# HG changeset patch
# User Sverre Rabbelier <srabbelier@gmail.com>
# Date 1240162964 0
# Node ID c29272f640b06a12ee4f9b29f037f0b4bc087ea0
# Parent  46e490606f3d6bdba04d85b79605a90b8a7a9459
Tweak the 'load balancing' algorithm

In order to reduce contention we randomly skipped jobs, but this
caused many jobs to end up stopping early. Now instead we keep on
going until we time out (also increased the chance of doing work).

Patch by: Sverre Rabbelier

diff -r 46e490606f3d -r c29272f640b0 app/soc/cron/job.py
--- a/app/soc/cron/job.py	Sun Apr 19 17:42:27 2009 +0000
+++ b/app/soc/cron/job.py	Sun Apr 19 17:42:44 2009 +0000
@@ -185,5 +185,13 @@
       db.run_in_transaction(self.failJob, job_key)
       return self.ERRORED
 
+  def iterate(self, jobs, retry_jobs):
+    """Trivial iterator that iterates over jobs then retry_jobs
+    """
+
+    for job in jobs:
+      yield job
+    while retry_jobs:
+      yield retry_jobs[0]
 
 handler = Handler()
diff -r 46e490606f3d -r c29272f640b0 app/soc/views/models/cron.py
--- a/app/soc/views/models/cron.py	Sun Apr 19 17:42:27 2009 +0000
+++ b/app/soc/views/models/cron.py	Sun Apr 19 17:42:44 2009 +0000
@@ -92,15 +92,23 @@
       queryGen = lambda: job_logic.getQueryForFields(filter=filter)
       jobs = job_logic.entityIterator(queryGen, batchSize=10)
 
-      for job in jobs:
-        if random.randint(0, 5) > 0:
+      good = True
+      retry_jobs = []
+
+      for job in handler.iterate(jobs, retry_jobs):
+        if random.randint(0, 3) > 0:
+          retry_jobs.append(job)
           continue
+
         job_key = job.key().id()
-        good = handler.handle(job_key)
+        status = handler.handle(job_key)
 
-        if not good:
+        if status is handler.OUT_OF_TIME:
           break
 
+        if status is handler.ERRORED:
+          retry_jobs.append(job)
+
         jobs_completed += 1
 
     response = 'Completed %d jobs in %d priority groups.' % (