[Findit] Strip platform from step_name before triggering try job.

appengine/findit/waterfall/trigger_swarming_task_pipeline.py

 # Copyright 2016 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.
 
 import copy
 from datetime import datetime
 import logging
 import time
 
 from google.appengine.ext import ndb
@@ skipping 36 matching lines @@
   ]
   new_request.extra_args.append('--gtest_filter=%s' % ':'.join(tests))
 
   # Remove the env setting for sharding.
   sharding_settings = ['GTEST_SHARD_INDEX', 'GTEST_TOTAL_SHARDS']
   new_request.env = [
       e for e in new_request.env if e['key'] not in sharding_settings
   ]
 
   # Reset tags for searching and monitoring.
+  ref_name = swarming_util.GetTagValue(new_request.tags, 'name') or step_name

[stgao, 2016/03/10 23:50:28]

Should get from the `ref_request` instead, right?

[stgao, 2016/03/10 23:50:28]

And should not default to the step_name.

[stgao, 2016/03/10 23:50:28]

I'm wondering why don't we get the name from the check_first_failure
stage/pipeline. Didn't we query all tasks for all Swarming steps earlier before
triggering any Swarming task?

Is there a reason we do it here instead of an earlier stage?

[chanli, 2016/03/11 00:49:41]

No, this is the referred task, the tag is 'name', 'ref_name' is the tag I added
to the findit tasks.

[chanli, 2016/03/11 00:49:41]

Done.

[chanli, 2016/03/11 00:49:41]

If we get the name when we list swarming tasks in detect_first_failure_pipeline,
we need to either carry it along the pipeline or save it somewhere and retrieve
it later...

While we are already parsing the referred request here, I think the name can be
treated as a part of request (in tags) anyway.

[stgao, 2016/03/11 01:03:20]

My thought is that ref_request is the original data (source of truth), while
new_request is not the source of truth. And we clear new_request.tags after this
line.

[chanli, 2016/03/11 03:47:00]

But before you clear new_request.tags, the tags in it are deepcopied from
ref_request. I don't see any problems here. If you still have doubt, I can
change it to

ref_name = swarming_util.GetTagValue(new_request.tags, 'name') or step_name

Although they are the same to me.

[stgao, 2016/03/11 06:42:58]

Using either new_request.tags or ref_request.tags has the same effect from the
perspective of program.
However, from the perspective of coding style, I believe using ref_request is
the right way to go.

   new_request.tags = []
   new_request.tags.append('purpose:deflake')
   new_request.tags.append('ref_master:%s' % master_name)
   new_request.tags.append('ref_buildername:%s' % builder_name)
   new_request.tags.append('ref_buildnumber:%s' % build_number)
   new_request.tags.append('ref_stepname:%s' % step_name)
   new_request.tags.append('ref_task_id:%s' % ref_task_id)
+  new_request.tags.append('ref_name:%s' % ref_name)
 
   return new_request
 
 
 @ndb.transactional
 def _NeedANewSwarmingTask(master_name, builder_name, build_number, step_name):
   swarming_task = WfSwarmingTask.Get(
       master_name, builder_name, build_number, step_name)
 
   if not swarming_task:
@@ skipping 79 matching lines @@
     # Save the task id.
     swarming_task = WfSwarmingTask.Get(
         master_name, builder_name, build_number, step_name)
     swarming_task.task_id = task_id
     swarming_task.parameters['tests'] = tests
     swarming_task.parameters['iterations_to_rerun'] = iterations_to_rerun
     swarming_task.put()
 
     logging.info('A Swarming task was triggered:%s', task_id)
     return task_id