has already been checked in validate_change.
-    if !current_api_client_authorization.nil? and self.auth_uuid == current_api_client_authorization.uuid
-      check_update_whitelist [:progress, :output]
-    else
-      super
-    end
+    # validate_change ensures owner_uuid can't be changed at all --
+    # except during create, which requires admin privileges. Checking
+    # permission here would be superfluous.
+    true
   end
 
   def set_timestamps
@@ -404,8 +395,21 @@ class Container < ArvadosModel
     end
   end
 
+  # Check that well-known runtime status keys have desired data types
+  def validate_runtime_status
+    [
+      'error', 'errorDetail', 'warning', 'warningDetail', 'activity'
+    ].each do |k|
+      if self.runtime_status.andand.include?(k) && !self.runtime_status[k].is_a?(String)
+        errors.add(:runtime_status, "'#{k}' value must be a string")
+      end
+    end
+  end
+
   def validate_change
     permitted = [:state]
+    progress_attrs = [:progress, :runtime_status, :log, :output]
+    final_attrs = [:exit_code, :finished_at]
 
     if self.new_record?
       permitted.push(:owner_uuid, :command, :container_image, :cwd,
@@ -415,24 +419,27 @@ class Container < ArvadosModel
     end
 
     case self.state
-    when Queued, Locked
+    when Locked
+      permitted.push :priority, :runtime_status, :log
+
+    when Queued
       permitted.push :priority
 
     when Running
-      permitted.push :priority, :progress, :output
+      permitted.push :priority, *progress_attrs
       if self.state_changed?
         permitted.push :started_at
       end
 
     when Complete
       if self.state_was == Running
-        permitted.push :finished_at, :output, :log, :exit_code
+        permitted.push *final_attrs, *progress_attrs
       end
 
     when Cancelled
       case self.state_was
       when Running
-        permitted.push :finished_at, :output, :log
+        permitted.push :finished_at, *progress_attrs
       when Queued, Locked
         permitted.push :finished_at, :log
       end
@@ -442,6 +449,15 @@ class Container < ArvadosModel
       return false
     end
 
+    if current_api_client_authorization.andand.uuid.andand == self.auth_uuid
+      # The contained process itself can update progress indicators,
+      # but can't change priority etc.
+      permitted = permitted & (progress_attrs + final_attrs + [:state] - [:log])
+    elsif self.locked_by_uuid && self.locked_by_uuid != current_api_client_authorization.andand.uuid
+      # When locked, progress fields cannot be updated by the wrong
+      # dispatcher, even though it has admin privileges.
+      permitted = permitted - progress_attrs
+    end
     check_update_whitelist permitted
   end
 
@@ -480,6 +496,19 @@ class Container < ArvadosModel
     end
   end
 
+  def update_cr_logs
+    # If self.final?, this update is superfluous: the final log/output
+    # update will be done when handle_completed calls finalize! on
+    # each requesting CR.
+    return if self.final? || !self.log_changed?
+    leave_modified_by_user_alone do
+      ContainerRequest.where(container_uuid: self.uuid).each do |cr|
+        cr.update_collections(container: self, collections: ['log'])
+        cr.save!
+      end
+    end
+  end
+
   def assign_auth
     if self.auth_uuid_changed?
       return errors.add :auth_uuid, 'is readonly'
@@ -536,13 +565,19 @@ class Container < ArvadosModel
     end
   end
 
+  def clear_runtime_status_when_queued
+    # Avoid leaking status messages between different dispatch attempts
+    if self.state_was == Locked && self.state == Queued
+      self.runtime_status = {}
+    end
+  end
+
   def handle_completed
     # This container is finished so finalize any associated container requests
     # that are associated with this container.
     if self.state_changed? and self.final?
       act_as_system_user do
 
-        ActiveRecord::Base.connection.execute('LOCK container_requests, containers IN EXCLUSIVE MODE')
         if self.state == Cancelled
           retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
         else