Parallelize LUCreateInstance

[ganeti-local] / lib / luxi.py

diff --git a/lib/luxi.py b/lib/luxi.py
index 7855434..f04aee7 100644 (file)
--- a/lib/luxi.py
+++ b/lib/luxi.py
@@ -31,18 +31,26 @@ The module is also be used by the master daemon.
 
 import socket
 import collections
-import simplejson
 import time
+import errno
 
-from ganeti import opcodes
+from ganeti import serializer
 from ganeti import constants
 
 
-KEY_REQUEST = 'request'
-KEY_DATA = 'data'
-REQ_SUBMIT = 'submit'
-REQ_ABORT = 'abort'
-REQ_QUERY = 'query'
+KEY_METHOD = 'method'
+KEY_ARGS = 'args'
+KEY_SUCCESS = "success"
+KEY_RESULT = "result"
+
+REQ_SUBMIT_JOB = "SubmitJob"
+REQ_WAIT_FOR_JOB_CHANGE = "WaitForJobChange"
+REQ_CANCEL_JOB = "CancelJob"
+REQ_ARCHIVE_JOB = "ArchiveJob"
+REQ_QUERY_JOBS = "QueryJobs"
+REQ_QUERY_INSTANCES = "QueryInstances"
+REQ_QUERY_NODES = "QueryNodes"
+REQ_QUERY_EXPORTS = "QueryExports"
 
 DEF_CTMO = 10
 DEF_RWTO = 60
@@ -68,22 +76,26 @@ class DecodingError(ProtocolError):
   """Decoding failure on the receiving side"""
 
 
-def SerializeJob(job):
-  """Convert a job description to a string format.
+class RequestError(ProtocolError):
+  """Error on request
+
+  This signifies an error in the request format or request handling,
+  but not (e.g.) an error in starting up an instance.
+
+  Some common conditions that can trigger this exception:
+    - job submission failed because the job data was wrong
+    - query failed because required fields were missing
 
   """
-  return simplejson.dumps(job.__getstate__())
 
 
-def UnserializeJob(data):
-  """Load a job from a string format"""
-  try:
-    new_data = simplejson.loads(data)
-  except Exception, err:
-    raise DecodingError("Error while unserializing: %s" % str(err))
-  job = opcodes.Job()
-  job.__setstate__(new_data)
-  return job
+class NoMasterError(ProtocolError):
+  """The master cannot be reached
+
+  This means that the master daemon is not running or the socket has
+  been removed.
+
+  """
 
 
 class Transport:
@@ -140,9 +152,13 @@ class Transport:
       try:
         self.socket.connect(address)
       except socket.timeout, err:
-        raise TimeoutError("Connection timed out: %s" % str(err))
+        raise TimeoutError("Connect timed out: %s" % str(err))
+      except socket.error, err:
+        if err.args[0] in (errno.ENOENT, errno.ECONNREFUSED):
+          raise NoMasterError((address,))
+        raise
       self.socket.settimeout(self._rwtimeout)
-    except socket.error:
+    except (socket.error, NoMasterError):
       if self.socket is not None:
         self.socket.close()
       self.socket = None
@@ -234,22 +250,65 @@ class Client(object):
       address = constants.MASTER_SOCKET
     self.transport = transport(address, timeouts=timeouts)
 
-  def SendRequest(self, request, data):
+  def CallMethod(self, method, args):
     """Send a generic request and return the response.
 
     """
-    msg = {KEY_REQUEST: request, KEY_DATA: data}
-    result = self.transport.Call(simplejson.dumps(msg))
+    # Build request
+    request = {
+      KEY_METHOD: method,
+      KEY_ARGS: args,
+      }
+
+    # Send request and wait for response
+    result = self.transport.Call(serializer.DumpJson(request, indent=False))
     try:
-      data = simplejson.loads(result)
+      data = serializer.LoadJson(result)
     except Exception, err:
       raise ProtocolError("Error while deserializing response: %s" % str(err))
-    return data
 
-  def SubmitJob(self, job):
-    """Submit a job"""
-    return self.SendRequest(REQ_SUBMIT, SerializeJob(job))
+    # Validate response
+    if (not isinstance(data, dict) or
+        KEY_SUCCESS not in data or
+        KEY_RESULT not in data):
+      raise DecodingError("Invalid response from server: %s" % str(data))
+
+    if not data[KEY_SUCCESS]:
+      # TODO: decide on a standard exception
+      raise RequestError(data[KEY_RESULT])
+
+    return data[KEY_RESULT]
+
+  def SubmitJob(self, ops):
+    ops_state = map(lambda op: op.__getstate__(), ops)
+    return self.CallMethod(REQ_SUBMIT_JOB, ops_state)
+
+  def CancelJob(self, job_id):
+    return self.CallMethod(REQ_CANCEL_JOB, job_id)
+
+  def ArchiveJob(self, job_id):
+    return self.CallMethod(REQ_ARCHIVE_JOB, job_id)
+
+  def WaitForJobChange(self, job_id, fields, prev_job_info, prev_log_serial):
+    timeout = (DEF_RWTO - 1) / 2
+    while True:
+      result = self.CallMethod(REQ_WAIT_FOR_JOB_CHANGE,
+                               (job_id, fields, prev_job_info,
+                                prev_log_serial, timeout))
+      if result != constants.JOB_NOTCHANGED:
+        break
+    return result
+
+  def QueryJobs(self, job_ids, fields):
+    return self.CallMethod(REQ_QUERY_JOBS, (job_ids, fields))
+
+  def QueryInstances(self, names, fields):
+    return self.CallMethod(REQ_QUERY_INSTANCES, (names, fields))
+
+  def QueryNodes(self, names, fields):
+    return self.CallMethod(REQ_QUERY_NODES, (names, fields))
+
+  def QueryExports(self, nodes):
+    return self.CallMethod(REQ_QUERY_EXPORTS, nodes)
 
-  def Query(self, data):
-    """Make a query"""
-    return self.SendRequest(REQ_QUERY, data)
+# TODO: class Server(object)