Add a RepositorySearchScore table and calculation to the RAC worker

This will be used in a followup PR to order search results instead of the RAC join. Currently, the join with the RAC table in search results in a lookup of ~600K rows, which causes searching to take ~6s. This PR denormalizes the data we need, as well as allowing us to score based on a wider band (6 months vs the current 1 week).

data/database.py

@@ -451,8 +451,8 @@ class User(BaseModel):
                                  TagManifest, AccessToken, OAuthAccessToken, BlobUpload,
                                  RepositoryNotification, OAuthAuthorizationCode,
                                  RepositoryActionCount, TagManifestLabel, Tag,
-                                 ManifestLabel, BlobUploading, TeamSync} | beta_classes
-
+                                 ManifestLabel, BlobUploading, TeamSync,
+                                 RepositorySearchScore} | beta_classes
       delete_instance_filtered(self, User, delete_nullable, skip_transitive_deletes)
 
 
@@ -584,10 +584,18 @@ class Repository(BaseModel):
     # These models don't need to use transitive deletes, because the referenced objects
     # are cleaned up directly
     skip_transitive_deletes = {RepositoryTag, RepositoryBuild, RepositoryBuildTrigger, BlobUpload,
-                               Image, TagManifest, TagManifestLabel, Label, DerivedStorageForImage} | beta_classes
+                               Image, TagManifest, TagManifestLabel, Label, DerivedStorageForImage,
+                               RepositorySearchScore} | beta_classes
+
     delete_instance_filtered(self, Repository, delete_nullable, skip_transitive_deletes)
 
 
+class RepositorySearchScore(BaseModel):
+  repository = ForeignKeyField(Repository, unique=True)
+  score = BigIntegerField(index=True, default=0)
+  last_updated = DateTimeField(null=True)
+
+
 class Star(BaseModel):
   user = ForeignKeyField(User)
   repository = ForeignKeyField(Repository)

data/migrations/versions/f30984525c86_add_repositorysearchscore_table.py

@@ -0,0 +1,35 @@
+"""Add RepositorySearchScore table
+
+Revision ID: f30984525c86
+Revises: be8d1c402ce0
+Create Date: 2017-04-04 14:30:13.270728
+
+"""
+
+# revision identifiers, used by Alembic.
+revision = 'f30984525c86'
+down_revision = 'be8d1c402ce0'
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects import mysql
+
+def upgrade(tables):
+    ### commands auto generated by Alembic - please adjust! ###
+    op.create_table('repositorysearchscore',
+    sa.Column('id', sa.Integer(), nullable=False),
+    sa.Column('repository_id', sa.Integer(), nullable=False),
+    sa.Column('score', sa.BigInteger(), nullable=False),
+    sa.Column('last_updated', sa.DateTime(), nullable=True),
+    sa.ForeignKeyConstraint(['repository_id'], ['repository.id'], name=op.f('fk_repositorysearchscore_repository_id_repository')),
+    sa.PrimaryKeyConstraint('id', name=op.f('pk_repositorysearchscore'))
+    )
+    op.create_index('repositorysearchscore_repository_id', 'repositorysearchscore', ['repository_id'], unique=True)
+    op.create_index('repositorysearchscore_score', 'repositorysearchscore', ['score'], unique=False)
+    ### end Alembic commands ###
+
+
+def downgrade(tables):
+    ### commands auto generated by Alembic - please adjust! ###
+    op.drop_table('repositorysearchscore')
+    ### end Alembic commands ###

data/model/__init__.py

@@ -133,6 +133,7 @@ from data.model import (
   oauth,
   organization,
   permission,
+  repositoryactioncount,
   release,
   repository,
   service_keys,

data/model/repository.py

@@ -12,7 +12,7 @@ from data.database import (Repository, Namespace, RepositoryTag, Star, Image, Im
                            Visibility, RepositoryPermission, RepositoryActionCount,
                            Role, RepositoryAuthorizedEmail, TagManifest, DerivedStorageForImage,
                            Label, TagManifestLabel, db_for_update, get_epoch_timestamp,
-                           db_random_func, db_concat_func)
+                           db_random_func, db_concat_func, RepositorySearchScore)
 from data.text import prefix_search
 from util.itertoolrecipes import take
 
@@ -42,6 +42,7 @@ def create_repository(namespace, name, creating_user, visibility='private', repo
 
   yesterday = datetime.now() - timedelta(days=1)
   RepositoryActionCount.create(repository=repo, count=0, date=yesterday)
+  RepositorySearchScore.create(repository=repo, score=0)
 
   if creating_user and not creating_user.organization:
     RepositoryPermission.create(user=creating_user, repository=repo, role=admin)

data/model/repositoryactioncount.py

@@ -0,0 +1,118 @@
+import logging
+
+from collections import namedtuple
+from peewee import IntegrityError
+
+from datetime import date, timedelta, datetime
+from data.database import (Repository, LogEntry, RepositoryActionCount, RepositorySearchScore,
+                           db_random_func, fn)
+
+logger = logging.getLogger(__name__)
+
+search_bucket = namedtuple('SearchBucket', ['delta', 'days', 'weight'])
+
+# Defines the various buckets for search scoring. Each bucket is computed using the given time
+# delta from today *minus the previous bucket's time period*. Once all the actions over the
+# bucket's time period have been collected, they are multiplied by the given modifier. The modifiers
+# for this bucket were determined via the integral of (2/((x/183)+1)^2)/183 over the period of days
+# in the bucket; this integral over 0..183 has a sum of 1, so we get a good normalize score result.
+SEARCH_BUCKETS = [
+  search_bucket(timedelta(days=1), 1, 0.010870),
+  search_bucket(timedelta(days=7), 6, 0.062815),
+  search_bucket(timedelta(days=31), 24, 0.21604),
+  search_bucket(timedelta(days=183), 152, 0.71028),
+]
+
+def find_uncounted_repository():
+  """ Returns a repository that has not yet had an entry added into the RepositoryActionCount
+      table for yesterday.
+  """
+  try:
+    # Get a random repository to count.
+    today = date.today()
+    yesterday = today - timedelta(days=1)
+    has_yesterday_actions = (RepositoryActionCount
+                             .select(RepositoryActionCount.repository)
+                             .where(RepositoryActionCount.date == yesterday))
+
+    to_count = (Repository
+                .select()
+                .where(~(Repository.id << (has_yesterday_actions)))
+                .order_by(db_random_func()).get())
+    return to_count
+  except Repository.DoesNotExist:
+    return 0
+
+
+def count_repository_actions(to_count):
+  """ Aggregates repository actions from the LogEntry table for the last day and writes them to
+      the RepositoryActionCount table. Return True if the repository was updated and False
+      otherwise.
+  """
+  today = date.today()
+  yesterday = today - timedelta(days=1)
+
+  actions = (LogEntry
+             .select()
+             .where(LogEntry.repository == to_count,
+                    LogEntry.datetime >= yesterday,
+                    LogEntry.datetime < today)
+             .count())
+
+  try:
+    RepositoryActionCount.create(repository=to_count, date=yesterday, count=actions)
+    return True
+  except IntegrityError:
+    logger.exception('Exception when writing count for repository')
+    return False
+
+
+def update_repository_score(repo):
+  """ Updates the repository score entry for the given table by retrieving information from
+      the RepositoryActionCount table. Note that count_repository_actions for the repo should
+      be called first. Returns True if the row was updated and False otherwise.
+  """
+  today = date.today()
+
+  # Retrieve the counts for each bucket and calculate the final score.
+  final_score = 0.0
+  last_end_timedelta = timedelta(days=0)
+
+  for bucket in SEARCH_BUCKETS:
+    start_date = today - bucket.delta
+    end_date = today - last_end_timedelta
+    last_end_timedelta = bucket.delta
+
+    query = (RepositoryActionCount
+             .select(fn.Sum(RepositoryActionCount.count), fn.Count(RepositoryActionCount.id))
+             .where(RepositoryActionCount.date >= start_date,
+                    RepositoryActionCount.date < end_date,
+                    RepositoryActionCount.repository == repo))
+
+    bucket_tuple = query.tuples()[0]
+    logger.debug('Got bucket tuple %s for bucket %s for repository %s', bucket_tuple, bucket,
+                 repo.id)
+
+    bucket_sum = bucket_tuple[0]
+    bucket_count = bucket_tuple[1]
+    if not bucket_count:
+      continue
+
+    bucket_score = bucket_sum / (bucket_count * 1.0)
+    final_score += bucket_score * bucket.weight
+
+  # Update the existing repo search score row or create a new one.
+  normalized_score = int(final_score * 100.0)
+  try:
+    try:
+      search_score_row = RepositorySearchScore.get(repository=repo)
+      search_score_row.last_updated = datetime.now()
+      search_score_row.score = normalized_score
+      search_score_row.save()
+      return True
+    except RepositorySearchScore.DoesNotExist:
+      RepositorySearchScore.create(repository=repo, score=normalized_score, last_updated=today)
+      return True
+  except IntegrityError:
+    logger.debug('RepositorySearchScore row already existed; skipping')
+    return False

data/model/test/test_repositoryactioncount.py

@@ -0,0 +1,38 @@
+from datetime import date, timedelta
+
+import pytest
+
+from data.database import RepositoryActionCount, RepositorySearchScore
+from data.model.repository import create_repository
+from data.model.repositoryactioncount import update_repository_score, SEARCH_BUCKETS
+from test.fixtures import database_uri, init_db_path, sqlitedb_file
+
+@pytest.mark.parametrize('bucket_sums,expected_score', [
+  ((0, 0, 0, 0), 0),
+
+  ((1, 6, 24, 152), 100),
+  ((2, 6, 24, 152), 101),
+  ((1, 6, 24, 304), 171),
+
+  ((100, 480, 24, 152), 703),
+  ((1, 6, 24, 15200), 7131),
+
+  ((300, 500, 1000, 0), 1733),
+  ((5000, 0, 0, 0), 5434),
+])
+def test_update_repository_score(bucket_sums, expected_score, database_uri):
+  # Create a new repository.
+  repo = create_repository('devtable', 'somenewrepo', None, repo_kind='image')
+
+  # Delete the RAC created in create_repository.
+  RepositoryActionCount.delete().where(RepositoryActionCount.repository == repo).execute()
+
+  # Add RAC rows for each of the buckets.
+  for index, bucket in enumerate(SEARCH_BUCKETS):
+    for day in range(0, bucket.days):
+      RepositoryActionCount.create(repository=repo,
+                                   count=(bucket_sums[index] / bucket.days * 1.0),
+                                   date=date.today() - bucket.delta + timedelta(days=day))
+
+  assert update_repository_score(repo)
+  assert RepositorySearchScore.get(repository=repo).score == expected_score

initdb.py

@@ -854,8 +854,13 @@ def populate_database(minimal=False, with_storage=False):
 
   model.user.create_user_prompt(new_user_4, 'confirm_username')
 
-  while repositoryactioncounter.count_repository_actions():
-    pass
+  while True:
+    to_count = model.repositoryactioncount.find_uncounted_repository()
+    if not to_count:
+      break
+
+    model.repositoryactioncount.count_repository_actions(to_count)
+    model.repositoryactioncount.update_repository_score(to_count)
 
 
 def find_models_missing_data():

workers/repositoryactioncounter.py

@@ -1,62 +1,39 @@
 import logging
 
-from datetime import date, timedelta
-
 from app import app # This is required to initialize the database.
-from data.database import Repository, LogEntry, RepositoryActionCount, db_random_func
+from data import model
 from workers.worker import Worker
 
 POLL_PERIOD_SECONDS = 10
 
 logger = logging.getLogger(__name__)
 
-def count_repository_actions():
-  """ Aggregates repository actions from the LogEntry table and writes them to
-      the RepositoryActionCount table. Returns the number of repositories for
-      which actions were logged. Returns 0 when there is no more work.
-      """
-  try:
-    # Get a random repository to count.
-    today = date.today()
-    yesterday = today - timedelta(days=1)
-    has_yesterday_actions = (RepositoryActionCount
-                             .select(RepositoryActionCount.repository)
-                             .where(RepositoryActionCount.date == yesterday))
-
-    to_count = (Repository
-                .select()
-                .where(~(Repository.id << (has_yesterday_actions)))
-                .order_by(db_random_func()).get())
-
-    logger.debug('Counting: %s', to_count.id)
-
-    actions = (LogEntry
-               .select()
-               .where(LogEntry.repository == to_count,
-                      LogEntry.datetime >= yesterday,
-                      LogEntry.datetime < today)
-               .count())
-
-    # Create the row.
-    try:
-      RepositoryActionCount.create(repository=to_count, date=yesterday, count=actions)
-      return 1
-    except:
-      logger.exception('Exception when writing count')
-  except Repository.DoesNotExist:
-    logger.debug('No further repositories to count')
-
-  return 0
-
-
 class RepositoryActionCountWorker(Worker):
   def __init__(self):
     super(RepositoryActionCountWorker, self).__init__()
     self.add_operation(self._count_repository_actions, POLL_PERIOD_SECONDS)
 
   def _count_repository_actions(self):
-    """ Counts actions for a random repository for the previous day. """
-    count_repository_actions()
+    """ Counts actions and aggregates search scores for a random repository for the
+        previous day. """
+    to_count = model.repositoryactioncount.find_uncounted_repository()
+    if to_count is None:
+      logger.debug('No further repositories to count')
+      return
+
+    logger.debug('Found repository #%s to count', to_count.id)
+    was_counted = model.repositoryactioncount.count_repository_actions(to_count)
+    if not was_counted:
+      logger.debug('Repository #%s was counted by another worker', to_count.id)
+      return
+
+    logger.debug('Updating search score for repository #%s', to_count.id)
+    was_updated = model.repositoryactioncount.update_repository_score(to_count)
+    if not was_updated:
+      logger.debug('Repository #%s had its search score updated by another worker', to_count.id)
+      return
+
+    logger.debug('Repository #%s search score updated', to_count.id)
 
 if __name__ == "__main__":
   worker = RepositoryActionCountWorker()

workers/test/test_repositoryactioncounter.py

@@ -0,0 +1,10 @@
+from data import model, database
+from test.fixtures import app, appconfig, database_uri, init_db_path, sqlitedb_file
+from workers.repositoryactioncounter import RepositoryActionCountWorker
+
+def test_repositoryactioncount(app):
+  database.RepositoryActionCount.delete().execute()
+  database.RepositorySearchScore.delete().execute()
+
+  rac = RepositoryActionCountWorker()
+  rac._count_repository_actions()