-
Notifications
You must be signed in to change notification settings - Fork 313
/
S3Aggregator.py
409 lines (360 loc) · 15.3 KB
/
S3Aggregator.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
import base64
import gzip
import hashlib
import io
import json
import queue
import random
import time
from collections import defaultdict
from typing import Any, DefaultDict, Dict, List, Optional
import boto3
import pandas as pd
import pyarrow as pa
import pyarrow.parquet as pq
import s3fs
from botocore.client import Config
from botocore.exceptions import ClientError, EndpointConnectionError
from pyarrow.filesystem import S3FSWrapper # noqa
from .BaseAggregator import (RECORD_TYPE_CONTENT, BaseAggregator, BaseListener,
BaseParams)
from .parquet_schema import PQ_SCHEMAS
CACHE_SIZE = 500
SITE_VISITS_INDEX = '_site_visits_index'
CONTENT_DIRECTORY = 'content'
CONFIG_DIR = 'config'
BATCH_COMMIT_TIMEOUT = 30 # commit a batch if no new records for N seconds
S3_CONFIG_KWARGS = {
'retries': {
'max_attempts': 20
}
}
S3_CONFIG = Config(**S3_CONFIG_KWARGS)
def listener_process_runner(base_params: BaseParams,
manager_params: Dict[str, Any],
instance_id: int) -> None:
"""S3Listener runner. Pass to new process"""
listener = S3Listener(base_params, manager_params, instance_id)
listener.startup()
while True:
listener.update_status_queue()
listener.save_batch_if_past_timeout()
if listener.should_shutdown():
break
try:
record = listener.record_queue.get(block=True, timeout=5)
listener.process_record(record)
except queue.Empty:
pass
listener.drain_queue()
listener.shutdown()
class S3Listener(BaseListener):
"""Listener that pushes aggregated records to S3.
Records for each page visit are stored in memory during a page visit. Once
the browser moves to another page, the data is written to S3 as part of
a parquet dataset. The schema for this dataset is given in
./parquet_schema.py
"""
def __init__(self,
base_params: BaseParams,
manager_params: Dict[str, Any],
instance_id: int) -> None:
self.dir = manager_params['s3_directory']
self._records: Dict[int, DefaultDict[str, List[Any]]] =\
dict() # maps visit_id and table to records
self._batches: DefaultDict[str, List[pa.RecordBatch]] = \
defaultdict(list) # maps table_name to a list of batches
self._unsaved_visit_ids: List[int] = \
list()
self._instance_id = instance_id
self._bucket = manager_params['s3_bucket']
self._s3_content_cache = set() # cache of filenames already uploaded
self._s3 = boto3.client('s3', config=S3_CONFIG)
self._s3_resource = boto3.resource('s3', config=S3_CONFIG)
self._fs = s3fs.S3FileSystem(
session=boto3.DEFAULT_SESSION,
config_kwargs=S3_CONFIG_KWARGS
)
self._s3_bucket_uri = 's3://%s/%s/visits/%%s' % (
self._bucket, self.dir)
# time last record was received
self._last_record_received: Optional[float] = None
super(S3Listener, self).__init__(*base_params)
def _get_records(self, visit_id: int) -> DefaultDict[str, List[Any]]:
"""Get the RecordBatch corresponding to `visit_id`"""
if visit_id not in self._records:
self._records[visit_id] = defaultdict(list)
return self._records[visit_id]
def _write_record(self, table, data, visit_id):
"""Insert data into a RecordBatch"""
records = self._get_records(visit_id)
# Add nulls
for item in PQ_SCHEMAS[table].names:
if item not in data:
data[item] = None
# Add instance_id (for partitioning)
data['instance_id'] = self._instance_id
records[table].append(data)
def _create_batch(self, visit_id):
"""Create record batches for all records from `visit_id`"""
if visit_id not in self._records:
# The batch for this `visit_id` was already created, skip
return
for table_name, data in self._records[visit_id].items():
try:
df = pd.DataFrame(data)
batch = pa.RecordBatch.from_pandas(
df, schema=PQ_SCHEMAS[table_name], preserve_index=False
)
self._batches[table_name].append(batch)
self.logger.debug(
"Successfully created batch for table %s and "
"visit_id %s" % (table_name, visit_id)
)
except pa.lib.ArrowInvalid:
self.logger.error(
"Error while creating record batch for table %s\n"
% table_name, exc_info=True
)
pass
self._unsaved_visit_ids.append(visit_id)
# We construct a special index file from the site_visits data
# to make it easier to query the dataset
if table_name == 'site_visits':
if SITE_VISITS_INDEX not in self._batches:
self._batches[SITE_VISITS_INDEX] = list()
for item in data:
self._batches[SITE_VISITS_INDEX].append(item)
del self._records[visit_id]
def _exists_on_s3(self, filename):
"""Check if `filename` already exists on S3"""
# Check local filename cache
if filename.split('/', 1)[1] in self._s3_content_cache:
self.logger.debug(
"File `%s` found in content cache." % filename)
return True
# Check S3
try:
self._s3_resource.Object(self._bucket, filename).load()
except ClientError as e:
if e.response['Error']['Code'] == "404":
return False
else:
raise
except EndpointConnectionError:
self.logger.error(
"Exception while checking if file exists %s" % filename,
exc_info=True
)
return False
# Add filename to local cache to avoid remote lookups on next request
# We strip the bucket name as its the same for all files
self._s3_content_cache.add(filename.split('/', 1)[1])
return True
def _write_str_to_s3(self, string, filename,
compressed=True, skip_if_exists=True):
"""Write `string` data to S3 with name `filename`"""
if skip_if_exists and self._exists_on_s3(filename):
self.logger.debug(
"File `%s` already exists on s3, skipping..." % filename)
return
if not isinstance(string, bytes):
string = string.encode('utf-8')
if compressed:
out_f = io.BytesIO()
with gzip.GzipFile(fileobj=out_f, mode='w') as writer:
writer.write(string)
out_f.seek(0)
else:
out_f = io.BytesIO(string)
# Upload to S3
try:
self._s3.upload_fileobj(out_f, self._bucket, filename)
self.logger.debug(
"Successfully uploaded file `%s` to S3." % filename)
# Cache the filenames that are already on S3
# We strip the bucket name as its the same for all files
if skip_if_exists:
self._s3_content_cache.add(filename.split('/', 1)[1])
except Exception:
self.logger.error(
"Exception while uploading %s" % filename, exc_info=True
)
pass
def _send_to_s3(self, force=False):
"""Copy in-memory batches to s3"""
should_send = force
for batches in self._batches.values():
if len(batches) > CACHE_SIZE:
should_send = True
if not should_send:
return
for table_name, batches in self._batches.items():
if table_name == SITE_VISITS_INDEX:
out_str = '\n'.join([json.dumps(x) for x in batches])
if not isinstance(out_str, bytes):
out_str = out_str.encode('utf-8')
fname = '%s/site_index/instance-%s-%s.json.gz' % (
self.dir, self._instance_id,
hashlib.md5(out_str).hexdigest()
)
self._write_str_to_s3(out_str, fname)
else:
if len(batches) == 0:
continue
try:
table = pa.Table.from_batches(batches)
pq.write_to_dataset(
table, self._s3_bucket_uri % table_name,
filesystem=self._fs,
partition_cols=['instance_id'],
compression='snappy',
flavor='spark'
)
except (pa.lib.ArrowInvalid, EndpointConnectionError):
self.logger.error(
"Error while sending records for: %s" % table_name,
exc_info=True
)
pass
# can't del here because that would modify batches
self._batches[table_name] = list()
for visit_id in self._unsaved_visit_ids:
self.mark_visit_complete(visit_id)
self._unsaved_visit_ids = list()
def save_batch_if_past_timeout(self):
"""Save the current batch of records if no new data has been received.
If we aren't receiving new data for this batch we commit early
regardless of the current batch size."""
if self._last_record_received is None:
return
if time.time() - self._last_record_received < BATCH_COMMIT_TIMEOUT:
return
self.logger.debug(
"Saving current record batches to S3 since no new data has "
"been written for %d seconds." %
(time.time() - self._last_record_received)
)
self.drain_queue()
self._last_record_received = None
def process_record(self, record):
"""Add `record` to database"""
if len(record) != 2:
self.logger.error("Query is not the correct length")
return
self._last_record_received = time.time()
table, data = record
if table == "create_table": # drop these statements
return
elif table == RECORD_TYPE_CONTENT:
self.process_content(record)
return
_, visit_id = self.update_records(table, data)
# Convert data to text type
for k, v in data.items():
if isinstance(v, bytes):
data[k] = str(v, errors='ignore')
elif callable(v):
data[k] = str(v)
# TODO: Can we fix this in the extension?
elif type(v) == dict:
data[k] = json.dumps(v)
# Save record to disk
self._write_record(table, data, visit_id)
def process_content(self, record):
"""Upload page content `record` to S3"""
if record[0] != RECORD_TYPE_CONTENT:
raise ValueError(
"Incorrect record type passed to `process_content`. Expected "
"record of type `%s`, received `%s`." % (
RECORD_TYPE_CONTENT, record[0])
)
content, content_hash = record[1]
content = base64.b64decode(content)
fname = "%s/%s/%s.gz" % (self.dir, CONTENT_DIRECTORY, content_hash)
self._write_str_to_s3(content, fname)
def drain_queue(self):
"""Process remaining records in queue and sync final files to S3"""
super(S3Listener, self).drain_queue()
# can't directly iterate because _create_batch modifies records
visit_ids = list(self._records.keys())
for visit_id in visit_ids:
self._create_batch(visit_id)
self._send_to_s3(force=True)
def run_visit_completion_tasks(self, visit_id: int,
is_shutdown: bool = False):
self._create_batch(visit_id)
self._send_to_s3(force=is_shutdown)
class S3Aggregator(BaseAggregator):
"""
Receives data records from other processes and aggregates them locally
per-site before pushing them to a remote S3 bucket. The remote files are
saved in a Paquet Dataset partitioned by the crawl_id and visit_id of
each record.
The visit and task ids are randomly generated to allow multiple writers
to write to the same S3 bucket. Every record should have a `visit_id`
(which identifies the site visit) and a `crawl_id` (which identifies the
browser instance) so we can associate records with the appropriate meta
data. Any records which lack this information will be dropped by the
writer.
Note: Parquet's partitioned dataset reader only supports integer partition
columns up to 32 bits. Currently, `instance_id` is the only partition
column, and thus can be no larger than 32 bits.
"""
def __init__(self, manager_params, browser_params):
super(S3Aggregator, self).__init__(manager_params, browser_params)
self.dir = manager_params['s3_directory']
self.bucket = manager_params['s3_bucket']
self.s3 = boto3.client('s3')
self._instance_id = random.getrandbits(32)
self._create_bucket()
def _create_bucket(self):
"""Create remote S3 bucket if it doesn't exist"""
resource = boto3.resource('s3')
try:
resource.meta.client.head_bucket(Bucket=self.bucket)
except ClientError as e:
error_code = int(e.response['Error']['Code'])
if error_code == 404:
resource.create_bucket(Bucket=self.bucket)
else:
raise
def save_configuration(self, openwpm_version, browser_version):
"""Save configuration details for this crawl to the database"""
# Save config keyed by task id
fname = "%s/%s/instance-%s_configuration.json" % (
self.dir, CONFIG_DIR, self._instance_id)
# Config parameters for update
out = dict()
out['manager_params'] = self.manager_params
out['openwpm_version'] = str(openwpm_version)
out['browser_version'] = str(browser_version)
out['browser_params'] = self.browser_params
out_str = json.dumps(out)
if not isinstance(out_str, bytes):
out_str = out_str.encode('utf-8')
out_f = io.BytesIO(out_str)
# Upload to S3 and delete local copy
try:
self.s3.upload_fileobj(out_f, self.bucket, fname)
except Exception:
self.logger.error("Exception while uploading %s" % fname)
raise
def get_next_visit_id(self):
"""Generate visit id as randomly generated positive integer less than 2^53.
Parquet can support integers up to 64 bits, but Javascript can only
represent integers up to 53 bits:
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Number/MAX_SAFE_INTEGER
Thus, we cap these values at 53 bits.
"""
return random.getrandbits(53)
def get_next_crawl_id(self):
"""Generate crawl id as randomly generated positive 32bit integer
Note: Parquet's partitioned dataset reader only supports integer
partition columns up to 32 bits.
"""
return random.getrandbits(32)
def launch(self):
"""Launch the aggregator listener process"""
super(S3Aggregator, self).launch(
listener_process_runner, self.manager_params, self._instance_id)