def __init__(self, cfg):
self.job_bundle = cfg.job_bundle
if cfg.input_filter:
self.input_filter = TelemetrySchema(
json.load(open(cfg.input_filter)))
else:
self.input_filter = None
if cfg.input_list_file:
self.input_list = cfg.input_list_file
else:
self.input_list = None
self.job_id = str(uuid4())
self.target_queue = cfg.target_queue
self.aws_key = cfg.aws_key
self.aws_secret_key = cfg.aws_secret_key
self.input_bucket = "telemetry-published-v1"
self.job_name = cfg.name
self.job_owner = cfg.owner
self.date_limit = cfg.date_limit
# Bucket with intermediate data for this analysis job
self.analysis_bucket = "jonasfj-telemetry-analysis"
self.s3_code_path = "batch-jobs/" + self.job_id + ".tar.gz"
# S3 region of operation
self.aws_region = "us-west-2"
self.task_size_limit = 400 * 1024 * 1024
self.sqs_input_name = cfg.sqs_queue
def _filter_to_schema(schema, filter_args):
new_schema = {"version": 1, "dimensions": []}
for i, dim in enumerate(schema["dimensions"]):
new_filter = {
"field_name":
schema["dimensions"][i].get("field_name", "field{}".format(i)),
"allowed_values":
"*"
}
if dim["field_name"] in filter_args:
new_filter["allowed_values"] = filter_args[dim["field_name"]]
new_schema["dimensions"].append(new_filter)
return TelemetrySchema(new_schema)
def test_v4execschema():
schema_spec = {
"version":
2,
"dimensions": [{
"field_name": "submissionDate",
"allowed_values": {
"max": "20150901"
}
}]
}
schema = TelemetrySchema(schema_spec)
found = set()
for f in s3util.list_heka_partitions(v4execbucket, schema=schema):
found.add(f.name)
assert (len(found) == 3)
assert ("20150901/20150901221519.541_ip-172-31-16-184" in found)
assert ("20150901/20150901223019.579_ip-172-31-16-184" in found)
assert ("20150901/20150901224519.623_ip-172-31-16-184" in found)
# Test with a prefix:
found = set()
for f in s3util.list_heka_partitions(
v4prefixbucket, prefix="telemetry-executive-summary-2",
schema=schema):
found.add(f.name)
assert (len(found) == 3)
assert (
"telemetry-executive-summary-2/20150901/20150901221519.541_ip-172-31-16-184"
in found)
assert (
"telemetry-executive-summary-2/20150901/20150901223019.579_ip-172-31-16-184"
in found)
assert (
"telemetry-executive-summary-2/20150901/20150901224519.623_ip-172-31-16-184"
in found)
# Test with a bunch of prefixes:
found = set()
for f in s3util.list_heka_partitions(multiprefixbucket,
prefix="a/b/c/d",
schema=schema):
found.add(f.name)
assert (len(found) == 3)
assert ("a/b/c/d/20150901/20150901221519.541_ip-172-31-16-184" in found)
assert ("a/b/c/d/20150901/20150901223019.579_ip-172-31-16-184" in found)
assert ("a/b/c/d/20150901/20150901224519.623_ip-172-31-16-184" in found)
def main():
parser = ArgumentParser(
description='Convert local Telemetry pings to server storage structure'
)
parser.add_argument("--input-dir", required=True)
parser.add_argument("--output-dir", required=True)
parser.add_argument("--schema",
type=file,
default='./telemetry/telemetry_schema.json')
parser.add_argument("--histogram-cache-dir",
default='/tmp/telemetry_histogram_cache')
args = parser.parse_args()
print "Getting pings from", args.input_dir, "converting them and storing them in", args.output_dir
schema = TelemetrySchema(json.load(args.schema))
cache = RevisionCache(args.histogram_cache_dir, 'hg.mozilla.org')
converter = Converter(cache, schema)
storage = StorageLayout(schema, args.output_dir, 500000000)
ping_dir = args.input_dir
ping_files = get_pings(ping_dir)
if len(ping_files) == 0:
# Try the usual ping dir (if the user just gave the Profile Dir)
ping_dir = os.path.join(args.input_dir, "saved-telemetry-pings")
ping_files = get_pings(ping_dir)
print "found", len(ping_files), "pings"
for ping_file in ping_files:
with open(os.path.join(ping_dir, ping_file), "r") as f:
ping = json.load(f)
reason = ping['reason']
key = ping['slug']
payload = ping['payload']
submission_date = date.today().strftime("%Y%m%d")
dims = schema.dimensions_from(payload, submission_date)
try:
parsed_data, dims = converter.convert_obj(payload, dims[-1])
serialized_data = converter.serialize(parsed_data)
data_version = Converter.VERSION_CONVERTED
try:
# Write to persistent storage
n = storage.write(key, serialized_data, dims, data_version)
print "Successfully saved ping", key, "to", n
except Exception, e:
traceback.print_exc()
except BadPayloadError, e:
print "Bad Payload:", e.msg
except Exception, e:
traceback.print_exc()
def test_v4schema():
schema_spec = {
"version":
2,
"dimensions": [{
"field_name": "submissionDate",
"allowed_values": "20150903"
}, {
"field_name": "sourceName",
"allowed_values": "*"
}, {
"field_name": "sourceVersion",
"allowed_values": "4"
}, {
"field_name": "docType",
"allowed_values": ["saved-session"]
}, {
"field_name": "appName",
"allowed_values": ["Firefox"]
}, {
"field_name": "appUpdateChannel",
"allowed_values": ["release"]
}, {
"field_name": "appVersion",
"allowed_values": "24.0"
}, {
"field_name": "appBuildId",
"allowed_values": "20130910160258"
}]
}
schema = TelemetrySchema(schema_spec)
found = set()
for f in s3util.list_heka_partitions(v4bucket, schema=schema):
found.add(f.name)
assert (len(found) == 3)
assert (
"20150903/telemetry/4/saved_session/Firefox/release/24.0/20130910160258/20150903051633.482_ip-172-31-16-184"
in found)
assert (
"20150903/telemetry/4/saved_session/Firefox/release/24.0/20130910160258/20150903051644.482_ip-172-31-16-184"
in found)
assert (
"20150903/telemetry/4/saved_session/Firefox/release/24.0/20130910160258/20150903051655.482_ip-172-31-16-184"
in found)
def test_schema(d):
schema_spec = {
"version":
1,
"dimensions": [{
"field_name": "reason",
"allowed_values": ["saved-session"]
}, {
"field_name": "appName",
"allowed_values": ["Firefox"]
}, {
"field_name": "appUpdateChannel",
"allowed_values": ["nightly"]
}, {
"field_name": "appVersion",
"allowed_values": ["27.0a1"]
}, {
"field_name": "appBuildID",
"allowed_values": ["20130918030202"]
}, {
"field_name": "submission_date",
"allowed_values": ["20131001"]
}]
}
schema = TelemetrySchema(schema_spec)
successfully_downloaded = []
failfully_downloaded = []
for f, r, err in d.get_schema(schema):
if err is not None:
print err
failfully_downloaded.append(f)
else:
print "Downloaded", f
successfully_downloaded.append(f)
assert len(failfully_downloaded) == 0
print "Successfully downloaded", len(successfully_downloaded)
assert len(successfully_downloaded) == 20
def test_v2schema():
schema_spec = {
"version":
1,
"dimensions": [{
"field_name": "reason",
"allowed_values": ["saved-session"]
}, {
"field_name": "appName",
"allowed_values": ["Firefox"]
}, {
"field_name": "appUpdateChannel",
"allowed_values": ["release"]
}, {
"field_name": "appVersion",
"allowed_values": ["24.0"]
}, {
"field_name": "appBuildID",
"allowed_values": ["20130910160258"]
}, {
"field_name": "submission_date",
"allowed_values": ["20131003", "20131004"]
}]
}
schema = TelemetrySchema(schema_spec)
found = set()
for f in s3util.list_partitions(v2bucket, schema=schema,
include_keys=True):
found.add(f.name)
assert (len(found) == 2)
assert (
"saved_session/Firefox/release/24.0/20130910160258.20131003.v2.log.25b53e7042c74188b08d71ce32e87237.lzma"
in found)
assert (
"saved_session/Firefox/release/24.0/20130910160258.20131004.v2.log.29afd7a250154729bd53c20253f8af78.lzma"
in found)
def __init__(self, config):
# Sanity check args.
if config.get("num_mappers") <= 0:
raise ValueError("Number of mappers must be greater than zero")
if config.get("num_reducers") <= 0:
raise ValueError("Number of reducers must be greater than zero")
if not os.path.isdir(config.get("data_dir")):
raise ValueError("Data dir must be a valid directory")
if not os.path.isdir(config.get("work_dir")):
raise ValueError("Work dir must be a valid directory")
if not os.path.isfile(config.get("job_script", "")):
raise ValueError("Job script must be a valid python file")
if not os.path.isfile(config.get("input_filter")):
raise ValueError("Input filter must be a valid json file")
self._input_dir = config.get("data_dir")
if self._input_dir[-1] == os.path.sep:
self._input_dir = self._input_dir[0:-1]
self._work_dir = config.get("work_dir")
with open(config.get("input_filter")) as filter_file:
self._input_filter = TelemetrySchema(json.load(filter_file))
self._allowed_values = self._input_filter.sanitize_allowed_values()
self._output_file = config.get("output")
self._num_mappers = config.get("num_mappers")
self._num_reducers = config.get("num_reducers")
self._local_only = config.get("local_only")
self._bucket_name = config.get("bucket")
self._aws_key = config.get("aws_key")
self._aws_secret_key = config.get("aws_secret_key")
self._profile = config.get("profile")
self._delete_data = config.get("delete_data")
with open(config.get("job_script")) as modulefd:
# let the job script import additional modules under its path
sys.path.append(os.path.dirname(config.get("job_script")))
## Lifted from FileDriver.py in jydoop.
self._job_module = imp.load_module("telemetry_job", modulefd,
config.get("job_script"),
('.py', 'U', 1))
def update_published_v4_files(sdb,
bucket,
bucket_prefix,
submission_date,
limit=None):
conn = boto.connect_s3(host=S3_DEFAULT_ENDPOINT)
metadata = conn.get_bucket(METADATA_BUCKET, validate=False)
schema_key = metadata.get_key("{}/schema.json".format(bucket_prefix))
schema_string = schema_key.get_contents_as_string()
schema = TelemetrySchema(json.loads(schema_string))
bucket = conn.get_bucket(bucket, validate=False)
added_count = 0
total_count = 0
start_time = datetime.now()
done = False
last_key = ''
batch = BatchPut(sdb)
prefix = "{}/{}".format(
bucket_prefix, submission_date) if submission_date else bucket_prefix
print "Bucket: {} - Prefix: {} - Date: {}".format(bucket.name,
bucket_prefix,
submission_date)
while not done:
try:
for key in bucket.list(marker=last_key, prefix=prefix):
last_key = key.name
if total_count % 1e5 == 0:
print("Looked at {} total records in {} seconds, added {}".
format(total_count, delta_sec(start_time),
added_count))
dims = schema.get_dimension_map(
schema.get_dimensions(".",
key.name[len(bucket_prefix) + 1:],
dirs_only=True))
if (dims["submissionDate"] == submission_date
) and dims["submissionDate"][:-2] in sdb:
batch.put(dims["submissionDate"][:-2], key.name, dims)
added_count += 1
total_count += 1
if total_count == limit:
done = True
break
except Exception as e:
print("Error listing keys: {}".format(e))
traceback.print_exc()
print("Continuing from last seen key: {}".format(last_key))
continue
break
batch.flush()
print("Overall, added {} of {} in {} seconds".format(
added_count, total_count, delta_sec(start_time)))
def main():
signal.signal(signal.SIGINT, handle_sigint)
parser = argparse.ArgumentParser(
description='Process incoming Telemetry data',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("-c",
"--config",
required=True,
type=file,
help="AWS Configuration file (json)")
parser.add_argument("-w",
"--work-dir",
required=True,
help="Location to cache downloaded files")
parser.add_argument("-o",
"--output-dir",
required=True,
help="Base dir to store processed data")
parser.add_argument("-i",
"--input-files",
type=file,
help="File containing a list of keys to process")
parser.add_argument("-b",
"--bad-data-log",
help="Save bad records to this file")
parser.add_argument("-l", "--log-file", help="Log output to this file")
parser.add_argument("-s",
"--stats-file",
help="Log statistics to this file")
parser.add_argument("--histogram-cache-path",
default="./histogram_cache",
help="Path to store a local cache of histograms")
parser.add_argument("-t",
"--telemetry-schema",
required=True,
help="Location of the desired telemetry schema")
parser.add_argument("-m",
"--max-output-size",
metavar="N",
type=int,
default=500000000,
help="Rotate output files after N bytes")
parser.add_argument("-D",
"--dry-run",
action="store_true",
help="Don't modify remote files")
parser.add_argument("-n",
"--no-clean",
action="store_true",
help="Don't clean out the output-dir before beginning")
parser.add_argument("-v",
"--verbose",
action="store_true",
help="Print more detailed output")
args = parser.parse_args()
if args.verbose:
# Turn on mp logging
multiprocessing.log_to_stderr(logging.DEBUG)
config = json.load(args.config)
# TODO: allow commandline args to override config values.
if not os.path.isdir(args.output_dir):
os.makedirs(args.output_dir)
schema_data = open(args.telemetry_schema)
schema = TelemetrySchema(json.load(schema_data))
schema_data.close()
cache = RevisionCache(args.histogram_cache_path, "hg.mozilla.org")
converter = Converter(cache, schema)
storage = StorageLayout(schema, args.output_dir, args.max_output_size)
logger = Log(args.log_file, "Master")
num_cpus = multiprocessing.cpu_count()
conn = None
incoming_bucket = None
incoming_queue = None
s3downloader = None
raw_readers = None
compressors = None
exporters = None
done = False
if args.no_clean:
logger.log("Not removing log files in {}".format(args.output_dir))
else:
# Remove existing log files from output_dir (to clean up after an
# incomplete previous run, for example).
logger.log("Removing log files in {}".format(args.output_dir))
for root, dirs, files in os.walk(args.output_dir):
for f in files:
if f.endswith(".log"):
full = os.path.join(root, f)
if args.dry_run:
logger.log("Would be deleting {}, except it's a " \
"dry run".format(full))
else:
try:
logger.log("Removing existing file: " + full)
os.remove(full)
except Exception, e:
logger.log("Error removing existing " \
" file {}: {}".format(full, e))
def main():
parser = argparse.ArgumentParser(
description='Split raw logs into partitioned files.',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("-m",
"--max-output-size",
metavar="N",
help="Rotate output files after N bytes",
type=int,
default=500000000)
parser.add_argument("-i",
"--input-file",
help="Filename to read from",
required=True)
parser.add_argument("-o",
"--output-dir",
help="Base directory to store split files",
required=True)
parser.add_argument("-t",
"--telemetry-schema",
help="Filename of telemetry schema spec",
required=True)
parser.add_argument("-f",
"--file-version",
help="Log file version (if omitted, we'll guess)")
args = parser.parse_args()
schema_data = open(args.telemetry_schema)
schema = TelemetrySchema(json.load(schema_data))
schema_data.close()
storage = StorageLayout(schema, args.output_dir, args.max_output_size)
expected_dim_count = len(schema._dimensions)
if not os.path.isdir(args.output_dir):
os.makedirs(args.output_dir)
record_count = 0
bad_record_count = 0
bytes_read = 0
start = datetime.now()
file_version = args.file_version
if not file_version:
file_version = fileutil.detect_file_version(args.input_file)
for r in fileutil.unpack(args.input_file, file_version=file_version):
record_count += 1
if r.error:
bad_record_count += 1
continue
# Incoming timestamps are in milliseconds, so convert to POSIX first
# (ie. seconds)
submission_date = date.fromtimestamp(r.timestamp /
1000).strftime("%Y%m%d")
# Deal with unicode
path = unicode(r.path, errors="replace")
data = unicode(r.data, errors="replace")
bytes_read += r.len_ip + r.len_path + r.len_data + fileutil.RECORD_PREAMBLE_LENGTH[
file_version]
#print "Path for record", record_count, path, "length of data:", r.len_data, "data:", data[0:5] + "..."
path_components = path.split("/")
if len(path_components) != expected_dim_count:
# We're going to pop the ID off, but we'll also add the submission
# date, so it evens out.
print "Found an invalid path in record", record_count, path
bad_record_count += 1
continue
key = path_components.pop(0)
info = {}
info["reason"] = path_components.pop(0)
info["appName"] = path_components.pop(0)
info["appVersion"] = path_components.pop(0)
info["appUpdateChannel"] = path_components.pop(0)
info["appBuildID"] = path_components.pop(0)
dimensions = schema.dimensions_from(info, submission_date)
#print " Converted path to filename", schema.get_filename(args.output_dir, dimensions)
storage.write(key, data, dimensions)
duration = timer.delta_sec(start)
mb_read = bytes_read / 1024.0 / 1024.0
print "Read %.2fMB in %.2fs (%.2fMB/s), %d of %d records were bad" % (
mb_read, duration, mb_read / duration, bad_record_count, record_count)
return 0
def main():
parser = argparse.ArgumentParser(
description='Process incoming Telemetry data',
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument("incoming_bucket",
help="The S3 bucket containing incoming files")
parser.add_argument("publish_bucket",
help="The S3 bucket to save processed files")
parser.add_argument("-k", "--aws-key", help="AWS Key", required=True)
parser.add_argument("-s",
"--aws-secret-key",
help="AWS Secret Key",
required=True)
parser.add_argument("-r",
"--aws-region",
help="AWS Region",
default="us-west-2")
parser.add_argument("-w",
"--work-dir",
help="Location to cache downloaded files",
required=True)
parser.add_argument("-o",
"--output-dir",
help="Base dir to store processed data",
required=True)
parser.add_argument("-i",
"--input-files",
help="File containing a list of keys to process",
type=file)
parser.add_argument("-b",
"--bad-data-log",
help="Save bad records to this file")
parser.add_argument("-q",
"--queue",
help="SQS Queue name to poll for incoming data")
parser.add_argument("-c",
"--histogram-cache-path",
help="Path to store a local cache of histograms",
default="./histogram_cache")
parser.add_argument("-t",
"--telemetry-schema",
help="Location of the desired telemetry schema",
required=True)
parser.add_argument("-m",
"--max-output-size",
metavar="N",
help="Rotate output files after N bytes",
type=int,
default=500000000)
parser.add_argument("-D",
"--dry-run",
help="Don't modify remote files",
action="store_true")
parser.add_argument("-C",
"--skip-conversion",
help="Skip validation/conversion of payloads",
action="store_true")
args = parser.parse_args()
if not os.path.isfile(S3FUNNEL_PATH):
print "ERROR: s3funnel not found at", S3FUNNEL_PATH
print "You can get it from github: https://github.com/sstoiana/s3funnel"
return -1
if not os.path.isdir(args.output_dir):
os.makedirs(args.output_dir)
schema_data = open(args.telemetry_schema)
schema = TelemetrySchema(json.load(schema_data))
schema_data.close()
cache = RevisionCache(args.histogram_cache_path, "hg.mozilla.org")
if args.skip_conversion:
converter = None
else:
converter = Converter(cache, schema)
storage = StorageLayout(schema, args.output_dir, args.max_output_size)
num_cpus = multiprocessing.cpu_count()
start = datetime.now()
conn = None
incoming_bucket = None
incoming_queue = None
incoming_queue_messages = []
if not args.dry_run:
conn = S3Connection(args.aws_key, args.aws_secret_key)
incoming_bucket = conn.get_bucket(args.incoming_bucket)
incoming_filenames = []
if args.queue is not None:
print "Fetching file list from queue", args.queue
if args.dry_run:
print "Dry run mode... can't read from the queue without messing things up..."
else:
q_conn = boto.sqs.connect_to_region(
args.aws_region,
aws_access_key_id=args.aws_key,
aws_secret_access_key=args.aws_secret_key)
incoming_queue = q_conn.get_queue(args.queue)
if incoming_queue is None:
print "Error: could not get queue", args.queue
return -2
# Sometimes we don't get all the messages, even if more are
# available, so keep trying until we have enough (or there aren't
# any left)
for i in range(num_cpus):
messages = incoming_queue.get_messages(num_cpus -
len(incoming_filenames))
for m in messages:
# TODO: Make sure this file exists in S3 first?
possible_filename = m.get_body()
key = incoming_bucket.get_key(possible_filename)
if key is None:
print "Could not find queued filename in bucket", args.incoming_bucket, ":", possible_filename
# try to delete it:
incoming_queue.delete_message(m)
else:
incoming_filenames.append(possible_filename)
incoming_queue_messages.append(m)
if len(messages) == 0 or len(incoming_filenames) >= num_cpus:
break
elif args.input_files:
print "Fetching file list from file", args.input_files
incoming_filenames = [l.strip() for l in args.input_files.readlines()]
else:
print "Fetching file list from S3..."
for f in incoming_bucket.list():
incoming_filenames.append(f.name)
print "Done"
if len(incoming_filenames) == 0:
print "Nothing to do!"
return 0
for f in incoming_filenames:
print " ", f
print "Verifying that we can write to", args.publish_bucket
if args.dry_run:
print "Dry run mode: don't care!"
else:
try:
publish_bucket = conn.get_bucket(args.publish_bucket)
print "Looks good!"
except S3ResponseError:
print "Bucket", args.publish_bucket, "not found. Attempting to create it."
publish_bucket = conn.create_bucket(args.publish_bucket)
result = 0
print "Downloading", len(incoming_filenames), "files..."
if args.dry_run:
print "Dry run mode: skipping download from S3"
else:
result = fetch_s3_files(incoming_filenames, args.work_dir,
incoming_bucket, args.aws_key,
args.aws_secret_key)
if result != 0:
print "Error downloading files. Return code of s3funnel was", result
return result
print "Done"
after_download = datetime.now()
local_filenames = [
os.path.join(args.work_dir, f) for f in incoming_filenames
]
# TODO: try a SimpleQueue
raw_files = Queue()
for l in local_filenames:
raw_files.put(l)
completed_files = Queue()
compressed_files = Queue()
# Begin reading raw input
raw_readers = start_workers(
num_cpus, "Reader", ReadRawStep, raw_files,
(completed_files, schema, converter, storage, args.bad_data_log))
# Tell readers when to stop:
for i in range(num_cpus):
raw_files.put(PipeStep.SENTINEL)
# Compress completed files.
compressors = start_workers(num_cpus, "Compressor", CompressCompletedStep,
completed_files, (compressed_files, ))
# Export compressed files to S3.
exporters = start_workers(
num_cpus, "Exporter", ExportCompressedStep, compressed_files,
(args.output_dir, args.aws_key, args.aws_secret_key,
args.publish_bucket, args.dry_run))
wait_for(raw_readers, "Raw Readers")
# `find <out_dir> -type f -not -name ".compressme"`
# Add them to completed_files
for root, dirs, files in os.walk(args.output_dir):
for f in files:
if f.endswith(".log"):
completed_files.put(os.path.join(root, f))
for i in range(num_cpus):
completed_files.put(PipeStep.SENTINEL)
wait_for(compressors, "Compressors")
for i in range(num_cpus):
compressed_files.put(PipeStep.SENTINEL)
wait_for(exporters, "Exporters")
print "Removing processed logs from S3..."
for f in incoming_filenames:
if args.dry_run:
print " Dry run, so not really deleting", f
else:
print " Deleting", f
incoming_bucket.delete_key(f)
# Delete file locally too.
os.remove(os.path.join(args.work_dir, f))
print "Done"
if len(incoming_queue_messages) > 0:
print "Removing processed messages from SQS..."
for m in incoming_queue_messages:
if args.dry_run:
print " Dry run, so not really deleting", m.get_body()
else:
print " Deleting", m.get_body()
if incoming_queue.delete_message(m):
print " Message deleted successfully"
else:
print " Failed to delete message :("
print "Done"
duration = timer.delta_sec(start)
print "All done in %.2fs (%.2fs excluding download time)" % (
duration, timer.delta_sec(after_download))
return 0
def setUp(self):
test_dir = self.get_test_dir()
self.schema = TelemetrySchema(self.get_schema_spec())
self.storage = StorageLayout(self.schema, test_dir, 10000)
assert not os.path.exists(test_dir)
os.makedirs(test_dir)
