aboutsummaryrefslogtreecommitdiffstats
path: root/python/fatcat_tools/harvest/pubmed.py
blob: fd4af1f4f61430e3c2859e45ebaacd2ff426e3c1 (plain)
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
"""
Pubmed harvest via FTP.

Assumptions:

* fixed hostname and directory structure
* XML files are gzip compressed
* accompanying HTML files contain correct dates
"""

import collections
import ftplib
import gzip
import io
import os
import re
import shutil
import socket
import sys
import tempfile
import time
import xml.etree.ElementTree as ET
import zlib
from urllib.parse import urlparse

import dateparser
from bs4 import BeautifulSoup
from confluent_kafka import KafkaException, Producer

from .harvest_common import HarvestState


class PubmedFTPWorker:
    """
    Access Pubmed FTP for daily updates.

    * Server directory: ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles
    * Docs: ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/README.txt

    Daily Update Files (02/2020)
    ----------------------------
    Each day, NLM produces update files that include new, revised and deleted
    citations. The first Update file to be loaded after loading the complete
    set of 2019 MEDLINE/PubMed Baseline files is pubmed20n1016.xml.

    Usually, three files per update, e.g.:

    * ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/pubmed20n1016_stats.html
    * ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/pubmed20n1016.xml.gz
    * ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/pubmed20n1016.xml.gz.md5

    Currently (02/2020) the HTML contains the date.

        <html>
        <head><title></title></head>
        <body>
        <h4>Filename: pubmed20n1019.xml -- Created: Wed Dec 18 14:31:09 EST 2019</h4>
        <table cellspacing="0" cellpadding="0" border="0" width="300">
        <tr>

    """
    def __init__(self, kafka_hosts, produce_topic, state_topic, start_date=None, end_date=None):
        self.name = 'Pubmed'
        self.host = 'ftp.ncbi.nlm.nih.gov'
        self.produce_topic = produce_topic
        self.state_topic = state_topic
        self.kafka_config = {
            'bootstrap.servers': kafka_hosts,
            'message.max.bytes': 20000000,  # ~20 MBytes; broker is ~50 MBytes
        }
        self.loop_sleep = 60 * 60  # how long to wait, in seconds, between date checks
        self.state = HarvestState(start_date, end_date)
        self.state.initialize_from_kafka(self.state_topic, self.kafka_config)
        self.producer = self._kafka_producer()
        self.date_file_map = None

    def _kafka_producer(self):
        def fail_fast(err, msg):
            if err is not None:
                print("Kafka producer delivery error: {}".format(err), file=sys.stderr)
                print("Bailing out...", file=sys.stderr)
                # TODO: should it be sys.exit(-1)?
                raise KafkaException(err)

        self._kafka_fail_fast = fail_fast

        producer_conf = self.kafka_config.copy()
        producer_conf.update({
            'delivery.report.only.error': True,
            'default.topic.config': {
                'request.required.acks': -1,  # all brokers must confirm
            },
        })
        return Producer(producer_conf)

    def fetch_date(self, date):
        """
        Fetch file or files for a given date and feed Kafka one article per
        message. If the fetched XML does not contain a PMID an exception is
        raised.

        If no date file mapping is found, this will fail.
        """
        if self.date_file_map is None:
            raise ValueError("cannot fetch date without date file mapping")

        date_str = date.strftime('%Y-%m-%d')
        paths = self.date_file_map.get(date_str)
        if paths is None:
            print("WARN: no pubmed update for this date: {} (UTC), available dates were: {}".format(date_str, self.date_file_map), file=sys.stderr)
            return False

        count = 0
        for path in paths:
            # Fetch and decompress file.
            url = "ftp://{}{}".format(self.host, path)
            filename = ftpretr(url)
            with tempfile.NamedTemporaryFile(prefix='fatcat-ftp-tmp-', delete=False) as decomp:
                try:
                    gzf = gzip.open(filename)
                    shutil.copyfileobj(gzf, decomp)
                except zlib.error as exc:
                    print('[skip] retrieving {} failed with {} (maybe empty, missing or broken gzip)'.format(
                        url, exc), file=sys.stderr)
                    continue

            # Here, blob is the unparsed XML; we peek into it to use PMID as
            # message key. We need streaming, since some updates would consume
            # GBs otherwise.
            # WARNING: Parsing foreign XML exposes us at some
            # https://docs.python.org/3/library/xml.html#xml-vulnerabilities
            # here.
            for blob in xmlstream(decomp.name, 'PubmedArticle', encoding='utf-8'):
                soup = BeautifulSoup(blob, 'xml')
                pmid = soup.find('PMID')
                if pmid is None:
                    raise ValueError("no PMID found, please adjust identifier extraction")
                count += 1
                if count % 50 == 0:
                    print("... up to {}".format(count), file=sys.stderr)
                self.producer.produce(self.produce_topic, blob, key=pmid.text, on_delivery=self._kafka_fail_fast)

            self.producer.flush()
            os.remove(filename)
            os.remove(decomp.name)

        return True

    def run(self, continuous=False):
        while True:
            self.date_file_map = generate_date_file_map(host=self.host)
            if len(self.date_file_map) == 0:
                raise ValueError("map from dates to files should not be empty, maybe the HTML changed?")

            current = self.state.next_span(continuous)
            if current:
                print("Fetching citations updated on {} (UTC)".format(current), file=sys.stderr)
                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("{} FTP ingest caught up".format(self.name))


def generate_date_file_map(host='ftp.ncbi.nlm.nih.gov'):
    """
    Generate a DefaultDict[string, set] mapping dates to absolute filepaths on
    the server (mostly we have one file, but sometimes more).

    Example: {"2020-01-02": set(["/pubmed/updatefiles/pubmed20n1016.xml.gz"]), ...}
    """
    mapping = collections.defaultdict(set)
    pattern = re.compile(r'Filename: ([^ ]*.xml) -- Created: ([^<]*)')
    ftp = ftplib.FTP(host)
    ftp.login()
    filenames = ftp.nlst('/pubmed/updatefiles')
    retries, retry_delay = 10, 60

    for name in filenames:
        if not name.endswith('.html'):
            continue
        sio = io.StringIO()
        for i in range(retries):
            try:
                # Previously, from 2020-12-14 to 2021-06-30 everything worked
                # fine, then a request for
                # /pubmed/updatefiles/pubmed21n1328_stats.html would always
                # fail with an EOFError, or when retried with a 32
                # BrokenPipeError. Suspecting the server for some unknown
                # reason dropped the connection.
                #
                # Using a fresh client, the exact same file would work just
                # fine. So when we retry, we setup a new client here as well.
                if i > 0:
                    ftp = ftplib.FTP(host)
                    ftp.login()
                    sio.truncate(0)
                ftp.retrlines('RETR {}'.format(name), sio.write)
            except (EOFError, ftplib.error_temp, socket.gaierror, BrokenPipeError) as exc:
                print("ftp retr on {} failed with {} ({}) ({} retries left)".format(
                    name, exc, type(exc), retries - (i + 1)), file=sys.stderr)
                if i + 1 == retries:
                    raise
                else:
                    time.sleep(retry_delay)
            else:
                break
        contents = sio.getvalue()
        match = pattern.search(contents)
        if match is None:
            print('pattern miss in {} on: {}, may need to adjust pattern: {}'.format(name, contents, pattern), file=sys.stderr)
            continue
        filename, filedate = match.groups()  # ('pubmed20n1017.xml', 'Tue Dec 17 15:23:32 EST 2019')
        date = dateparser.parse(filedate)
        fullpath = '/pubmed/updatefiles/{}.gz'.format(filename)
        date_str = date.strftime('%Y-%m-%d')
        mapping[date_str].add(fullpath)
        print('added entry for {}: {}'.format(date_str, fullpath), file=sys.stderr)

    print('generated date-file mapping for {} dates'.format(len(mapping)), file=sys.stderr)
    return mapping


def ftpretr(url):
    """
    Note: This might move into a generic place in the future.

    Fetch (RETR) a remote file given by its URL (e.g.
    "ftp://ftp.ncbi.nlm.nih.gov/pubmed/updatefiles/pubmed20n1016.xml.gz") to a
    local temporary file. Returns the name of the local, closed temporary file.

    It is the reponsibility of the caller to cleanup the temporary file.
    """
    parsed = urlparse(url)
    server, path = parsed.netloc, parsed.path
    ftp = ftplib.FTP(server)
    ftp.login()
    with tempfile.NamedTemporaryFile(prefix='fatcat-ftp-tmp-', delete=False) as f:
        print('retrieving {} from {} to {} ...'.format(path, server, f.name), file=sys.stderr)
        ftp.retrbinary('RETR %s' % path, f.write)
    ftp.close()
    return f.name


def xmlstream(filename, tag, encoding='utf-8'):
    """
    Note: This might move into a generic place in the future.

    Given a path to an XML file and a tag name (without namespace), stream
    through the XML and yield elements denoted by tag as string.

    for snippet in xmlstream("sample.xml", "sometag"):
        print(len(snippet))

    Known vulnerabilities: https://docs.python.org/3/library/xml.html#xml-vulnerabilities
    """
    def strip_ns(tag):
        if not '}' in tag:
            return tag
        return tag.split('}')[1]

    # https://stackoverflow.com/a/13261805, http://effbot.org/elementtree/iterparse.htm
    context = iter(ET.iterparse(filename, events=(
        'start',
        'end',
    )))
    try:
        _, root = next(context)
    except StopIteration:
        return

    for event, elem in context:
        if not strip_ns(elem.tag) == tag or event == 'start':
            continue

        yield ET.tostring(elem, encoding=encoding)
        root.clear()