When speaking to version 0.2-beta of the build worker, properly lookup the cached commands and see if we have a matching image/tag in the repository

buildman/jobutil/buildjob.py

@@ -31,10 +31,58 @@ class BuildJob(object):
           'Could not parse repository build job config with ID %s' % self._job_details['build_uuid']
       )
 
-  def determine_cached_tag(self):
+  def determine_cached_tag(self, base_image_id=None, cache_comments=None):
     """ Returns the tag to pull to prime the cache or None if none. """
-    # TODO(jschorr): Change this to use the more complicated caching rules, once we have caching
-    # be a pull of things besides the constructed tags.
+    cached_tag = None
+    if base_image_id and cache_comments:
+      cached_tag = self._determine_cached_tag_by_comments(base_image_id, cache_comments)
+
+    if not cached_tag:
+      cached_tag = self._determine_cached_tag_by_tag()
+
+    return cached_tag
+
+  def _determine_cached_tag_by_comments(self, base_image_id, cache_commands):
+    """ Determines the tag to use for priming the cache for this build job, by matching commands
+        starting at the given base_image_id. This mimics the Docker cache checking, so it should,
+        in theory, provide "perfect" caching.
+    """
+    # Lookup the base image in the repository. If it doesn't exist, nothing more to do.
+    repo_namespace = self._repo_build.repository.namespace_user.username
+    repo_name = self._repo_build.repository.name
+
+    repository = model.get_repository(repo_namespace, repo_name)
+    if repository is None:
+      # Should never happen, but just to be sure.
+      return None
+
+    current_image = model.get_image(repository, base_image_id)
+    if current_image is None:
+      return None
+
+    # For each cache comment, find a child image that matches the command.
+    for cache_command in cache_commands:
+      print current_image.docker_image_id
+
+      current_image = model.find_child_image(repository, current_image, cache_command)
+      if current_image is None:
+        return None
+
+    # Find a tag associated with the image, if any.
+    # TODO(jschorr): We should just return the image ID instead of a parent tag, OR we should
+    # make this more efficient.
+    for tag in model.list_repository_tags(repo_namespace, repo_name):
+      tag_image = tag.image
+      ancestor_index = '/%s/' % current_image.id
+      if ancestor_index in tag_image.ancestors:
+        return tag.name
+
+    return None
+
+  def _determine_cached_tag_by_tag(self):
+    """ Determines the cached tag by looking for one of the tags being built, and seeing if it
+        exists in the repository. This is a fallback for when no comment information is available.
+    """
     tags = self._build_config.get('docker_tags', ['latest'])
     existing_tags = model.list_repository_tags(self._repo_build.repository.namespace_user.username,
                                                self._repo_build.repository.name)

buildman/jobutil/workererror.py

@@ -57,6 +57,11 @@ class WorkerError(object):
         'io.quay.builder.missingorinvalidargument': {
             'message': 'Missing required arguments for builder',
             'is_internal': True
+        },
+
+        'io.quay.builder.cachelookupissue': {
+            'message': 'Error checking for a cached tag',
+            'is_internal': True
         }
     }