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
|
import re
import sys
import csv
import json
import time
import requests
import itertools
import datetime
from pykafka import KafkaClient
from fatcat_tools.workers.worker_common import most_recent_message
# Skip pylint due to:
# AttributeError: 'NoneType' object has no attribute 'scope'
# in 'astroid/node_classes.py'
# pylint: skip-file
DATE_FMT = "%Y-%m-%d"
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.kafka = KafkaClient(hosts=kafka_hosts, broker_version="1.0.0")
self.is_update_filter = is_update_filter
# these are both optional, and should be datetime.date
self.start_date = start_date
self.end_date = end_date
self.loop_sleep = 60*60 # how long to wait, in seconds, between date checks
self.api_batch_size = 50
# for crossref, it's "from-index-date"
self.name = "Crossref"
def get_latest_date(self):
state_topic = self.kafka.topics[self.state_topic]
latest = most_recent_message(state_topic)
if latest:
latest = datetime.datetime.strptime(latest.decode('utf-8'), DATE_FMT).date()
print("Latest date found: {}".format(latest))
return latest
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 fetch_date(self, date):
state_topic = self.kafka.topics[self.state_topic]
produce_topic = self.kafka.topics[self.produce_topic]
date_str = date.strftime(DATE_FMT)
params = self.params(date_str)
headers = {
'User-Agent': 'fatcat_tools/0.1.0 (https://fatcat.wiki; mailto:{}) python-requests'.format(self.contact_email),
}
count = 0
with produce_topic.get_producer() as producer:
while True:
http_resp = requests.get(self.api_host_url, params, headers=headers)
if http_resp.status_code == 503:
# crud backoff
print("got HTTP {}, pausing for 30 seconds".format(http_resp.status_code))
time.sleep(30.0)
continue
assert http_resp.status_code == 200
resp = http_resp.json()
items = self.extract_items(resp)
count += len(items)
print("... got {} ({} of {}) in {}".format(len(items), count,
self.extract_total(resp), http_resp.elapsed))
#print(json.dumps(resp))
for work in items:
producer.produce(json.dumps(work).encode('utf-8'))
if len(items) < self.api_batch_size:
break
params = self.update_params(params, resp)
# record our completion state
with state_topic.get_sync_producer() as producer:
producer.produce(date.strftime(DATE_FMT).encode('utf-8'))
def extract_items(self, resp):
return resp['message']['items']
def extract_total(self, resp):
return resp['message']['total-results']
def run_once(self):
today_utc = datetime.datetime.utcnow().date()
if self.start_date is None:
self.start_date = self.get_latest_date()
if self.start_date:
# if we are continuing, start day after last success
self.start_date = self.start_date + datetime.timedelta(days=1)
if self.start_date is None:
# bootstrap to yesterday (don't want to start on today until it's over)
self.start_date = datetime.datetime.utcnow().date()
if self.end_date is None:
# bootstrap to yesterday (don't want to start on today until it's over)
self.end_date = today_utc - datetime.timedelta(days=1)
print("Harvesting from {} through {}".format(self.start_date, self.end_date))
current = self.start_date
while current <= self.end_date:
print("Fetching DOIs updated on {} (UTC)".format(current))
self.fetch_date(current)
current += datetime.timedelta(days=1)
print("{} DOI ingest caught up through {}".format(self.name, self.end_date))
return self.end_date
def run_loop(self):
while True:
last = self.run_once()
self.start_date = last
self.end_date = None
print("Sleeping {} seconds...".format(self.loop_sleep))
time.sleep(self.loop_sleep())
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 update_params(self, params, resp):
params['page[number]'] = resp['meta']['page'] + 1
return params
|
