Child pages
  • Moving From DAX3 to DAX4
Skip to end of metadata
Go to start of metadata

Changes:

  • moved from camelCase to snake_case for functions
  • job.uses(file, link=Link.INPUT) is now job.uses(file, type=LinkType.INPUT)

  • job.uses(file, type=LinkType.INPUT, transfer=True, register=True) is now job.uses(file, type=LinkType.INPUT, stage_out=True, register_replica=True

  • add_PFN(PFN("uri", "site_name")) is now just add_pfn(*uris, site="local"))

  • add_profile(Profile(namespace, key, value)) is now just add_profile(namespace, key, value)

  • job.uses(....) is now split up into job.add_inputs(*inputs) and job.add_outputs(*outputs, stage_out=True, register_replica=True)  (stage_out and register_replica would apply to all of *outputs, and so this can be called multiple times if different values of stage_out and register_replica must be set for some subset of *outputs)

  • transformation.uses(exec) is now transformation.uses(*execs)

  • obj.metadata(key, value) is now obj.add_metadata(key, value)

  • dag.depends(parent=some_parent, child=some_child) is now dag.add_dependency(parent, *children)

Given what we have in DAX4 (middle column below), we could rewrite it to be something like what we have in the right most column below. Functionality it is the same. At least in this example, it looks a little less cluttered and simpler to read than the middle column while saving a few lines in the process...


DAX3 Black Diamond
#!/usr/bin/env python

from Pegasus.DAX3 import *

# Create a DAX
diamond = ADAG("diamond")

# Add some metadata
diamond.metadata("name", "diamond")
diamond.metadata("createdby", "Gideon Juve")

# Add input file to the DAX-level replica catalog
a = File("f.a")
a.addPFN(PFN("gsiftp://site.com/inputs/f.a","site"))
a.metadata("size", "1024")
diamond.addFile(a)

# Add executables to the DAX-level replica catalog
e_preprocess = Executable(namespace="diamond", name="preprocess", version="4.0", os="linux", arch="x86_64")
e_preprocess.metadata("size", "2048")
e_preprocess.addPFN(PFN("gsiftp://site.com/bin/preprocess","site"))
diamond.addExecutable(e_preprocess)

e_findrange = Executable(namespace="diamond", name="findrange", version="4.0", os="linux", arch="x86_64")
e_findrange.addPFN(PFN("gsiftp://site.com/bin/findrange","site"))
diamond.addExecutable(e_findrange)

e_analyze = Executable(namespace="diamond", name="analyze", version="4.0", os="linux", arch="x86_64")
e_analyze.addPFN(PFN("gsiftp://site.com/bin/analyze","site"))
diamond.addExecutable(e_analyze)

# Add a preprocess job
preprocess = Job(e_preprocess)
preprocess.addProfile(Profile(Namespace.DAGMAN, "RETRY", 0))
preprocess.metadata("time", "60")
b1 = File("f.b1")
b2 = File("f.b2")
preprocess.addArguments("-a preprocess","-T60","-i",a,"-o",b1,b2)
preprocess.uses(a, link=Link.INPUT)
preprocess.uses(b1, link=Link.OUTPUT, transfer=True)
preprocess.uses(b2, link=Link.OUTPUT, transfer=True)
diamond.addJob(preprocess)

# Add left Findrange job
frl = Job(e_findrange)
frl.metadata("time", "60")
c1 = File("f.c1")
frl.addArguments("-a findrange","-T60","-i",b1,"-o",c1)
frl.uses(b1, link=Link.INPUT)
frl.uses(c1, link=Link.OUTPUT, transfer=True)
diamond.addJob(frl)

# Add right Findrange job
frr = Job(e_findrange)
frr.metadata("time", "60")
c2 = File("f.c2")
frr.addArguments("-a findrange","-T60","-i",b2,"-o",c2)
frr.uses(b2, link=Link.INPUT)
frr.uses(c2, link=Link.OUTPUT, transfer=True)
diamond.addJob(frr)

# Add Analyze job
analyze = Job(e_analyze)
analyze.metadata("time", "60")
d = File("f.d")
analyze.addArguments("-a analyze","-T60","-i",c1,c2,"-o",d)
analyze.uses(c1, link=Link.INPUT)
analyze.uses(c2, link=Link.INPUT)
analyze.uses(d, link=Link.OUTPUT, transfer=True, register=True)
diamond.addJob(analyze)

# Add dependencies
diamond.depends(parent=preprocess, child=frl)
diamond.depends(parent=preprocess, child=frr)
diamond.depends(parent=frl, child=analyze)
diamond.depends(parent=frr, child=analyze)

# Write the DAX to stdout
import sys
diamond.writeXML(sys.stdout)

# Write the DAX to a file
f = open("diamond.dax","w")
diamond.writeXML(f)
f.close()
DAX4 Black Diamond
#!/usr/bin/env python

from Pegasus.DAX4 import *

# Create a DAX
diamond = Workflow("diamond")

# Add some metadata
diamond.add_metadata("name", "diamond")
diamond.add_metadata("createdby", "Gideon Juve")

# Add input file to the DAX-level replica catalog
a = File("f.a")
a.add_PFN("gsiftp://site.com/inputs/f.a","site")
a.add_metadata("size", "1024")
diamond.add_file(a)

# Add executables to the DAX-level replica catalog
e_preprocess = Transformation(namespace="diamond", name="preprocess", version="4.0", os="linux", arch="x86_64")
e_preprocess.add_metadata("size", "2048")
e_preprocess.add_PFN("gsiftp://site.com/bin/preprocess","site")
diamond.add_transformation(e_preprocess)

e_findrange = Transformation(namespace="diamond", name="findrange", version="4.0", os="linux", arch="x86_64")
e_findrange.add_PFN("gsiftp://site.com/bin/findrange","site")
diamond.add_transformation(e_findrange)

e_analyze = Transformation(namespace="diamond", name="analyze", version="4.0", os="linux", arch="x86_64")
e_analyze.add_PFN("gsiftp://site.com/bin/analyze","site")
diamond.add_transformation(e_analyze)

# Add a preprocess job
preprocess = Job(e_preprocess)
preprocess.add_profile(Namespace.DAGMAN, "RETRY", 0)
preprocess.add_metadata("time", "60")
b1 = File("f.b1")
b2 = File("f.b2")
preprocess.add_arguments("-a preprocess","-T60","-i",a,"-o",b1,b2)
preprocess.add_inputs(a)
preprocess.add_outputs(b1, b2, stage_out=True, register_replica=False)
diamond.add_job(preprocess)

# Add left Findrange job
frl = Job(e_findrange)
frl.add_metadata("time", "60")
c1 = File("f.c1")
frl.add_arguments("-a findrange","-T60","-i",b1,"-o",c1)
frl.add_inputs(b1)
frl.add_outputs(c1, stage_out=True)
diamond.add_job(frl)

# Add right Findrange job
frr = Job(e_findrange)
frr.add_metadata("time", "60")
c2 = File("f.c2")
frr.add_arguments("-a findrange","-T60","-i",b2,"-o",c2)
frr.add_inputs(b2)
frr.add_outputs(c3, stage_out=True)
diamond.add_job(frr)

# Add Analyze job
analyze = Job(e_analyze)
analyze.add_metadata("time", "60")
d = File("f.d")
analyze.add_arguments("-a analyze","-T60","-i",c1,c2,"-o",d)
analyze.add_inputs(c1, c2)
analyze.add_outputs(d, stage_out=True, register_replica=True)
diamond.add_job(analyze)

# Add dependencies
diamond.add_dependency(preprocess, frl, frr) # Workflow.add_dependency(parent, *children)
diamond.add_dependency(frl, analyze) # Workflow.add_dependency(parent, *children)
diamond.add_dependency(frr, analyze) # Workflow.add_dependency(parent, *children)

# Write the DAX to stdout
import sys
diamond.write_yaml(sys.stdout)

# Write the DAX to a file
f = open("diamond.yml","w")
diamond.write_yaml(f)
f.close()

DAX4 Black Diamond, catalogs included
'''
------- Catalogs -------
Pegasus.DAX4.SiteCatalog
- add_site(name)
- get_site(name)
- has_site(name)
- write_catalog(path)
...

Pegasus.DAX4.ReplicaCatalog
- add_replica(lfn, iri, site, regex=false)
- get_replica(lfn)
- has_replica(lfn)
- remove_replica(lfn)
- write_catalog(path)
...

Pegasus.DAX4.TransformationCatalog
- add_transformation(namespace, name, version, os, arch, ...)
- get_transformation(namespace, name, version, os, arch, ...)
- has_transformation(namespace, name, version, os, arch, ...)
- remove_transformation(namespace, name, version, os, arch, ...)
- write_catalog(path)
...

------- Workflow -------
Pegasus.DAX4.Workflow <------ requires only 'pegasus' (version), 'name', and 'jobs' (jobs and parent-child dependencies)
- add_job(transformation, *args)
- add_dependency(parent, *children)
- add_site_catalog(sc)            # add inline OR iri if string is given
- add_replica_catalog(rc)         # add inline
- add_transformation_catalog(tc)  # add inline
...

'''


#!/usr/bin/env python

from Pegasus.DAX4 import *

# Create a DAX
diamond = Workflow("diamond")

# Add some metadata
diamond.add_metadata("name", "diamond")\
    .add_metadata("createdby", "Gideon Juve")

# Add input file to the replica_catalog
rc = ReplicaCatalog()
a = rc.add_file("f.a", "gsiftp://site.com/inputs/f.a", "site", regex=False)\
        .add_metadata("size", "1024")

# Add executables to the transformation catalog
tc = TransformationCatalog()
e_preprocess = tc.add_transformation(namespace="diamond", name="preprocess", version="4.0", os="linux", arch="x86_64")\
                .add_metadata("size", "2048")\
                .add_site(site="site", uri="gsiftp://site.com/bin/preprocess", type="installed")

e_findrange = tc.add_transformation(namespace="diamond", name="findrange", version="4.0", os="linux", arch="x86_64")\
                .add_site(site="site", uri="gsiftp://site.com/bin/findrange", type="installed")

e_analyze = tc.add_transformation(namespace="diamond", name="analyze", version="4.0", os="linux", arch="x86_64")\
                .add_site(site="site", uri="gsiftp://site.com/bin/analyze", type="installed")

# Add a preprocess job
b1 = File("f.b1")
b2 = File("f.b2")
preprocess = diamond.add_job(e_preprocess, "-a preprocess", "-T60", "-i", a, "-o", b1, b2)\ # ADAG.add_job(exec, *args)
                .add_profile(Namespace.DAGMAN, "RETRY", 0)\
                .add_metadata("time", "60")\
                .add_inputs(a)\
                .add_outputs(b1, b2, stage_out=True, register_replica=False) # job.add_outputs(*lfns, stage_out=True, register_replica=False)

# Add left Findrange job
c1 = File("f.c1")
frl = diamond.add_job(e_findrange, "-a findrange", "-T60", "-i", b1, "-o", c1)\ # ADAG.add_job(exec, *args)
        .add_metadata("time", "60")\
        .add_inputs(b1)\
        .add_outputs(c1, stage_out=True)

# Add right Findrange job
c2 = File("f.c2")
frr = diamond.add_job(e_findrange, "-a findrange", "-T60", "-i", b2, "-o", c2)\ # ADAG.add_job(exec, *args)
        .add_metadata("time", "60")\
        .add_inputs(b2)\
        .add_outputs(c2, stage_out=True)\

# Add Analyze job
d = File("f.d")
analyze = diamond.add_job(e_analyze, *["-a analyze", "-T60", "-i", c1, c2, "-o", d])\ # ADAG.add_job(exec, *args)
            .add_metadata("time", "60")\
            .add_inputs(c1, c2)\
            .add_outputs(d, stage_out=True, register_replica=True)

# Add dependencies
diamond.add_dependency(preprocess, frl, frr)\
    .add_dependency(frl, analyze)\
    .add_dependency(frr, analyze)


If you wanted the catalogs inline, then:
diamond.add_replica_catalog(rc)
diamond.add_transformation_catalog(tc)
diamond.add_site_catalog(sc) # not defined above, but if it was


# Write the DAX to a file
f = open("diamond.yml","w")
diamond.write_yaml(f)
f.close()

'''
# Additionally, we could use contextlib and rewrite like the following:
with ReplicaCatalog("rc.yml") as rc, \
     SiteCatalog("sc.yml") as sc, \
     TransformationCatalog("tc.yml") as tc, \
     Workflow("diamond") as wf:

     # do stuff

# with block ends and all the catalogs + workflow are written to separate files
'''


DAX3 Invokes
from Pegasus.DAX3 import *

# Create a DAX
adag = ADAG("workflow")

# Add invoke
adag.addInvoke(Invoke("start", "/usr/bin/echo hello"))
# adag.invoke("start", "/usr/bin/echo hello")

DAX4 Hooks
from Pegasus.DAX4 import *

# Create a DAX
adag = Workflow("workflow")

# Add shell hook
adag.add_shell_hook("start", "/usr/bin/echo hello")

# Add web hook hook
adag.add_webhook_hook("start", ...)

DAX3 Transformations
from Pegasus.DAX3 import *

# Create a DAX
adag = ADAG("workflow")

# Create some executables
e1 = Executable("executable_1")
e2 = Executable("executable_2")
e3 = Executable("executable_3")

# Create a transformation
t = Transformation(e3)
t.uses(e1)
t.uses(e2)
t.uses(e3)
DAX4 Transformations
from Pegasus.DAX4 import *

# Create a DAX
adag = Workflow("workflow")

# Create some transformations
e1 = Transformation("executable_1")
e2 = Transformation("executable_2")


# Create a transformation that requires other transformations
t = Transformation("executable_3")
t.requires(e1, e2)


Jobs
'''
Trying to be declarative, but then workflow.add(...) might need to be
workflow.jobs(Job(...)) to be consistent. Don't like how the function names
aren't verbs.
'''
job = workflow.add(Job(transformation, arguments))\
    .profiles(namespace, "key", "value")\
    .metadata({"time": "60"})\
    .metadata({"key": "value"})\
    .hooks(ShellHook(args))\
    .inputs(a)\
    .outputs(b1, b2, stage_out=True)

'''
Verbose, explicitly stating operations on job.
The following two assignments are the same.
'''
job = workflow.add_job(transformation, arguments...)\
    .add_profile(namespace, "key", "value")\
    .add_metadata(**{"time": "60", "key": "value"})\
    .add_shell_hook(args)\
    .add_inputs(a)\
    .add_outputs(b1, b2, stage_out=True)

job = workflow.add_job(Job(transformation, args))\
    .add_profile(Profile(namespace, "key", "value"))\
    .add_metadata(*[Metadata("time", "60"), Metadata("key", "value")])\
    .add_shell_hook(args)\
    .add_inputs(a)\
    .add_outputs(b1, b2, stage_out=True)

'''
List/Set semantics.Can't chain as job.<member>.add(something) won't
return back a ref to the job (unless we make it but that would look awkward)
'''
job = workflow.add(Job(transformation, arguments))
# List semantics for (job.args)
job.arguments.append("A") 
job.arguments.extend( ["A", "B"] )
# Set semantics for (job.profiles)
job.profiles.add(namespace, "key", "value")
# Dict semantics for job.{<namespace>,metadata}.*
job.env.key = "value"
job.env["key"] = "value" 
job.env.update( {"key1": "value1", "key2": "value2"} )
job.metadata.add({"time": "60"})
job.metadata.add({"key": "value"})
job.hooks.add(ShellHook(args))
job.inputs.add(a)
job.outputs.add(b1, b2, stage_out=True)

DAX4 Workflow "adds"
#!/usr/bin/env python
from Pegasus.DAX4 import *

# Create a DAX
diamond = Workflow("diamond")

# Add some metadata
diamond.add_metadata("name", "diamond")

# Add input file to the replica_catalog
rc = ReplicaCatalog()
sc = SiteCatalog()
# say that we've added some replicas and sites to the catalogs

tc = TransformationCatalog()
e_preprocess = tc.add_transformation(name="preprocess")\
                .add_site(site="site", uri="gsiftp://site.com/bin/preprocess", type="installed")

e_findrange = tc.add_transformation(name="findrange")\
                .add_site(site="site", uri="gsiftp://site.com/bin/findrange", type="installed")

e_analyze = tc.add_transformation(name="analyze")\
                .add_site(site="site", uri="gsiftp://site.com/bin/analyze", type="installed")
# Add jobs
job1 = diamond.add_job(e_preprocess, args)
job2 = diamond.add_job(e_findrange, args)
job3 = diamond.add_job(e_findrange, args)
job4 = diamond.add_job(e_analyze, args)

# Add dependencies
diamond.add_dependency(job1, job2, job3)\
    .add_dependency(job2, job4)\
    .add_dependency(job3, job4)

diamond.add_replica_catalog(rc)
diamond.add_transformation_catalog(tc)
diamond.add_site_catalog(sc)

# Write the DAX to a file
f = open("diamond.yml","w")
diamond.write_yaml(f)
f.close()
DAX4 Workflow "generic add"
#!/usr/bin/env python
from Pegasus.DAX4 import *

# Create a DAX
diamond = Workflow("diamond")

# Add some metadata
diamond.add({"name": "diamond"})

# Add input file to the replica_catalog
rc = ReplicaCatalog()
sc = SiteCatalog()
# say that we've added some replicas and sites to the catalogs

tc = TransformationCatalog()
e_preprocess = tc.add_transformation(name="preprocess")\
                .add_site(site="site", uri="gsiftp://site.com/bin/preprocess", type="installed")

e_findrange = tc.add_transformation(name="findrange")\
                .add_site(site="site", uri="gsiftp://site.com/bin/findrange", type="installed")

e_analyze = tc.add_transformation(name="analyze")\
                .add_site(site="site", uri="gsiftp://site.com/bin/analyze", type="installed")
# Add jobs
job1 = diamond.add(e_preprocess, args)
job2 = diamond.add(e_findrange, args)
job3 = diamond.add(e_findrange, args)
job4 = diamond.add(e_analyze, args)

# Add dependencies
diamond.add(job1, job2, job3)
diamond.add(job2, job4)
diamond.add(job3, job4)

diamond.add(rc)
diamond.add(tc)
diamond.add(sc)

# Write the DAX to a file
f = open("diamond.yml","w")
diamond.write_yaml(f)
f.close()

'''
Using Workflow.add(*args) as supposed to:
- Workflow.add_job(transformation, *args)
- Workflow.add_dependency(parent, *children)
- Workflow.add_metadata(dict or key, value)
- Workflow.add_transformation_catalog(tc)
- Workflow.add_site_catalog(sc)
- Workflow.add_replica_catalog(rc)

Could work as functools.singledispatch makes the dispatch happen
based only on the first argument, and in our function signatures above,
the first arguments are all unique.

If we go with this route, then chaining things together may not be ideal
as it is not obvious what Workflow.add(...) will return. When we say
Workflow.add_job(transformation, *args), it might be more apparent that
a Job would be returned, and therefore you can chain job related functions
together.
'''
  • No labels

4 Comments

  1. add_PFN should be add_pfn

    Consider changing signature to

    def add_pfn(*pfns, site="local")

    Same for .add_inputs .add_outputs

    .depends(parent=preprocess, child=frl) could be

    .depends(parent, *children)

    1. okay, I agree with those changes,

      although I like add_dependency better.


      Also for add_pfn, is there any reason to have multiple uri's for the same site?


      and the function signatures will look like this:

      Job.add_inputs(*inputs)

      Job.add_outputs(*outputs, stage_out=True, register_replica=False)

      ADAG.add_dependency(parent, *children)

      FileOrExecutable.add_pfn(*uri, site="local")


      1. Actually multiple PFN won't be needed as I think it will violate Primary Key (lfn, pfn, site).

  2. Note:

    • ADAG() can be changed to Workflow()
    • change Executable to Transformation as this is how it is in the schema 
    • change Transformation.uses to Transformation.requires