def setup_class(cls):
client = make_logged_in_client('test')
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
# Make home directory
cluster.fs.setuser(cluster.superuser)
if not cluster.fs.exists("/user/test"):
cluster.fs.mkdir("/user/test")
cluster.fs.chown("/user/test", "test", "test")
cluster.fs.setuser("test")
cls.cluster = cluster
cls.client = client
cls.jobsubd = jobsubd
def setup_class(cls):
client = make_logged_in_client('test')
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
# Make home directory
cluster.fs.setuser(cluster.superuser)
if not cluster.fs.exists("/user/test"):
cluster.fs.mkdir("/user/test")
cluster.fs.chown("/user/test", "test", "test")
if not cluster.fs.exists("/tmp"):
cluster.fs.mkdir("/tmp")
cluster.fs.chmod("/tmp", int('777', 8))
cluster.fs.setuser("test")
cls.cluster = cluster
cls.client = client
cls.jobsubd = jobsubd
def test_jobsub_setup_and_samples():
"""
Merely exercises jobsub_setup, and then runs
all the examples.
"""
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
try:
c = make_logged_in_client()
# Create a job, to make sure that it sticks around
response = c.post("/jobsub/new/jar", dict(
name="should_stick_around",
jarfile="foo",
arguments="foo", submit="Save"))
design_id = response.context["saved"]
import jobsub.management.commands.jobsub_setup as jobsub_setup
if not jobsub_setup.Command().has_been_setup():
jobsub_setup.Command().handle()
# Make sure we have three job designs now.
assert_equal(3, JobDesign.objects.filter(name__startswith="Example: ").count())
# Make sure "should_stick_around" is still there
assert_equal(1, JobDesign.objects.filter(name="should_stick_around").count())
# Make sure sample user got created.
assert_equal(1, User.objects.filter(username="******").count())
assert_equal(1, User.objects.filter(username="******").count())
# And now submit and run the samples
# pi Example
# Irritatingly, /user/test needs to exist first
setup_cluster_fs(cluster)
id = JobDesign.objects.get(name__contains="Example: Pi").id
response = c.get("/jobsub/submit/%d" % id)
assert_true("Iterations per mapper" in response.content)
assert_true("Num of mappers" in response.content)
response = c.post("/jobsub/submit/%d" % id, dict(
iterations_per_mapper=10,
num_of_mappers=1))
response = watch_till_complete(c, parse_out_id(response))
assert_true("Estimated value of Pi is" in response.context["job_data"].stdout_tail)
assert_true("bin/hadoop returned 0" in response.content)
# Wordcount example
id = JobDesign.objects.get(name__contains="Example: Streaming Wordcount").id
response = c.get("/jobsub/submit/%d" % id)
response = c.post("/jobsub/submit/%d" % id, dict(
output="/user/test/jobsub-streaming-test"))
response = watch_till_complete(c, parse_out_id(response))
assert_true("streaming.StreamJob: Job complete:" in response.context["job_data"].stderr_tail)
assert_true(cluster.fs.exists("/user/test/jobsub-streaming-test/part-00000"))
# Not running sleep example, since it adds little.
finally:
jobsubd.exit()
cluster.shutdown()
def test_job_submission():
JARNAME = posixpath.basename(hadoop.conf.HADOOP_EXAMPLES_JAR.get())
c = make_logged_in_client()
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
# Remember the number of pending jobs beforehand
n_pending = c.get("/jobsub/status_bar/").context["pending_count"]
try:
# Create a job
response = c.post("/jobsub/new/jar", dict(
name="wordcount",
jarfile="/user/test/%s" % JARNAME,
arguments="wordcount $input $output", submit="Save"))
design_id = response.context["saved"]
# Submission should get a parameterization form
response = c.get("/jobsub/submit/%d" % design_id)
assert_true("<form " in response.content)
# Create home dir
setup_cluster_fs(cluster)
# Prepare sample data
f = cluster.fs.open("/user/test/input", "w")
f.write("alpha beta gamma\nepsilon zeta theta\nalpha beta\n")
f.close()
# We also have to upload the jar file
src = file(hadoop.conf.HADOOP_EXAMPLES_JAR.get())
try:
dst = cluster.fs.open("/user/test/%s" % JARNAME, "w")
try:
shutil.copyfileobj(src, dst)
finally:
dst.close()
finally:
src.close()
# Status_bar should be at original
assert_equal(n_pending, c.get("/jobsub/status_bar/").context["pending_count"])
# Let's parameterize and submit
INPUT, OUTPUT = "/user/test/input", "/user/test/output"
response = c.post("/jobsub/submit/%d" % design_id,
dict(input=INPUT, output=OUTPUT))
watch_id = parse_out_id(response)
# Status bar at original + 1
assert_equal(n_pending + 1, c.get("/jobsub/status_bar/").context["pending_count"])
# Let's take a look
response = watch_till_complete(c, watch_id)
assert_equal(1, len(response.context["job_data"].hadoop_job_ids),
"Should have launched and captured exactly one Hadoop job")
submission = Submission.objects.get(id=watch_id)
assert_equal(["wordcount", INPUT, OUTPUT],
submission.submission_plan.steps[1].bin_hadoop_step.arguments[2:])
hadoop_job_id = response.context["job_data"].hadoop_job_ids[0]
# Status bar back to original
assert_equal(n_pending, c.get("/jobsub/status_bar/").context["pending_count"])
# Make sure the counts are right:
lines = cluster.fs.open("/user/test/output/part-r-00000").read().splitlines()
counts = {}
for line in lines:
word, count = line.split("\t", 2)
count = int(count)
counts[word] = count
assert_equal(dict(alpha=2, beta=2, gamma=1, epsilon=1, zeta=1, theta=1), counts)
# And check that the output file has correct permissions.
assert_equal("test", cluster.fs.stats("/user/test/output/part-r-00000")["user"],
"Wrong username for job output.")
assert_equal("test", cluster.fs.stats("/user/test/output/part-r-00000")["group"],
"Wrong groupname for job output.")
# Just to be sure it really happened, check the Job struct
# There's no way to get just one job (eek!)...
job_map = dict([ (x.jobID.asString, x) for x in cluster.jt.completed_jobs().jobs ])
this_job = job_map[hadoop_job_id]
# Check username and group
assert_equal("test", this_job.profile.user)
# Let's kill the temporary directory, and make sure watch
# output still works. We do file deletion very explicitly,
# because tests that might mistakenly delete your home directory
# tend to cause unhappiness.
server_id = Submission.objects.get(id=watch_id).submission_handle.id
tmp_dir = ServerSubmissionState.objects.get(id=server_id).tmp_dir
for filename in ("jobs", "stderr", "stdout", os.path.join("work", "tmp.jar")):
os.remove(os.path.join(tmp_dir, filename))
os.rmdir(os.path.join(tmp_dir, "work"))
os.rmdir(tmp_dir)
response = c.get("/jobsub/watch/%d" % watch_id)
assert_true("No longer available" in response.content)
finally:
cluster.shutdown()
jobsubd.exit()
def test_jobsub_setup_and_samples():
"""
Merely exercises jobsub_setup, and then runs
all the examples.
"""
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
try:
c = make_logged_in_client()
# Create a job, to make sure that it sticks around
response = c.post("/jobsub/new/jar", dict(
name="should_stick_around",
jarfile="foo",
arguments="foo", submit="Save"))
design_id = response.context["saved"]
import jobsub.management.commands.jobsub_setup as jobsub_setup
if not jobsub_setup.Command().has_been_setup():
jobsub_setup.Command().handle()
# Make sure we have three job designs now.
assert_equal(3, JobDesign.objects.filter(name__startswith="Example: ").count())
# Make sure "should_stick_around" is still there
assert_equal(1, JobDesign.objects.filter(name="should_stick_around").count())
# Make sure sample user got created.
assert_equal(1, User.objects.filter(username="******").count())
assert_equal(1, User.objects.filter(username="******").count())
# And now submit and run the samples
# pi Example
# Irritatingly, /user/test needs to exist first
setup_cluster_fs(cluster)
id = JobDesign.objects.get(name__contains="Example: Pi").id
response = c.get("/jobsub/submit/%d" % id)
assert_true("Iterations per mapper" in response.content)
assert_true("Num of mappers" in response.content)
response = c.post("/jobsub/submit/%d" % id, dict(
iterations_per_mapper=10,
num_of_mappers=1))
response = watch_till_complete(c, parse_out_id(response))
assert_true("Estimated value of Pi is" in response.context["job_data"].stdout_tail)
assert_true("bin/hadoop returned 0" in response.content)
# Wordcount example
id = JobDesign.objects.get(name__contains="Example: Streaming Wordcount").id
response = c.get("/jobsub/submit/%d" % id)
response = c.post("/jobsub/submit/%d" % id, dict(
output="/user/test/jobsub-streaming-test"))
response = watch_till_complete(c, parse_out_id(response))
assert_true("streaming.StreamJob: Job complete:" in response.context["job_data"].stderr_tail)
assert_true(cluster.fs.exists("/user/test/jobsub-streaming-test/part-00000"))
# Not running sleep example, since it adds little.
finally:
jobsubd.exit()
cluster.shutdown()
def test_job_submission():
JARNAME = posixpath.basename(hadoop.conf.HADOOP_EXAMPLES_JAR.get())
c = make_logged_in_client()
cluster = mini_cluster.shared_cluster(conf=True)
jobsubd = in_process_jobsubd(cluster.config_dir)
# Remember the number of pending jobs beforehand
n_pending = c.get("/jobsub/status_bar/").context["pending_count"]
try:
# Create a job
response = c.post("/jobsub/new/jar", dict(
name="wordcount",
jarfile="/user/test/%s" % JARNAME,
arguments="wordcount $input $output", submit="Save"))
design_id = response.context["saved"]
# Submission should get a parameterization form
response = c.get("/jobsub/submit/%d" % design_id)
assert_true("<form " in response.content)
# Create home dir
setup_cluster_fs(cluster)
# Prepare sample data
f = cluster.fs.open("/user/test/input", "w")
f.write("alpha beta gamma\nepsilon zeta theta\nalpha beta\n")
f.close()
# We also have to upload the jar file
src = file(hadoop.conf.HADOOP_EXAMPLES_JAR.get())
try:
dst = cluster.fs.open("/user/test/%s" % JARNAME, "w")
try:
shutil.copyfileobj(src, dst)
finally:
dst.close()
finally:
src.close()
# Status_bar should be at original
assert_equal(n_pending, c.get("/jobsub/status_bar/").context["pending_count"])
# Let's parameterize and submit
INPUT, OUTPUT = "/user/test/input", "/user/test/output"
response = c.post("/jobsub/submit/%d" % design_id,
dict(input=INPUT, output=OUTPUT))
watch_id = parse_out_id(response)
# Status bar at original + 1
assert_equal(n_pending + 1, c.get("/jobsub/status_bar/").context["pending_count"])
# Let's take a look
response = watch_till_complete(c, watch_id)
assert_equal(1, len(response.context["job_data"].hadoop_job_ids),
"Should have launched and captured exactly one Hadoop job")
submission = Submission.objects.get(id=watch_id)
assert_equal(["wordcount", INPUT, OUTPUT],
submission.submission_plan.steps[1].bin_hadoop_step.arguments[2:])
hadoop_job_id = response.context["job_data"].hadoop_job_ids[0]
# Status bar back to original
assert_equal(n_pending, c.get("/jobsub/status_bar/").context["pending_count"])
# Make sure the counts are right:
lines = cluster.fs.open("/user/test/output/part-r-00000").read().splitlines()
counts = {}
for line in lines:
word, count = line.split("\t", 2)
count = int(count)
counts[word] = count
assert_equal(dict(alpha=2, beta=2, gamma=1, epsilon=1, zeta=1, theta=1), counts)
# And check that the output file has correct permissions.
assert_equal("test", cluster.fs.stats("/user/test/output/part-r-00000")["user"],
"Wrong username for job output.")
assert_equal("test", cluster.fs.stats("/user/test/output/part-r-00000")["group"],
"Wrong groupname for job output.")
# Just to be sure it really happened, check the Job struct
# There's no way to get just one job (eek!)...
job_map = dict([ (x.jobID.asString, x) for x in cluster.jt.completed_jobs().jobs ])
this_job = job_map[hadoop_job_id]
# Check username and group
assert_equal("test", this_job.profile.user)
# Let's kill the temporary directory, and make sure watch
# output still works. We do file deletion very explicitly,
# because tests that might mistakenly delete your home directory
# tend to cause unhappiness.
server_id = Submission.objects.get(id=watch_id).submission_handle.id
tmp_dir = ServerSubmissionState.objects.get(id=server_id).tmp_dir
for filename in ("jobs", "stderr", "stdout", os.path.join("work", "tmp.jar")):
os.remove(os.path.join(tmp_dir, filename))
os.rmdir(os.path.join(tmp_dir, "work"))
os.rmdir(tmp_dir)
response = c.get("/jobsub/watch/%d" % watch_id)
assert_true("No longer available" in response.content)
finally:
cluster.shutdown()
jobsubd.exit()
