X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/62a0bbe9c1f6e3376b25d0f415302eb7dcdf55f3..6dd8a072ec2e305df687f72dd294d760ae6c8e23:/services/api/app/models/job.rb

diff --git a/services/api/app/models/job.rb b/services/api/app/models/job.rb
index 90fba3c797..fa38ece244 100644
--- a/services/api/app/models/job.rb
+++ b/services/api/app/models/job.rb
@@ -1,3 +1,4 @@
+require 'log_reuse_info'
 require 'safe_json'
 
 class Job < ArvadosModel
@@ -5,6 +6,7 @@ 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
@@ -237,45 +239,83 @@ class Job < ArvadosModel
     end
 
     # Search for a reusable Job, and return it if found.
-    candidates = Job.
-      readable_by(current_user).
-      where('state = ? or (owner_uuid = ? and state in (?))',
-            Job::Complete, current_user.uuid, [Job::Queued, Job::Running]).
-      where('script_parameters_digest = ?', Job.sorted_hash_digest(attrs[:script_parameters])).
-      where('nondeterministic is distinct from ?', true).
-      order('state desc, created_at') # prefer Running jobs over Queued
+    candidates = Job.readable_by(current_user)
+    log_reuse_info { "starting with #{candidates.count} jobs readable by current user #{current_user.uuid}" }
+
+    candidates = candidates.where(
+      'state = ? or (owner_uuid = ? and state in (?))',
+      Job::Complete, current_user.uuid, [Job::Queued, Job::Running])
+    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(candidates) { "after filtering on script_parameters_digest #{digest}" }
+
+    candidates = candidates.where('nondeterministic is distinct from ?', true)
+    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(candidates) { "after filtering on repo, script, and custom filters #{filters.inspect}" }
+
     chosen = nil
     incomplete_job = nil
     candidates.each do |j|
       if j.state != Job::Complete
-        # We'll use this if we don't find a job that has completed
-        incomplete_job ||= j
-        next
-      end
-
-      if chosen == false
-        # We have already decided not to reuse any completed job
-        next
+        if !incomplete_job
+          # We'll use this if we don't find a job that has completed
+          log_reuse_info { "job #{j.uuid} is reusable, but unfinished; continuing search for completed jobs" }
+          incomplete_job = j
+        else
+          log_reuse_info { "job #{j.uuid} is unfinished and we already have #{incomplete_job.uuid}; ignoring" }
+        end
+      elsif chosen == false
+        # 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 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
+          # output is readable.
+          log_reuse_info { "job #{j.uuid} output #{j.output} unavailable to user; continuing search" }
+        end
       elsif chosen
         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.
+          log_reuse_info { "job #{j.uuid} output #{j.output} disagrees; forgetting about #{chosen.uuid} and ignoring any other finished jobs (see reuse_job_if_outputs_differ in application.default.yml)" }
           chosen = false
+        else
+          log_reuse_info { "job #{j.uuid} output #{j.output} agrees with chosen #{chosen.uuid}; continuing search in case other candidates have different outputs" }
         end
         # ...and that's the only thing we need to do once we've chosen
         # a job to reuse.
       elsif !Collection.readable_by(current_user).find_by_portable_data_hash(j.output)
-        # As soon as the output we will end up returning (if any) is
-        # decided, check whether it will be visible to the user; if
-        # not, any further investigation of reusable jobs is futile.
+        # This user cannot read the output of this job. Any other
+        # completed job will have either the same output (making it
+        # unusable) or a different output (making it unusable because
+        # reuse_job_if_outputs_different is turned off). Therefore,
+        # 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
       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
       end
     end
-    chosen || incomplete_job
+    j = chosen || incomplete_job
+    if j
+      log_reuse_info { "done, #{j.uuid} was selected" }
+    else
+      log_reuse_info { "done, nothing suitable" }
+    end
+    return j
   end
 
   def self.default_git_filters(attr_name, repo_name, refspec)