X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/f159fab8f9d6bc4254192ce43432defd5bd400aa..d97388bdbfeb6a43cb86996012a1db0ba4a8871f:/services/api/app/models/container.rb

diff --git a/services/api/app/models/container.rb b/services/api/app/models/container.rb
index 7176bda926..abcfdbd296 100644
--- a/services/api/app/models/container.rb
+++ b/services/api/app/models/container.rb
@@ -23,11 +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
@@ -35,7 +37,9 @@ class Container < ArvadosModel
   after_validation :assign_auth
   before_save :sort_serialized_attrs
   before_save :update_secret_mounts_md5
-  before_save :scrub_secret_mounts
+  before_save :scrub_secrets
+  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 }
@@ -58,10 +62,13 @@ 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
     t.add :scheduling_parameters
+    t.add :runtime_user_uuid
+    t.add :runtime_auth_scopes
   end
 
   # Supported states for a container
@@ -86,15 +93,15 @@ class Container < ArvadosModel
   end
 
   def self.full_text_searchable_columns
-    super - ["secret_mounts", "secret_mounts_md5"]
+    super - ["secret_mounts", "secret_mounts_md5", "runtime_token"]
   end
 
   def self.searchable_columns *args
-    super - ["secret_mounts_md5"]
+    super - ["secret_mounts_md5", "runtime_token"]
   end
 
   def logged_attributes
-    super.except('secret_mounts')
+    super.except('secret_mounts', 'runtime_token')
   end
 
   def state_transitions
@@ -141,17 +148,37 @@ class Container < ArvadosModel
   # Create a new container (or find an existing one) to satisfy the
   # given container request.
   def self.resolve(req)
-    c_attrs = {
-      command: req.command,
-      cwd: req.cwd,
-      environment: req.environment,
-      output_path: req.output_path,
-      container_image: resolve_container_image(req.container_image),
-      mounts: resolve_mounts(req.mounts),
-      runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
-      scheduling_parameters: req.scheduling_parameters,
-      secret_mounts: req.secret_mounts,
-    }
+    if req.runtime_token.nil?
+      runtime_user = if req.modified_by_user_uuid.nil?
+                       current_user
+                     else
+                       User.find_by_uuid(req.modified_by_user_uuid)
+                     end
+      runtime_auth_scopes = ["all"]
+    else
+      auth = ApiClientAuthorization.validate(token: req.runtime_token)
+      if auth.nil?
+        raise ArgumentError.new "Invalid runtime token"
+      end
+      runtime_user = User.find_by_id(auth.user_id)
+      runtime_auth_scopes = auth.scopes
+    end
+    c_attrs = act_as_user runtime_user do
+      {
+        command: req.command,
+        cwd: req.cwd,
+        environment: req.environment,
+        output_path: req.output_path,
+        container_image: resolve_container_image(req.container_image),
+        mounts: resolve_mounts(req.mounts),
+        runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
+        scheduling_parameters: req.scheduling_parameters,
+        secret_mounts: req.secret_mounts,
+        runtime_token: req.runtime_token,
+        runtime_user_uuid: runtime_user.uuid,
+        runtime_auth_scopes: runtime_auth_scopes
+      }
+    end
     act_as_system_user do
       if req.use_existing && (reusable = find_reusable(c_attrs))
         reusable
@@ -196,7 +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).
@@ -205,13 +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
@@ -276,9 +301,10 @@ 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
@@ -320,7 +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
 
@@ -338,7 +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
 
@@ -349,6 +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
 
@@ -356,6 +392,19 @@ class Container < ArvadosModel
     [Complete, Cancelled].include?(self.state)
   end
 
+  def self.for_current_token
+    return if !current_api_client_authorization
+    _, _, _, container_uuid = Thread.current[:token].split('/')
+    if container_uuid.nil?
+      Container.where(auth_uuid: current_api_client_authorization.uuid).first
+    else
+      Container.where('auth_uuid=? or (uuid=? and runtime_token=?)',
+                      current_api_client_authorization.uuid,
+                      container_uuid,
+                      current_api_client_authorization.token).first
+    end
+  end
+
   protected
 
   def fill_field_defaults
@@ -372,24 +421,11 @@ 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
@@ -402,37 +438,54 @@ 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,
                      :environment, :mounts, :output_path, :priority,
                      :runtime_constraints, :scheduling_parameters,
-                     :secret_mounts)
+                     :secret_mounts, :runtime_token,
+                     :runtime_user_uuid, :runtime_auth_scopes)
     end
 
     case self.state
-    when Queued, Locked
+    when Locked
+      permitted.push :priority, :runtime_status, :log, :lock_count
+
+    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
+        permitted.push :finished_at, :log, :runtime_status
       end
 
     else
@@ -440,6 +493,19 @@ class Container < ArvadosModel
       return false
     end
 
+    if self.state == Running &&
+       !current_api_client_authorization.nil? &&
+       (current_api_client_authorization.uuid == self.auth_uuid ||
+        current_api_client_authorization.token == self.runtime_token)
+      # The contained process itself can write final attrs but can't
+      # change priority or log.
+      permitted.push *final_attrs
+      permitted = permitted - [:log, :priority]
+    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
 
@@ -478,9 +544,22 @@ 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'
+         return errors.add :auth_uuid, 'is readonly'
     end
     if not [Locked, Running].include? self.state
       # don't need one
@@ -491,16 +570,29 @@ class Container < ArvadosModel
       # already have one
       return
     end
-    cr = ContainerRequest.
-      where('container_uuid=? and priority>0', self.uuid).
-      order('priority desc').
-      first
-    if !cr
-      return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
+    if self.runtime_token.nil?
+      if self.runtime_user_uuid.nil?
+        # legacy behavior, we don't have a runtime_user_uuid so get
+        # the user from the highest priority container request, needed
+        # when performing an upgrade and there are queued containers,
+        # and some tests.
+        cr = ContainerRequest.
+               where('container_uuid=? and priority>0', self.uuid).
+               order('priority desc').
+               first
+        if !cr
+          return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
+        end
+        self.runtime_user_uuid = cr.modified_by_user_uuid
+        self.runtime_auth_scopes = ["all"]
+      end
+
+      # generate a new token
+      self.auth = ApiClientAuthorization.
+                    create!(user_id: User.find_by_uuid(self.runtime_user_uuid).id,
+                            api_client_id: 0,
+                            scopes: self.runtime_auth_scopes)
     end
-    self.auth = ApiClientAuthorization.
-      create!(user_id: User.find_by_uuid(cr.modified_by_user_uuid).id,
-              api_client_id: 0)
   end
 
   def sort_serialized_attrs
@@ -516,6 +608,9 @@ class Container < ArvadosModel
     if self.scheduling_parameters_changed?
       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
     end
+    if self.runtime_auth_scopes_changed?
+      self.runtime_auth_scopes = self.runtime_auth_scopes.sort
+    end
   end
 
   def update_secret_mounts_md5
@@ -525,12 +620,20 @@ class Container < ArvadosModel
     end
   end
 
-  def scrub_secret_mounts
+  def scrub_secrets
     # this runs after update_secret_mounts_md5, so the
     # secret_mounts_md5 will still reflect the secrets that are being
     # scrubbed here.
     if self.state_changed? && self.final?
       self.secret_mounts = {}
+      self.runtime_token = nil
+    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
 
@@ -555,7 +658,11 @@ class Container < ArvadosModel
             container_image: self.container_image,
             mounts: self.mounts,
             runtime_constraints: self.runtime_constraints,
-            scheduling_parameters: self.scheduling_parameters
+            scheduling_parameters: self.scheduling_parameters,
+            secret_mounts: self.secret_mounts_was,
+            runtime_token: self.runtime_token_was,
+            runtime_user_uuid: self.runtime_user_uuid,
+            runtime_auth_scopes: self.runtime_auth_scopes
           }
           c = Container.create! c_attrs
           retryable_requests.each do |cr|