def precheck(self):
"""
Checks to see if the preconditions for this HappyJob have been met.
If so, returns true, and the HappyJob is executed.
It is expected that this method will be overidden to implement custom
checks in a subclass (use lamdbas instead?)
@return: STATUS_READY if the HappyJob's preconditions are met and the job can be run.
STATUS_WAIT if the job is not ready to be run
STATUS_SKIP if the job has already been run
STATUS_ERROR if we should abort
"""
if (not dfs.exists(self.outputpath)):
logger.debug("precheck(%s): outputpath %s does not exist, ready to run."
% (self, self.outputpath))
return 'ready'
inTSs = [dfs.modtime(file) for file in self.inputpaths]
outTS = dfs.modtime(self.outputpath)
newer = reduce(lambda x,y: x or y, [(inTS>outTS) for inTS in inTSs])
logger.debug("Input timestamps: %s" % inTSs)
logger.debug("Output timestamp: %s" % outTS)
if newer:
logger.debug("At least one input file is newer than outputfile, ready to run.")
dfs.delete(self.outputpath)
return 'ready'
else:
logger.debug("All input files are newer than outputfile, skipping.")
return 'skip'
def testTopoSort(self):
""" Topological sort of the DAG """
logger.info("In TestFlow.testSort ...")
logger.debug("Setting up DAG ...")
a = HappyJobNode()
b = HappyJobNode()
c = HappyJobNode()
d = HappyJobNode()
e = HappyJobNode()
f = HappyJobNode()
g = HappyJobNode()
h = HappyJobNode()
i = HappyJobNode()
j = HappyJobNode()
a.addChild(b)
a.addChild(c)
b.addChild(d)
b.addChild(e)
c.addChild(f)
c.addChild(g)
d.addChild(h)
g.addChild(i)
i.addChild(j)
e.addChild(j)
b.addChild(i)
a.addChild(j)
logger.debug("Testing topological sort ...")
sort = a.sort()
self.assertEqual(sort, [a, b, c, d, e, f, g, h, i, j])
logger.info("DONE.")
def __init__(self, **kwargs):
logger.debug("Creating TripleQueryNode.")
DAG.__init__(self, **kwargs)
self.inputpaths = kwargs.get("inputpaths", None)
self.outputpath = kwargs.get("outputpath", None)
self.force = kwargs.get("force", False)
self.query = kwargs.get("query", None)
self.status = None
def __init__(self, job=NullNode(name="root"), workingDir="tmp", cleanupTemp=False, inputpaths=None, outputpath=None, **kwargs):
logger.debug("Creating Flow() object, workingDir=%s" % workingDir)
self.startNode = job
self.lastNode = job
self.workingDir = workingDir
self.cleanupTemp = cleanupTemp
self.default_inputpaths = inputpaths
self.default_outputpath = outputpath
def dictize(self):
"""
Create a dictionary representation of this DAG. Requires subclasses to
"""
dict = {}
for node in self.sort():
logger.debug("Dictize: id %s has name %s" % (node._id, node.name))
x = node._kwargs()
dict[node._id]={"klass":node.__class__.__name__,
"kwargs": x,
"children":[child._id for child in node.children()]}
return dict
def fire(self):
"""
Runs this node's TripleQuery job. Blocks until completed.
"""
job = TripleQuery(self.query, self.inputpaths, self.outputpath)
try:
job.run()
logger.debug("TripleQuery run. Setting status to done.")
self.status = 'done'
except Exception:
logger.error("Caught exception in TripleQuery. Setting status to fail and deleting output.")
dfs.delete(self.outputpath)
self.status = 'fail'
def __copy__(self):
"""
Jython copy.copy() does not work by default.
"""
logger.debug("Copying Flow() object.")
c = Flow()
c.workingDir = self.workingDir
c.cleanupTemp = self.cleanupTemp
c.default_inputpaths = self.default_inputpaths
c.default_outputpath = self.default_outputpath
c.startNode = self.startNode
c.lastNode = self.lastNode
return c
def testDAG(self):
""" Set up with many relationships """
logger.info("In TestFlow.testDAG ...")
logger.debug("Setting up DAG ...")
a = HappyJobNode()
b = HappyJobNode()
c = HappyJobNode()
d = HappyJobNode()
e = HappyJobNode()
f = HappyJobNode()
g = HappyJobNode()
h = HappyJobNode()
i = HappyJobNode()
a.addChild(b)
a.addChild(c)
b.addChild(c)
d.addChild(f)
e.addChild(f)
c.addChild(g)
f.addChild(g)
h.addChild(g)
g.addChild(i)
logger.debug("Testing parent/child relationships ...")
self.assertEqual(a.parents(), [])
self.assertEqual(a.children(), [b, c])
self.assertEqual(c.parents(), [a, b])
self.assertEqual(f.parents(), [d, e])
self.assertEqual(g.parents(), [c, f, h])
self.assertEqual(g.children(), [i])
logger.debug("Testing node retrieval ...")
nodes0 = set([a,b,c,d,e,f,g,h,i])
nodes1 = a.nodes()
nodes2 = e.nodes()
self.assertEqual(nodes0, nodes1)
self.assertEqual(nodes0, nodes2)
logger.debug("Testing sinks and sources ...")
sinks = a.sinks()
self.assertEqual(sinks, [i])
sources = a.sources()
self.assertEqual(sources, [a, d, e, h])
logger.debug("Testing isAncestorOf() and isDecendentOf() ...")
self.assert_(a.isAncestorOf(b))
self.assert_(a.isAncestorOf(g))
self.assert_(a.isAncestorOf(i))
self.assert_(not a.isAncestorOf(d))
self.assert_(i.isDecendentOf(g))
self.assert_(i.isDecendentOf(a))
self.assert_(i.isDecendentOf(b))
self.assert_(i.isDecendentOf(e))
self.assert_(not f.isDecendentOf(a))
logger.info("DONE.")
def fire(self):
"""
Runs this node's HappyJob. Blocks until completed.
"""
if (self.job):
job = self.job
try:
job.run()
logger.debug("Job run. Setting status to done.")
self.status = 'done'
except Exception:
logger.error("Caught exception. Setting status to fail and deleting output.")
dfs.delete(self.outputpath)
self.status = 'fail'
def postcheck(self):
"""
Checks to see if the postconditions for this HappyJob have been met.
If so, returns true, and this Node's children are fired.
It is expected that this method will be overidden to implement custom
checks in a subclass (use lamdbas instead?)
The default implementation returns the status set by fire():
'ready' if the job completed, 'fail' is the job threw an exception
@return: True if this HappyJob's postonditions are met. Children exec.
False if the children jobs should not be fired.
"""
logger.debug("Postcheck status is %s" % self.status)
return self.status
def run(self, force=False, workingDir=None):
"""
Runs the entire job chain (ie DAG) that contains this node.
"""
logger.debug("Calling HappyJobNode.run(), workingDir=%s" % workingDir)
self.linkNodes(workingDir)
if force:
self.deleteOutFiles(onlytmp=False)
# stack = self.sources()
stack = self.sort()
logger.info("Stack order is: %s" % (", ".join([str(x._id) for x in stack],)))
ok_children = self.sources()
while stack:
node = stack.pop(0)
putChildren = False
if (not node in ok_children):
logger.warn("Branch terminated: node %s not in ok_children list %s." % (node, ok_children))
continue
pre = node.precheck()
if node.force:
logger.info("FORCING %s [%s --> %s] (delete %s first)" % (node, node.inputpaths, node.outputpath, node.outputpath))
dfs.delete(node.outputpath)
node.fire()
elif (pre =='ready'):
logger.info("Running %s [%s --> %s]" % (node, node.inputpaths, node.outputpath))
node.fire()
else:
logger.info("Skipping job %s: already done" % node)
putChildren = True
self.status = 'skip'
post = node.postcheck()
if (post == 'done'):
logger.info("Job %s completed successfully. " % node)
putChildren = True
elif (post == 'fail'):
logger.info("Job %s failed. Not adding children." % node)
if putChildren:
if (node.isSink()):
logger.info("Job %s is a sink, no children." % node)
else:
newChildren = [child for child in node.children() if child not in ok_children]
logger.info("Placing children %s of job %s on stack." % (newChildren, node))
ok_children.extend(newChildren)
def __init__(self, **kwargs):
logger.debug("Creating NullNode.")
self._inputpaths = []
self._outputpath = None
self._parents = []
self._children = []
self.job = NullJob
HappyJobNode.__init__(self, **kwargs)
self._name = kwargs.get("name", None)
self.force = kwargs.get("force", False)
if kwargs.has_key("inputpaths"):
self._inputpaths = kwargs.get("inputpaths")
logger.debug("Using inputpaths=%s" % self._inputpaths)
if len(self.job.inputpaths) > 1:
raise ValueException, "NullNodes can only handle a single inputpath."
self._outputpath = self.inputpaths[0]
self.status = None
def __init__(self, **kwargs):
logger.debug("Creating HappyJobNode.")
jobparam = kwargs.get("job", None)
if (not jobparam):
self.job = NullJob()
elif (type(jobparam)==str):
self.job = eval(jobparam)
else:
self.job = jobparam
DAG.__init__(self, **kwargs)
self.force = kwargs.get("force", False)
if kwargs.has_key("inputpaths"):
self.job.inputpaths = kwargs.get("inputpaths")
if kwargs.has_key("outputpath"):
if (kwargs.get("outputpath") and not type(kwargs.get("outputpath"))==str):
raise ValueError, "HappyJobNode.outputpath only accepts a single file; got %s." % kwargs.get("outputpath")
self.job.outputpath = kwargs.get("outputpath")
self.status = None
def __init__(self, **kwargs):
"""
Create a new DAG node.
@param: name -- the name of the node
@param: parents -- parents of this node
@param: children -- children of this node
@param: _id -- the id of the node (for deserialization only)
"""
logger.debug("Creating DAG: %s" % kwargs)
self._id = kwargs.get("_id", DAG._nextId())
self.name = kwargs.get("name", None)
self._parents = []
self._children = []
if (kwargs.has_key("parents")):
self.addParent(kwargs["parents"])
if (kwargs.has_key("children")):
self.addChild(kwargs["children"])
if not self.name:
self.name = self._auto_name()
def linkNodes(self, workingDir=None):
"""
Assures that every parent/child pair have a matching file in
their inFile / outFile lists. Creates files if necessary.
@param workingDir: the directory to create temp files in.
"""
if workingDir:
logger.info("Linking nodes, using workingDir = %s" % (workingDir))
if dfs.exists(workingDir):
fs = dfs.fileStatus(workingDir)
if not fs.isDir():
raise FlowException, "%s is a file, not a directory." % (workingDir)
else:
logger.info("Creating working directory %s." % (workingDir))
# dfs.mkdir(workingDir)
stack = self.sources()
for source in stack:
if ((not source.inputpaths) or len(source.inputpaths)<1):
raise FlowException, "Source node %s has no inputpaths defined." % source
while stack:
node = stack.pop(0)
if node.outputpath:
logger.trace("linkNodes(): %s has an outputpath '%s'. Using it." % (node, node.outputpath))
filename = node.outputpath
else:
filename = "tmp.%s" % (node.name)
if workingDir:
filename = "%s/%s" % (workingDir, filename)
logger.trace("linkNodes(): Created temp outfile '%s' for %s." % (filename, node))
node.outputpath = filename
for child in node.children():
if ((not child.inputpaths) or
(len(set(node.outputpath) & set(child.inputpaths)) == 0)):
logger.debug("linkNodes(): Linked %s and %s with file '%s'." % (node, child, filename))
child.inputpaths = castList(child.inputpaths) + [filename]
stack.append(child)
logger.debug("%s has inputs %s and outputs %s" % (node, node.inputpaths, node.outputpath))
def testFlow2(self):
logger.info("In TestFlow.testFlow2() ...")
test_flow = Flow2(inputpaths=['/data/graph/latest/crawl'], outputpath='namelist')
(names, types) = test_flow.split()
names.chain(HappyJobNode(name='get_names',
job=FilterExact(filterkey='propname',
filtervalues=['/type/object/name', '/common/topic/alias'],
keyfield='a:guid', mapfields={'value':'name'})))
types.chain(HappyJobNode(name='get_types',
job=FilterExact(filterkey='propname',
filtervalues=['/type/object/type'],
keyfield='b:guid', mapfields={'target':'type'})))
names.chain(HappyJobNode(name='join_name_types',
job=InnerJoin(joinkeys=['a:guid', 'b:guid'], outkey='guid'),
force=True),
join=types)
names.chain(HappyJobNode(name='filter_people',
job=FilterExact(filterkey='type',
filtervalues=['/people/person'],
keyfield='guid', mapfields={'type':'type', 'name':'name'})))
names.chain(HappyJobNode(name='invert_names',
job=AggregateJson(aggkey='name', aggfunc='agg.list("guid")')))
logger.debug("DAG: \n%s\n" % names.startNode.dictize())
names.run(force=False)
def _auto_name(self, job):
"""
Generates a unique name for this node, if one was not provided.
"""
root = job.__class__.__name__
nodes = list(self.lastNode.nodes())
matches = [node.name for node in nodes if node.name.startswith(root)]
logger.debug("Node names: %s" % nodes)
if (len(matches)==0):
return root + '_1'
try:
iter_str = [name.split('_')[-1] for name in matches]
logger.debug("Node iter_str: %s" % iter_str)
iters = [int(i) for i in iter_str]
logger.debug("Node iters: %s" % iter_str)
max_iter = max(iters) + 1
logger.debug("max_iter: %s" % max_iter)
return root + '_' + str(max_iter)
except:
logger.warn("Could not determine iteration: %s " % matches)
return root + '_1'
def chain(self, node=None, join=None):
"""
Add a new node to the chain at lastNode return the modified object
"""
if not node: node = NullJob()
if join:
logger.debug("In Flow.chain(): join=%s" % join)
if (type(join) != type(list())):
join = [join]
for jn in join:
logger.debug("Joining node %s into chain %s." % (jn, node))
jn.lastNode.addChild(node)
logger.debug("Chaining %s to %s." % (self, node))
self.lastNode.addChild(node)
self.lastNode = node
return self
def run(self):
"""
Runs the flow.
"""
logger.debug("Preparing to run Flow.")
sources = self.startNode.sources()
logger.debug("Sources: %s" % sources)
for node in sources:
if not node.inputpaths:
logger.debug("Source %s does not have inputpaths, setting to: %s" % (node, self.default_inputpaths))
node.inputpaths = self.default_inputpaths
sinks = self.startNode.sinks()
logger.debug("Sinks: %s" % sinks)
for node in sinks:
if not node.outputpath:
logger.debug("Source %s does not have outputpath, setting to: %s" % (node, self.default_outputpath))
node.outputpath = self.default_outputpath
logger.debug("Calling HappyJobNode.run(), workingDir = %s" % self.workingDir)
self.startNode.run(force=False, workingDir=self.workingDir)
def _getoutputpath(self):
logger.debug("Called NullNode._getoutputpath() = %s" % self._outputpath)
return self._outputpath
def _setoutputpath(self, outputpath):
logger.debug("Called NullNode._setoutputpath(%s)" % outputpath)
self._inputpaths = [outputpath]
self._outputpath = outputpath
def _getinputpaths(self):
logger.debug("Called NullNode._getinputpaths() = %s" % self._inputpaths)
return self._inputpaths
def _setinputpaths(self, inputpaths):
logger.debug("Called NullNode._setinputpaths(%s)" % inputpaths)
self._inputpaths = uniq(inputpaths)
self._outputpath = inputpaths[0]
