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
|
import re
import sys
import csv
import json
import time
import itertools
import datetime
import requests
from confluent_kafka import Producer
from fatcat_tools.workers import most_recent_message
from .harvest_common import HarvestState, requests_retry_session
class HarvestCrossrefWorker:
"""
Notes on crossref API:
- from-index-date is the updated time
- is-update can be false, to catch only new or only old works
https://api.crossref.org/works?filter=from-index-date:2018-11-14,is-update:false&rows=2
I think the design is going to have to be a cronjob or long-running job
(with long sleeps) which publishes "success through" to a separate state
queue, as simple YYYY-MM-DD strings.
Within a day, will need to use a resumption token. Maybe should use a
crossref library... meh.
will want to have some mechanism in kafka consumer (pushing to fatcat) to group
in batches as well. maybe even pass through as batches? or just use timeouts on
iteration.
logic of this worker:
- on start, fetch latest date from state feed
- in a function (unit-testable), decide which dates to ingest
- for each date needing update:
- start a loop for just that date, using resumption token for this query
- when done, publish to state feed, with immediate sync
TODO: what sort of parallelism? I guess multi-processing on dates, but need
to be careful how state is serialized back into kafka.
"""
def __init__(self, kafka_hosts, produce_topic, state_topic, contact_email,
api_host_url="https://api.crossref.org/works", start_date=None,
end_date=None, is_update_filter=None):
self.api_host_url = api_host_url
self.produce_topic = produce_topic
self.state_topic = state_topic
self.contact_email = contact_email
self.is_update_filter = is_update_filter
self.kafka_config = {
'bootstrap.servers': kafka_hosts,
'delivery.report.only.error': True,
'message.max.bytes': 20000000, # ~20 MBytes; broker is ~50 MBytes
'default.topic.config': {
'request.required.acks': 'all',
},
}
self.state = HarvestState(start_date, end_date)
self.state.initialize_from_kafka(self.state_topic, self.kafka_config)
self.loop_sleep = 60*60 # how long to wait, in seconds, between date checks
self.api_batch_size = 50
self.name = "Crossref"
def params(self, date_str):
filter_param = 'from-index-date:{},until-index-date:{}'.format(
date_str, date_str)
if self.is_update_filter is not None:
filter_param += ',is_update:{}'.format(bool(self.is_update_filter))
return {
'filter': filter_param,
'rows': self.api_batch_size,
'cursor': '*',
}
def update_params(self, params, resp):
params['cursor'] = resp['message']['next-cursor']
return params
def extract_key(self, obj):
return obj['DOI'].encode('utf-8')
def fetch_date(self, date):
def fail_fast(err, msg):
if err is not None:
print("Kafka producer delivery error: {}".format(err))
print("Bailing out...")
# TODO: should it be sys.exit(-1)?
raise KafkaException(err)
producer = Producer(self.kafka_config)
date_str = date.isoformat()
params = self.params(date_str)
http_session = requests_retry_session()
http_session.headers.update({
'User-Agent': 'fatcat_tools/0.1.0 (https://fatcat.wiki; mailto:{}) python-requests'.format(
self.contact_email),
})
count = 0
while True:
http_resp = http_session.get(self.api_host_url, params=params)
if http_resp.status_code == 503:
# crude backoff; now redundant with session exponential
# backoff, but allows for longer backoff/downtime on remote end
print("got HTTP {}, pausing for 30 seconds".format(http_resp.status_code))
# keep kafka producer connection alive
producer.poll(0)
time.sleep(30.0)
continue
http_resp.raise_for_status()
resp = http_resp.json()
items = self.extract_items(resp)
count += len(items)
print("... got {} ({} of {}), HTTP fetch took {}".format(len(items), count,
self.extract_total(resp), http_resp.elapsed))
#print(json.dumps(resp))
for work in items:
producer.produce(
self.produce_topic,
json.dumps(work).encode('utf-8'),
key=self.extract_key(work),
on_delivery=fail_fast)
producer.poll(0)
if len(items) < self.api_batch_size:
break
params = self.update_params(params, resp)
producer.flush()
def extract_items(self, resp):
return resp['message']['items']
def extract_total(self, resp):
return resp['message']['total-results']
def run(self, continuous=False):
while True:
current = self.state.next(continuous)
if current:
print("Fetching DOIs updated on {} (UTC)".format(current))
self.fetch_date(current)
self.state.complete(current,
kafka_topic=self.state_topic,
kafka_config=self.kafka_config)
continue
if continuous:
print("Sleeping {} seconds...".format(self.loop_sleep))
time.sleep(self.loop_sleep)
else:
break
print("{} DOI ingest caught up".format(self.name))
class HarvestDataciteWorker(HarvestCrossrefWorker):
"""
datacite has a REST API as well as OAI-PMH endpoint.
have about 8 million
bulk export notes: https://github.com/datacite/datacite/issues/188
fundamentally, very similar to crossref. don't have a scrape... maybe
could/should use this script for that, and dump to JSON?
"""
def __init__(self, kafka_hosts, produce_topic, state_topic, contact_email,
api_host_url="https://api.datacite.org/works",
start_date=None, end_date=None):
super().__init__(kafka_hosts=kafka_hosts,
produce_topic=produce_topic,
state_topic=state_topic,
api_host_url=api_host_url,
contact_email=contact_email,
start_date=start_date,
end_date=end_date)
# for datecite, it's "from-update-date"
self.name = "Datacite"
def params(self, date_str):
return {
'from-update-date': date_str,
'until-update-date': date_str,
'page[size]': self.api_batch_size,
'page[number]': 1,
}
def extract_items(self, resp):
return resp['data']
def extract_total(self, resp):
return resp['meta']['total']
def extract_key(self, obj):
return obj['attributes']['doi'].encode('utf-8')
def update_params(self, params, resp):
params['page[number]'] = resp['meta']['page'] + 1
return params
|
