X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/df55660fc8748a317219043153f0bb54150086a8..a079a9f1627188e3ef0bb73e004defa2129d25f2:/services/api/app/models/job.rb

diff --git a/services/api/app/models/job.rb b/services/api/app/models/job.rb
index 5344d45fbc..8de3897eab 100644
--- a/services/api/app/models/job.rb
+++ b/services/api/app/models/job.rb
@@ -1,3 +1,8 @@
+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: AGPL-3.0
+
+require 'log_reuse_info'
 require 'safe_json'
 
 class Job < ArvadosModel
@@ -5,6 +10,7 @@ class Job < ArvadosModel
   include KindAndEtag
   include CommonApiTemplate
   extend CurrentApiClient
+  extend LogReuseInfo
   serialize :components, Hash
   attr_protected :arvados_sdk_version, :docker_image_locator
   serialize :script_parameters, Hash
@@ -198,17 +204,6 @@ class Job < ArvadosModel
     filters
   end
 
-  # log_reuse_info logs whatever the given block returns, if
-  # log_reuse_decisions is enabled. It accepts a block instead of a
-  # string because in some cases constructing the strings involves
-  # doing database queries, and we want to skip those queries when
-  # logging is disabled.
-  def self.log_reuse_info
-    if Rails.configuration.log_reuse_decisions
-      Rails.logger.info("find_reusable: " + yield)
-    end
-  end
-
   def self.find_reusable attrs, params, filters, read_users
     if filters.empty?  # Translate older creation parameters into filters.
       filters =
@@ -254,22 +249,23 @@ class Job < ArvadosModel
     candidates = candidates.where(
       'state = ? or (owner_uuid = ? and state in (?))',
       Job::Complete, current_user.uuid, [Job::Queued, Job::Running])
-    log_reuse_info { "have #{candidates.count} candidates after filtering on job state ((state=Complete) or (state=Queued/Running and (submitted by current user)))" }
+    log_reuse_info(candidates) { "after filtering on job state ((state=Complete) or (state=Queued/Running and (submitted by current user)))" }
 
     digest = Job.sorted_hash_digest(attrs[:script_parameters])
     candidates = candidates.where('script_parameters_digest = ?', digest)
-    log_reuse_info { "have #{candidates.count} candidates after filtering on script_parameters_digest #{digest}" }
+    log_reuse_info(candidates) { "after filtering on script_parameters_digest #{digest}" }
 
     candidates = candidates.where('nondeterministic is distinct from ?', true)
-    log_reuse_info { "have #{candidates.count} candidates after filtering on !nondeterministic" }
+    log_reuse_info(candidates) { "after filtering on !nondeterministic" }
 
     # prefer Running jobs over Queued
     candidates = candidates.order('state desc, created_at')
 
     candidates = apply_filters candidates, filters
-    log_reuse_info { "have #{candidates.count} candidates after filtering on repo, script, and custom filters #{filters.inspect}" }
+    log_reuse_info(candidates) { "after filtering on repo, script, and custom filters #{filters.inspect}" }
 
     chosen = nil
+    chosen_output = nil
     incomplete_job = nil
     candidates.each do |j|
       if j.state != Job::Complete
@@ -284,17 +280,25 @@ class Job < ArvadosModel
         # Ignore: we have already decided not to reuse any completed
         # job.
         log_reuse_info { "job #{j.uuid} with output #{j.output} ignored, see above" }
+      elsif j.output.nil?
+        log_reuse_info { "job #{j.uuid} has nil output" }
+      elsif j.log.nil?
+        log_reuse_info { "job #{j.uuid} has nil log" }
       elsif Rails.configuration.reuse_job_if_outputs_differ
-        if Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
-          log_reuse_info { "job #{j.uuid} with output #{j.output} is reusable; decision is final." }
-          return j
-        else
-          # Ignore: keep locking for an incomplete job or one whose
+        if !Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
+          # Ignore: keep looking for an incomplete job or one whose
           # output is readable.
           log_reuse_info { "job #{j.uuid} output #{j.output} unavailable to user; continuing search" }
+        elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.log)
+          # Ignore: keep looking for an incomplete job or one whose
+          # log is readable.
+          log_reuse_info { "job #{j.uuid} log #{j.log} unavailable to user; continuing search" }
+        else
+          log_reuse_info { "job #{j.uuid} with output #{j.output} is reusable; decision is final." }
+          return j
         end
-      elsif chosen
-        if chosen.output != j.output
+      elsif chosen_output
+        if chosen_output != j.output
           # If two matching jobs produced different outputs, run a new
           # job (or use one that's already running/queued) instead of
           # choosing one arbitrarily.
@@ -313,9 +317,15 @@ class Job < ArvadosModel
         # any further investigation of reusable jobs is futile.
         log_reuse_info { "job #{j.uuid} output #{j.output} is unavailable to user; this means no finished job can be reused (see reuse_job_if_outputs_differ in application.default.yml)" }
         chosen = false
+      elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.log)
+        # This user cannot read the log of this job, don't try to reuse the
+        # job but consider if the output is consistent.
+        log_reuse_info { "job #{j.uuid} log #{j.log} is unavailable to user; continuing search" }
+        chosen_output = j.output
       else
         log_reuse_info { "job #{j.uuid} with output #{j.output} can be reused; continuing search in case other candidates have different outputs" }
         chosen = j
+        chosen_output = j.output
       end
     end
     j = chosen || incomplete_job