Upload basic info about completed swarming tasks to event_mon

appengine/swarming/server/task_scheduler.py

Index: appengine/swarming/server/task_scheduler.py
diff --git a/appengine/swarming/server/task_scheduler.py b/appengine/swarming/server/task_scheduler.py
index a4366492c8fcf7e5eca6cb8992f8ca977698587b..e00515f9fc8ea7faf2141f07c8287661df49dd2f 100644
--- a/appengine/swarming/server/task_scheduler.py
+++ b/appengine/swarming/server/task_scheduler.py
@@ -19,6 +19,7 @@ from components import datastore_utils
 from components import pubsub
 from components import utils
 
+import event_mon_metrics
 import ts_mon_metrics
 
 from server import acl
@@ -812,6 +813,7 @@ def bot_update_task(
     return None
   _update_stats(run_result, bot_id, request, task_completed)
   if task_completed:
+    event_mon_metrics.send_task_event(smry)

[M-A Ruel, 2016/11/28 16:56:32]

I'm concerned in adding more synchronous I/O in the hot path. I was concerned
about ts_mon_metrics too and it did cause a fair amount of failure once deployed
to prod.

I'm fine with testing on prod and looking at the failure rate.

[Paweł Hajdan Jr., 2016/11/29 14:37:33]

I'm also concerned. How about proactively requesting async interfaces for both
monitoring pipelines?

[M-A Ruel, 2016/11/29 14:48:40]

As long as it is quick to disable the functionality by adding an early return in
send_task_event(), it's not worth overdesigning it.

     ts_mon_metrics.update_jobs_completed_metrics(smry)
   return run_result.state