Changeset ca489e8

Timestamp:

Jul 27, 2009 9:41:25 AM (15 years ago)

Author:

Ted Faber <faber@…>

Branches:

axis_example, compt_changes, info-ops, master, version-1.30, version-2.00, version-3.01, version-3.02

Children:

8f32dc0

Parents:

65f3f29

Message:

Add override fedids that can access all experiments. These are admins,
basically.

Also add a force parameter to terminate requests, to force the termination (and
really deletion of internal state) for experiments in odd states or with
internal misconfigurations.

Location:

fedd

Files:

• fedd_client.py (modified) (3 diffs)
• federation/experiment_control.py (modified) (6 diffs)
• wsdl/fedd_types.xsd (modified) (1 diff)

fedd/fedd_client.py

@@ -157 +157 @@
                 help="data to extract")
 
+class fedd_terminate_opts(fedd_exp_data_opts):
+    def __init__(self):
+        fedd_exp_data_opts.__init__(self)
+        self.add_option("--force", dest="force",
+                action="store_true", default=False,
+                help="Force termination if experiment is in strange state")
+
 class fedd_multi_exp_data_opts(fedd_client_opts):
     def __init__(self):
@@ -1024 +1031 @@
         """
         # Process the options using the customized option parser defined above
-        parser = fedd_exp_data_opts()
+        parser = fedd_terminate_opts()
 
         (opts, args) = parser.parse_args()
@@ -1061 +1068 @@
                     "--experiment_name");
 
-        req = { 'experiment': exp_id }
+        req = { 'experiment': exp_id, 'force': opts.force }
 
         try:

fedd/federation/experiment_control.py

@@ -252 +252 @@
         # NB for internal master/slave ops, not experiment setup
         self.ssh_type = config.get("experiment_control", "sshkeytype", "rsa")
+
+        self.overrides = set([])
+        ovr = config.get('experiment_control', 'overrides')
+        if ovr:
+            for o in ovr.split(","):
+                o = o.strip()
+                if o.startswith('fedid:'): o = o[len('fedid:'):]
+                self.overrides.add(fedid(hexstr=o))
+
         self.state = { }
         self.state_lock = Lock()
@@ -408 +417 @@
                             if f.has_key('fedid') ]:
                     self.auth.set_attribute(self.state[k]['owner'], eid)
+                    # allow overrides to control experiments as well
+                    for o in self.overrides:
+                        self.auth.set_attribute(o, eid)
             except KeyError, e:
                 self.log.warning("[read_state]: State ownership or identity " +\
@@ -2009 +2021 @@
         self.auth.set_attribute(fid, expid)
         self.auth.set_attribute(expid, expid)
+        # Override fedids can manipulate state as well
+        for o in self.overrides:
+            self.auth.set_attribute(o, expid)
 
         # Create a logger that logs to the experiment's state object as well as
@@ -2270 +2285 @@
             raise service_error(service_error.req,
                     "Bad request format (no TerminateRequestBody)")
+        force = req.get('force', False)
         exp = req.get('experiment', None)
         if exp:
@@ -2298 +2314 @@
             if status:
                 if status == 'starting':
-                    self.state_lock.release()
-                    raise service_error(service_error.partial, 
-                            'Experiment still being created')
+                    if not force:
+                        self.state_lock.release()
+                        raise service_error(service_error.partial, 
+                                'Experiment still being created')
+                    else:
+                        self.log.warning('Experiment in starting state ' + \
+                                'being terminated by admin.')
             else:
                 # No status??? trouble
@@ -2361 +2381 @@
 
             # release the allocations (failed experiments have done this
-            # already)
-            if status != 'failed':
+            # already, and starting experiments may be in odd states, so we
+            # ignore errors releasing those allocations
+            try: 
                 for tb in tbparams.keys():
                     self.release_access(tb, tbparams[tb]['aid'])
+            except service_error, e:
+                if status != 'failed' and not force:
+                    raise e
 
             # Remove the terminated experiment

fedd/wsdl/fedd_types.xsd

@@ -595 +595 @@
     <xsd:sequence>
       <xsd:element name="experiment" type="tns:IDType"/>
+      <xsd:element name="force" type="xsd:boolean" minOccurs="0" maxOccurs="1"/>
     </xsd:sequence>
   </xsd:complexType>