[Planetlab-devel] initscripts

[Faiyaz Ahmed]

Hi Amine and Thierry,

You're right in that the slice attribute name for initscripts changed in 
the API and NM was not updated.  I've updated the key used by nm to use 
the correct name.

As for using names rather than ids, the change in NM would clearly be 
simple.  But, I believe things may break in the API.

Tony, do you have any thoughts on what the change might entail?


Faiyaz


Modified:
    NodeManager/trunk/sm.py
Log:
s/plc_initscript_id/initscript
The attribute name used in the API changed and nm was never updated. 
Thanks Amine & Thierry.

Modified: NodeManager/trunk/sm.py
===================================================================
--- NodeManager/trunk/sm.py	2007-12-20 21:45:00 UTC (rev 7583)
+++ NodeManager/trunk/sm.py	2007-12-20 21:59:33 UTC (rev 7584)
@@ -90,10 +90,11 @@
  ### Emulab-specific hack ends here


-    logger.verbose ('dealing with initscripts')
-    initscripts_by_id = {}
+    # Take intscripts (global) returned by API, make dict
+    initscripts = {}
      for is_rec in data['initscripts']:
-        initscripts_by_id[str(is_rec['initscript_id'])] = is_rec['script']
+        logger.verbose("initscript: %s" % is_rec['name'])
+        initscripts[str(is_rec['initscript_id'])] = is_rec['script']

      for sliver in data['slivers']:
          logger.verbose("sm:GetSlivers in slivers loop")
@@ -116,9 +117,9 @@
              rec['type'] = 'delegate'

          rec.setdefault('vref', attr_dict.get('vref', 'default'))
-        is_id = attr_dict.get('plc_initscript_id')
-        if is_id is not None and is_id in initscripts_by_id:
-            rec['initscript'] = initscripts_by_id[is_id]
+        is_id = attr_dict.get('initscript')
+        if is_id is not None and is_id in initscripts:
+            rec['initscript'] = initscripts[is_id]
          else:
              rec['initscript'] = ''
          rec.setdefault('delegations', attr_dict.get("delegations", []))


Thierry Parmentelat wrote:
> Hi Fayaiz, and all
> 
> We're still having problems with this initscript thing
> 
> Here is the situation, as reported by Amine
> ==
> I start with a new Myplc based on the last build:
> http://build.one-lab.org/planetlab/2007.12.20--planetlab-f7-linux32/
> that contains nodemanager from the svn trunk
> 
> I associated an Initscript to a slice(with the id returned by the db 
> when I added the initscript)
> then locally in the node I restarted the NM , but nothing happens.
> 
> My understanding reading the code is that, when the sm.py calls 
> GetSlivers(),
> the nodemanager looks for a slice attribute named "plc_initscript_id", 
> while we naively use the predefined one simply named 'initscript', and 
> so our db contains something like this
> {'instantiation': u'plc-instantiated', 'name': u'Slice_Name', 
> 'slice_id': *, 'keys': [], 'expires': 1199363409L, 'attributes': 
> [{'name': u'initscript', 'value': u'*'}
>                                                                                                                                                                                                       ****
> So based on this part of the sm.py code:
> ******************
> for sliver in data['slivers']:
>    logger.verbose("sm:GetSlivers in slivers loop")
>   .....
>   ........
>   .........
> 
>   rec.setdefault('vref', attr_dict.get('vref', 'default'))
>   is_id = attr_dict.get('plc_initscript_id')
>   if is_id is not None and is_id in initscripts_by_id:
>        rec['initscript'] = initscripts_by_id[is_id]
>   else:
>    rec['initscript'] = '' "
> ************
> the rec['initscript'] that gets passed to the sliver_vs.py class,
> will be empty..
> 
> ==
> 
> I haven't run the code myself, but if the above is correct, we would 
> advocate to either
> (a) change the predefined slice attribute name to what is expected by 
> the nm so that the system works 
> (b) or change the node manager code to use the predefined name instead
> I guess there are implications in terms of legacy that need be taken 
> into account.
> In any case, it would of course make sense to add the feature that we've 
> discussed already, since an initscript name would probably be a 
> preferred method for slices to refer to initscripts. To this respect, we 
> could simply check whether the value attached to the slice attribute is 
> an integer or not, and react accordingly. 
> 
> We'd appreciate to hear your comments on this matter
> 
> Thanks - Thierry
> ==
> PS. fyi - we've started to add a unit test about this in our system test 
> suite.
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Devel mailing list
> Devel at lists.planet-lab.org
> https://lists.planet-lab.org/mailman/listinfo/devel