X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/b3f88934721fc97973b40786cae97e55e84b31d5..704dd826a498db11a052fe77a4953b0caa245c10:/services/api/app/models/container.rb

diff --git a/services/api/app/models/container.rb b/services/api/app/models/container.rb
index a22d6204b0..b013776b98 100644
--- a/services/api/app/models/container.rb
+++ b/services/api/app/models/container.rb
@@ -1,3 +1,8 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+require 'log_reuse_info'
 require 'whitelist_update'
 require 'safe_json'
 
@@ -8,6 +13,7 @@ class Container < ArvadosModel
   include WhitelistUpdate
   extend CurrentApiClient
   extend DbCurrentTime
+  extend LogReuseInfo
 
   serialize :environment, Hash
   serialize :mounts, Hash
@@ -18,6 +24,7 @@ class Container < ArvadosModel
   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, less_than_or_equal_to: 1000 }
   validate :validate_state_change
   validate :validate_change
   validate :validate_lock
@@ -25,6 +32,7 @@ class Container < ArvadosModel
   after_validation :assign_auth
   before_save :sort_serialized_attrs
   after_save :handle_completed
+  after_save :propagate_priority
 
   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
@@ -82,11 +90,25 @@ class Container < ArvadosModel
       self.priority = ContainerRequest.
         where(container_uuid: uuid,
               state: ContainerRequest::Committed).
-        maximum('priority')
+        maximum('priority') || 0
       self.save!
     end
   end
 
+  def propagate_priority
+    if self.priority_changed?
+      act_as_system_user do
+         # Update the priority of child container requests to match new priority
+         # of the parent container.
+         ContainerRequest.where(requesting_container_uuid: self.uuid,
+                                state: ContainerRequest::Committed).each do |cr|
+           cr.priority = self.priority
+           cr.save
+         end
+       end
+    end
+  end
+
   # Create a new container (or find an existing one) to satisfy the
   # given container request.
   def self.resolve(req)
@@ -175,41 +197,76 @@ class Container < ArvadosModel
   end
 
   def self.find_reusable(attrs)
-    candidates = Container.
-      where_serialized(:command, attrs[:command]).
-      where('cwd = ?', attrs[:cwd]).
-      where_serialized(:environment, attrs[:environment]).
-      where('output_path = ?', attrs[:output_path]).
-      where('container_image = ?', resolve_container_image(attrs[:container_image])).
-      where_serialized(:mounts, resolve_mounts(attrs[:mounts])).
-      where_serialized(:runtime_constraints, resolve_runtime_constraints(attrs[:runtime_constraints]))
-
-    # Check for Completed candidates whose output and log are both readable.
+    log_reuse_info { "starting with #{Container.all.count} container records in database" }
+    candidates = Container.where_serialized(:command, attrs[:command])
+    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])
+    log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
+
+    candidates = candidates.where('output_path = ?', attrs[:output_path])
+    log_reuse_info(candidates) { "after filtering on output_path #{attrs[:output_path].inspect}" }
+
+    image = resolve_container_image(attrs[:container_image])
+    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]))
+    log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
+
+    candidates = candidates.where_serialized(:runtime_constraints, resolve_runtime_constraints(attrs[:runtime_constraints]))
+    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..." }
+
     select_readable_pdh = Collection.
       readable_by(current_user).
       select(:portable_data_hash).
       to_sql
-    usable = candidates.
-      where(state: Complete).
-      where(exit_code: 0).
-      where("log IN (#{select_readable_pdh})").
-      where("output IN (#{select_readable_pdh})").
-      order('finished_at ASC').
-      limit(1).
-      first
-    return usable if usable
+
+    usable = candidates.where(state: Complete, exit_code: 0)
+    log_reuse_info(usable) { "with state=Complete, exit_code=0" }
+
+    usable = usable.where("log IN (#{select_readable_pdh})")
+    log_reuse_info(usable) { "with readable log" }
+
+    usable = usable.where("output IN (#{select_readable_pdh})")
+    log_reuse_info(usable) { "with readable output" }
+
+    usable = usable.order('finished_at ASC').limit(1).first
+    if usable
+      log_reuse_info { "done, reusing container #{usable.uuid} with state=Complete" }
+      return usable
+    end
 
     # Check for Running candidates and return the most likely to finish sooner.
+    log_reuse_info { "checking for state=Running..." }
     running = candidates.where(state: Running).
-      order('progress desc, started_at asc').limit(1).first
-    return running if not running.nil?
+              order('progress desc, started_at asc').
+              limit(1).first
+    if running
+      log_reuse_info { "done, reusing container #{running.uuid} with state=Running" }
+      return running
+    else
+      log_reuse_info { "have no containers in Running state" }
+    end
 
     # Check for Locked or Queued ones and return the most likely to start first.
-    locked_or_queued = candidates.where("state IN (?)", [Locked, Queued]).
-      order('state asc, priority desc, created_at asc').limit(1).first
-    return locked_or_queued if not locked_or_queued.nil?
+    locked_or_queued = candidates.
+                       where("state IN (?)", [Locked, Queued]).
+                       order('state asc, priority desc, created_at asc').
+                       limit(1).first
+    if locked_or_queued
+      log_reuse_info { "done, reusing container #{locked_or_queued.uuid} with state=#{locked_or_queued.state}" }
+      return locked_or_queued
+    else
+      log_reuse_info { "have no containers in Locked or Queued state" }
+    end
 
-    # No suitable candidate found.
+    log_reuse_info { "done, no reusable container found" }
     nil
   end
 
@@ -256,17 +313,13 @@ class Container < ArvadosModel
   end
 
   def self.readable_by(*users_list)
-    if users_list.select { |u| u.is_admin }.any?
-      return self
+    # Load optional keyword arguments, if they exist.
+    if users_list.last.is_a? Hash
+      kwargs = users_list.pop
+    else
+      kwargs = {}
     end
-    user_uuids = users_list.map { |u| u.uuid }
-    uuid_list = user_uuids + users_list.flat_map { |u| u.groups_i_can(:read) }
-    uuid_list.uniq!
-    permitted = "(SELECT head_uuid FROM links WHERE link_class='permission' AND tail_uuid IN (:uuids))"
-    joins(:container_requests).
-      where("container_requests.uuid IN #{permitted} OR "+
-            "container_requests.owner_uuid IN (:uuids)",
-            uuids: uuid_list)
+    Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").exists)
   end
 
   def final?
@@ -281,7 +334,7 @@ class Container < ArvadosModel
     self.runtime_constraints ||= {}
     self.mounts ||= {}
     self.cwd ||= "."
-    self.priority ||= 1
+    self.priority ||= 0
     self.scheduling_parameters ||= {}
   end
 
@@ -348,7 +401,7 @@ class Container < ArvadosModel
       when Running
         permitted.push :finished_at, :output, :log
       when Queued, Locked
-        permitted.push :finished_at
+        permitted.push :finished_at, :log
       end
 
     else
@@ -384,12 +437,10 @@ class Container < ArvadosModel
     # that a container cannot "claim" a collection that it doesn't otherwise
     # have access to just by setting the output field to the collection PDH.
     if output_changed?
-      c = Collection.unscoped do
-        Collection.
-            readable_by(current_user).
+      c = Collection.
+            readable_by(current_user, {include_trash: true}).
             where(portable_data_hash: self.output).
             first
-      end
       if !c
         errors.add :output, "collection must exist and be readable by current user."
       end
@@ -464,7 +515,7 @@ class Container < ArvadosModel
             cr.with_lock do
               # Use row locking because this increments container_count
               cr.container_uuid = c.uuid
-              cr.save
+              cr.save!
             end
           end
         end
@@ -475,11 +526,21 @@ class Container < ArvadosModel
           cr.finalize!
         end
 
-        # Try to cancel any outstanding container requests made by this container.
-        ContainerRequest.where(requesting_container_uuid: uuid,
-                               state: ContainerRequest::Committed).each do |cr|
-          cr.priority = 0
-          cr.save
+        # Cancel outstanding container requests made by this container.
+        ContainerRequest.
+          includes(:container).
+          where(requesting_container_uuid: uuid,
+                state: ContainerRequest::Committed).each do |cr|
+          cr.update_attributes!(priority: 0)
+          cr.container.reload
+          if cr.container.state == Container::Queued || cr.container.state == Container::Locked
+            # If the child container hasn't started yet, finalize the
+            # child CR now instead of leaving it "on hold", i.e.,
+            # Queued with priority 0.  (OTOH, if the child is already
+            # running, leave it alone so it can get cancelled the
+            # usual way, get a copy of the log collection, etc.)
+            cr.update_attributes!(state: ContainerRequest::Final)
+          end
         end
       end
     end