swarming: Add server-side implementation for supplemental bot_config

appengine/swarming/handlers_bot.py

Index: appengine/swarming/handlers_bot.py
diff --git a/appengine/swarming/handlers_bot.py b/appengine/swarming/handlers_bot.py
index 06d30e62bf66d0c6195839969569d2282618ac87..bfb52da06778f05dc51966cd85f169ee3bc70ad9 100644
--- a/appengine/swarming/handlers_bot.py
+++ b/appengine/swarming/handlers_bot.py
@@ -240,6 +240,8 @@ class _ProcessResult(object):
   dimensions = None
   # Instance of BotGroupConfig with server-side bot config (from bots.cfg).
   bot_group_cfg = None
+  # Supplemental bot_config to inject into the bot.
+  bot_config = None
   # Bot quarantine message (or None if the bot is not in a quarantine).
   quarantined_msg = None
 
@@ -437,6 +439,20 @@ class BotHandshakeHandler(_BotBaseHandler):
         'dimensions': res.bot_group_cfg.dimensions,
       },
     }
+    if res.bot_group_cfg.bot_config_script:

[Vadim Sh., 2017/02/10 22:08:00]

I've realized there's no mechanism that triggers a reload of config script
changes. The existing mechanism based on 'bot_group_cfg_version' will trigger a
reload if _name_ of bot_config_script changes, not its content.

And even if we extend to include bot_config_script body into
bot_group_cfg_version hash, this mechanism currently reboots bots, which is too
heavyweight for config changes (e.g. we don't want to reboot the entire Chrome
pool when changing configs). 

+      _, script = config.config.get_self_config(
+          'scripts/' + res.bot_group_cfg.bot_config_script,
+          store_last_good=True)
+      logging.info(
+          'Injecting %s: %d bytes',
+          res.bot_group_cfg.bot_config_script,
+          len(script or ''))
+      if not script:
+        # Return an internal error, so that the bot retries slowly.
+        msg = 'Failed to find %s' % res.bot_group_cfg.bot_config_script
+        logging.error(msg)
+        self.abort(503, msg)
+      data['bot_config'] = script
     self.send_response(data)