def wait_for_data(wait_for_config, options):
"""Wait for data before kicking off hive jobs"""
# Step 1 - read meta data.
hive_mysql_connector.configure(options.hive_masternode,
options.ssh_keyfile)
if options.hive_init:
hive_mysql_connector.run_hive_init()
# Step 2 - wait for all the data partitions are available
boto_util.initialize_creds_from_file()
s3conn = boto.connect_s3()
s3bucket = s3conn.get_bucket('ka-mapreduce')
max_wait = datetime.timedelta(hours=options.max_wait)
start = datetime.datetime.now()
for d in wait_for_config:
table = d['table_name']
table_location = hive_mysql_connector.get_table_location(table)
for p in d['partitions']:
partition_location = table_location + '/' + p
#TODO(yunfang): abstract the following to wait_for_partition
# for boto_util
while True:
if partition_available(s3bucket, partition_location):
g_logger.info("%s is available" % (partition_location))
break
if (datetime.datetime.now() - start) > max_wait:
# Wait for a long time already. Can't wait any more
g_logger.fatal("Wait for too long. "
"Data is still not available."
"Exiting...")
sys.exit(1)
# Wait for a minute to check again
g_logger.info("Waiting for %s to be available... " % (
partition_location))
time.sleep(60)
def wait_for_data(wait_for_config, options):
"""Wait for data before kicking off hive jobs"""
# Step 1 - read meta data.
hive_mysql_connector.configure(options.hive_masternode,
options.ssh_keyfile)
if options.hive_init:
hive_mysql_connector.run_hive_init()
# Step 2 - wait for all the data partitions are available
boto_util.initialize_creds_from_file()
s3conn = boto.connect_s3()
s3bucket = s3conn.get_bucket('ka-mapreduce')
max_wait = datetime.timedelta(hours=options.max_wait)
start = datetime.datetime.now()
for d in wait_for_config:
table = d['table_name']
table_location = hive_mysql_connector.get_table_location(table)
for p in d['partitions']:
partition_location = table_location + '/' + p
#TODO(yunfang): abstract the following to wait_for_partition
# for boto_util
while True:
if partition_available(s3bucket, partition_location):
g_logger.info("%s is available" % (partition_location))
break
if (datetime.datetime.now() - start) > max_wait:
# Wait for a long time already. Can't wait any more
g_logger.fatal("Wait for too long. "
"Data is still not available."
"Exiting...")
sys.exit(1)
# Wait for a minute to check again
g_logger.info("Waiting for %s to be available... " % (
partition_location))
time.sleep(60)
def main(table_location, score_type, options):
boto_util.initialize_creds_from_file()
# TODO(benkomalo): factor some of this boilerplate out to boto_util
# Open our input connections.
s3conn = boto.connect_s3()
bucket = s3conn.get_bucket('ka-mapreduce')
s3keys = bucket.list(
prefix=table_location[len('s3://ka-mapreduce/'):] + '/')
# Mapping of video key to the info on the other videos which best match
# that video
# vid_key -> VideoInfo
video_infos = {}
# Note: a table's data may be broken down into multiple files on disk.
delimiter = '\01'
lines_read = 0
for key in s3keys:
if key.name.endswith('_$.folder$'):
# S3 meta data - not useful.
continue
contents = key.get_contents_as_string()
for line in contents.split('\n'):
if not line:
# EOF
break
lines_read += 1
parts = line.rstrip().split(delimiter)
if len(parts) != 3:
# TODO(benkomalo): error handling
continue
vid1_key, vid2_key, score = parts
try:
score = float(score)
except ValueError:
# Some of the values were invalid - deal with it.
# TODO(benkomalo): error handling.
continue
add_score(video_infos, vid1_key, vid2_key, score)
if lines_read % 1000 == 0:
print "Read %s lines..." % lines_read
total_pairs = sum([len(info.best_matches)
for info in video_infos.values()])
print "\nSummary of collected data:"
print "\tScore type: [%s]" % score_type
print ("\tDetected %d videos, with a total of %d video pair data" %
(len(video_infos), total_pairs))
print "Target: %s" % oauth_util.consts.SERVER_URL
if raw_input("Proceed to upload? [Y/n]: ").lower() in ['', 'y', 'yes']:
upload_to_gae(score_type, video_infos)
def main(table_location,
target_db,
target_collection,
column_info,
partition_cols,
options):
boto_util.initialize_creds_from_file()
# TODO(benkomalo): handle other formats? It may also be compressed, so
# we may have to do more work.
# Default row format for the tables are delimited by this control char.
# when stored on disk.
delimiter = '\01'
key_index = options.key_index
num_cols = len(column_info)
col_names, col_types, _ = zip(*column_info)
if key_index >= num_cols:
raise Exception("Invalid key index (there aren't enough columns)")
base_partition_values = {}
for partition in partition_cols:
key, value = partition.split('=')
base_partition_values[key] = value
table_location += _format_path_from_partition(partition_cols)
# Open our target db connection
mongo_conn = pymongo.Connection(options.report_db_host,
port=options.report_db_port)
mongodb = mongo_conn[target_db]
mongo_collection = mongodb[target_collection]
# If drop flag was set, nuke any pre-existing data
if options.drop:
print "\nDropping existing data in collection %s." % target_collection
# Even though the option is called 'drop', I am instead
# calling 'remove', because it leaves any indexes while deleting data.
mongo_collection.remove()
# Open our input connections.
s3conn = boto.connect_s3()
bucket = s3conn.get_bucket('ka-mapreduce')
path_prefix = table_location[len('s3://ka-mapreduce/'):] + '/'
s3keys = bucket.list(prefix=path_prefix)
# TODO(benkomalo): add a flag to bail on any errors so no partial data is
# saved?
# Counts of rows saved, and rows with errors.
saved = 0
errors = 0
NULL_STRING = '\N'
docs_to_insert = []
batch_size = max(1, options.batch_size)
lines_parsed = 0
# Note: a table's data may be broken down into multiple files on disk.
for key in s3keys:
if key.name.endswith('_$folder$'):
# S3 meta data - not useful.
continue
# Find out if files are under additional partition "directories"
# e.g. "start_dt=2012-01-01/end_dt=2012-02-01/some_file"
base_path = key.name[len(path_prefix):]
path_parts = base_path.split('/')
partition_values = base_partition_values.copy()
for partition in path_parts[:-1]:
name, value = partition.split('=')
partition_values[name] = value
contents = key.get_contents_as_string()
for line in contents.split('\n'):
if not line:
# EOF
break
# HACK: some files are '\t' delimited? Switch delimiters.
# TODO(benkomalo): read format from metadata in Hive master?
if delimiter not in line:
delimiter = '\t'
parts = line.strip().split(delimiter)
if len(parts) != num_cols:
# TODO(benkomalo): properly handle? shouldn't happen though.
sys.stderr.write(
"Unexpected number of columns in row (expected [%s]):\n" %
num_cols)
print >> sys.stderr, parts
continue
doc = {}
for i, (name, type) in enumerate(zip(col_names, col_types)):
# TODO(benkomalo): deal with other types and possible UTF-8
# issues?
try:
if type == 'int':
value = int(parts[i])
elif type == 'boolean':
value = parts[i] == 'true'
elif type in ['double', 'float']:
value = float(parts[i])
else:
value = parts[i]
except Exception:
if parts[i] == NULL_STRING:
# TODO(benkomalo): figure out why our data has this.
# It seems that sometimes Hive likes to put in
# NULL values for ints and booleans? They don't parse
# well. This is unfortunate - just skip the row since
# it's pretty rare for now.
doc = None
break
raise
doc[name] = value
if key_index > -1:
# mongo primary keys are labelled as "_id"
doc['_id'] = parts[key_index]
if doc:
doc.update(partition_values)
# If we have doc IDs, perform a single upsert because bulk
# upserts are not currently supported by pymongo according to
# this old answer: http://stackoverflow.com/questions/5292370
if '_id' in doc:
mongo_collection.save(doc)
saved += 1
else:
docs_to_insert.append(doc)
else:
errors += 1
# Bulk insert docs in batches
if len(docs_to_insert) >= batch_size:
mongo_collection.insert(docs_to_insert)
saved += len(docs_to_insert)
docs_to_insert = []
if lines_parsed % 100 == 0:
print "\rSaved %s docs with %s errors..." % (saved, errors),
sys.stdout.flush()
lines_parsed += 1
if docs_to_insert:
mongo_collection.insert(docs_to_insert)
saved += len(docs_to_insert)
print "\nSummary of results:"
print "\tSaved [%s] documents" % saved
print "\tSkipped [%s] documents with errors" % errors
try:
hipchat.config.init_cfg("hipchat.cfg")
except Exception, e:
# No hipchat.cfg file found - the token will be empty and handled below.
pass
if not hipchat.config.token:
print >> sys.stderr, (
"Can't find HipChat token. Make a hipchat.cfg file "
+ 'with a single line "token = <token_value>" '
+ "(don't forget to chmod 600) either in this directory "
+ "or in your $HOME directory"
)
sys.exit(-1)
boto_util.initialize_creds_from_file()
def popen_results(args):
proc = subprocess.Popen(args, stdout=subprocess.PIPE)
return proc.communicate()[0]
def parse_git_version():
return popen_results(["git", "rev-parse", "HEAD"]).strip()
def parse_git_message():
return popen_results(["git", "show", "-s", "--pretty=format:%s"]).strip()
def main(table_location, score_type, options):
boto_util.initialize_creds_from_file()
# TODO(benkomalo): factor some of this boilerplate out to boto_util
# Open our input connections.
s3conn = boto.connect_s3()
bucket = s3conn.get_bucket('ka-mapreduce')
path = table_location[len('s3://ka-mapreduce/'):]
if not path.endswith('/'):
path = path + '/'
s3keys = bucket.list(path)
# Mapping of video key to the info on the other videos which best match
# that video
# vid_key -> VideoInfo
video_infos = {}
# Note: a table's data may be broken down into multiple files on disk.
delimiter = '\01'
lines_read = 0
version = None # Use a datestamp as a version.
for key in s3keys:
if key.name.endswith('_$.folder$'):
# S3 meta data - not useful.
continue
contents = key.get_contents_as_string()
version = max(version, key.last_modified)
for line in contents.split('\n'):
if not line:
# EOF
break
lines_read += 1
parts = line.rstrip().split(delimiter)
if len(parts) != 3:
# TODO(benkomalo): error handling
continue
vid1_key, vid2_key, score = parts
try:
score = float(score)
except ValueError:
# Some of the values were invalid - deal with it.
# TODO(benkomalo): error handling.
continue
add_score(video_infos, vid1_key, vid2_key, score)
if lines_read % 1000 == 0:
print "Read %s lines..." % lines_read
# Convert version datestamp to a more sane ISO8601 from RFC822
version = rfc822.parsedate_tz(version)[:6] # extract YMDHMS from tuple
version = datetime.datetime(*version).isoformat()
total_pairs = sum(
[len(info.best_matches) for info in video_infos.values()])
print "\nSummary of collected data:"
print "\tScore type: [%s]" % score_type
print "\tVersion: [%s]" % version
print("\tDetected %d videos, with a total of %d video pair data" %
(len(video_infos), total_pairs))
print "Target: %s" % oauth_util.consts.SERVER_URL
if raw_input("Proceed to upload? [Y/n]: ").lower() in ['', 'y', 'yes']:
upload_to_gae(score_type, version, video_infos)
print "Success!"
print "Run the following to make it live:"
print "set_video_matrix_version.py '%s' '%s'" % (score_type, version)
def main(table_location, score_type, options):
boto_util.initialize_creds_from_file()
# TODO(benkomalo): factor some of this boilerplate out to boto_util
# Open our input connections.
s3conn = boto.connect_s3()
bucket = s3conn.get_bucket('ka-mapreduce')
path = table_location[len('s3://ka-mapreduce/'):]
if not path.endswith('/'):
path = path + '/'
s3keys = bucket.list(path)
# Mapping of video key to the info on the other videos which best match
# that video
# vid_key -> VideoInfo
video_infos = {}
# Note: a table's data may be broken down into multiple files on disk.
delimiter = '\01'
lines_read = 0
version = None # Use a datestamp as a version.
for key in s3keys:
if key.name.endswith('_$.folder$'):
# S3 meta data - not useful.
continue
contents = key.get_contents_as_string()
version = max(version, key.last_modified)
for line in contents.split('\n'):
if not line:
# EOF
break
lines_read += 1
parts = line.rstrip().split(delimiter)
if len(parts) != 3:
# TODO(benkomalo): error handling
continue
vid1_key, vid2_key, score = parts
try:
score = float(score)
except ValueError:
# Some of the values were invalid - deal with it.
# TODO(benkomalo): error handling.
continue
add_score(video_infos, vid1_key, vid2_key, score)
if lines_read % 1000 == 0:
print "Read %s lines..." % lines_read
# Convert version datestamp to a more sane ISO8601 from RFC822
version = rfc822.parsedate_tz(version)[:6] # extract YMDHMS from tuple
version = datetime.datetime(*version).isoformat()
total_pairs = sum([len(info.best_matches)
for info in video_infos.values()])
print "\nSummary of collected data:"
print "\tScore type: [%s]" % score_type
print "\tVersion: [%s]" % version
print ("\tDetected %d videos, with a total of %d video pair data" %
(len(video_infos), total_pairs))
print "Target: %s" % oauth_util.consts.SERVER_URL
if raw_input("Proceed to upload? [Y/n]: ").lower() in ['', 'y', 'yes']:
upload_to_gae(score_type, version, video_infos)
print "Success!"
print "Run the following to make it live:"
print "set_video_matrix_version.py '%s' '%s'" % (
score_type, version)
def main(table_location, target_db, target_collection, column_info,
partition_cols, options):
boto_util.initialize_creds_from_file()
# TODO(benkomalo): handle other formats? It may also be compressed, so
# we may have to do more work.
# Default row format for the tables are delimited by this control char.
# when stored on disk.
delimiter = '\01'
key_index = options.key_index
num_cols = len(column_info)
col_names, col_types, _ = zip(*column_info)
if key_index >= num_cols:
raise Exception("Invalid key index (there aren't enough columns)")
base_partition_values = {}
for partition in partition_cols:
key, value = partition.split('=')
base_partition_values[key] = value
table_location += _format_path_from_partition(partition_cols)
# Open our target db connection
mongo_conn = pymongo.Connection(options.report_db_host,
port=options.report_db_port)
mongodb = mongo_conn[target_db]
mongo_collection = mongodb[target_collection]
# If drop flag was set, nuke any pre-existing data
if options.drop:
print "\nDropping existing data in collection %s." % target_collection
# Even though the option is called 'drop', I am instead
# calling 'remove', because it leaves any indexes while deleting data.
mongo_collection.remove()
elif options.drop_partition:
print(
"\nDropping existing data in collection {0} "
"on partition spec {1}".format(target_collection, partition_cols))
mongo_collection.remove(base_partition_values)
# Open our input connections.
s3conn = boto.connect_s3()
bucket = s3conn.get_bucket('ka-mapreduce')
path_prefix = table_location[len('s3://ka-mapreduce/'):] + '/'
s3keys = bucket.list(prefix=path_prefix)
# TODO(benkomalo): add a flag to bail on any errors so no partial data is
# saved?
# Counts of rows saved, and rows with errors.
saved = 0
errors = 0
NULL_STRING = '\N'
docs_to_insert = []
batch_size = max(1, options.batch_size)
lines_parsed = 0
# Note: a table's data may be broken down into multiple files on disk.
for key in s3keys:
if key.name.endswith('_$folder$'):
# S3 meta data - not useful.
continue
# Find out if files are under additional partition "directories"
# e.g. "start_dt=2012-01-01/end_dt=2012-02-01/some_file"
base_path = key.name[len(path_prefix):]
path_parts = base_path.split('/')
partition_values = base_partition_values.copy()
for partition in path_parts[:-1]:
name, value = partition.split('=')
partition_values[name] = value
contents = key.get_contents_as_string()
for line in contents.split('\n'):
if not line:
# EOF
break
# HACK: some files are '\t' delimited? Switch delimiters.
# TODO(benkomalo): read format from metadata in Hive master?
if delimiter not in line:
delimiter = '\t'
parts = line.strip().split(delimiter)
if len(parts) != num_cols:
# TODO(benkomalo): properly handle? shouldn't happen though.
sys.stderr.write(
"Unexpected number of columns in row (expected [%s]):\n" %
num_cols)
print >> sys.stderr, parts
continue
doc = {}
for i, (name, type) in enumerate(zip(col_names, col_types)):
# TODO(benkomalo): deal with other types and possible UTF-8
# issues?
try:
if type == 'int':
value = int(parts[i])
elif type == 'boolean':
value = parts[i] == 'true'
elif type in ['double', 'float']:
value = float(parts[i])
else:
value = parts[i]
except ValueError:
if parts[i] == NULL_STRING:
# Some queries (especially OUTER JOINS) can result in
# null values in non string type fields.
# In case the person who wrote the query does not
# replace resultant NULLs with default values
# the importer should still be able to copy the table
value = None
else:
doc = None
break
doc[name] = value
if key_index > -1 and doc:
# mongo primary keys are labelled as "_id"
doc['_id'] = parts[key_index]
if doc:
doc.update(partition_values)
# If we have doc IDs, perform a single upsert because bulk
# upserts are not currently supported by pymongo according to
# this old answer: http://stackoverflow.com/questions/5292370
if '_id' in doc:
mongo_collection.save(doc)
saved += 1
else:
docs_to_insert.append(doc)
else:
errors += 1
# Bulk insert docs in batches
if len(docs_to_insert) >= batch_size:
mongo_collection.insert(docs_to_insert)
saved += len(docs_to_insert)
docs_to_insert = []
if lines_parsed % 100 == 0:
print "\rSaved %s docs with %s errors..." % (saved, errors),
sys.stdout.flush()
lines_parsed += 1
if docs_to_insert:
mongo_collection.insert(docs_to_insert)
saved += len(docs_to_insert)
print "\nSummary of results:"
print "\tSaved [%s] documents" % saved
print "\tSkipped [%s] documents with errors" % errors
"python deploy.py geoip"
For even more information about GeoIP refer to Makefile in this directory
"""
import optparse
import os
import subprocess
import sys
import boto
sys.path.append(os.path.join(os.path.dirname(__file__), '..', 'src'))
import boto_util
import notify
boto_util.initialize_creds_from_file()
def popen_results(args):
proc = subprocess.Popen(args, stdout=subprocess.PIPE)
return proc.communicate()[0]
def parse_git_version():
return popen_results(['git', 'rev-parse', 'HEAD']).strip()
def parse_git_message():
return popen_results(['git', 'show', '-s', '--pretty=format:%s']).strip()