From e07ab5f8c1f6fdcc3e715f1e6630e9cc6631ccd2 Mon Sep 17 00:00:00 2001
From: Rayson Zhu <yuxzhu@redhat.com>
Date: Aug 16 2016 13:32:04 +0000
Subject: optimize table joins for faster counting performance


Tests shows that changing inner joins to left outer joins on non-null fields will significantly improve the counting performance on PostgreSQL server 9.X.

Only large tables (task, build, and rpminfo) are involved in this commit.

---

diff --git a/hub/kojihub.py b/hub/kojihub.py
index c88946c..9bc17cd 100644
--- a/hub/kojihub.py
+++ b/hub/kojihub.py
@@ -3526,7 +3526,7 @@ def list_rpms(buildID=None, buildrootID=None, imageID=None, componentBuildrootID
               ('rpminfo.metadata_only', 'metadata_only'),
               ('rpminfo.extra', 'extra'),
              ]
-    joins = ['external_repo ON rpminfo.external_repo_id = external_repo.id']
+    joins = ['LEFT JOIN external_repo ON rpminfo.external_repo_id = external_repo.id']
     clauses = []
 
     if buildID != None:
@@ -9047,10 +9047,10 @@ class RootExports(object):
                   ('users.id', 'owner_id'), ('users.name', 'owner_name')]
 
         tables = ['build']
-        joins = ['events ON build.create_event = events.id',
-                 'package ON build.pkg_id = package.id',
-                 'volume ON build.volume_id = volume.id',
-                 'users ON build.owner = users.id']
+        joins = ['LEFT JOIN events ON build.create_event = events.id',
+                 'LEFT JOIN package ON build.pkg_id = package.id',
+                 'LEFT JOIN volume ON build.volume_id = volume.id',
+                 'LEFT JOIN users ON build.owner = users.id']
         clauses = []
         if packageID != None:
             clauses.append('package.id = %(packageID)i')
@@ -9763,7 +9763,7 @@ class RootExports(object):
             queryOpts = {}
 
         tables = ['task']
-        joins = ['users ON task.owner = users.id']
+        joins = ['LEFT JOIN users ON task.owner = users.id']
         flist = Task.fields + (
                     ('task.request', 'request'),
                     ('task.result', 'result'),