import io import json import mrjob import pytest import struct import responses import happybase_mock from unittest import mock from extraction_cdx_grobid import MRExtractCdxGrobid FAKE_PDF_BYTES = b"%PDF SOME JUNK" + struct.pack("!q", 112853843) @pytest.fixture def job(): """ Note: this mock only seems to work with job.run_mapper(), not job.run(); the later results in a separate instantiation without the mock? """ job = MRExtractCdxGrobid(['--no-conf', '-']) conn = happybase_mock.Connection() conn.create_table('wbgrp-journal-extract-test', {'file': {}, 'grobid0': {}, 'f': {}}) job.hb_table = conn.table('wbgrp-journal-extract-test') return job @mock.patch('extraction_cdx_grobid.MRExtractCdxGrobid.fetch_warc_content', return_value=(FAKE_PDF_BYTES, None)) @responses.activate def test_mapper_lines(mock_fetch, job): with open('tests/files/23b29ea36382680716be08fc71aa81bd226e8a85.xml', 'r') as f: real_tei_xml = f.read() responses.add(responses.POST, 'http://localhost:8070/api/processFulltextDocument', status=200, body=real_tei_xml, content_type='text/xml') raw = io.BytesIO(b""" com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 301 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz eu,eui,cadmus)/bitstream/handle/1814/36635/rscas_2015_03.pdf;jsessionid=761393014319a39f40d32ae3eb3a853f?sequence=1 20170705062202 http://cadmus.eui.eu/bitstream/handle/1814/36635/RSCAS_2015_03.pdf%3Bjsessionid%3D761393014319A39F40D32AE3EB3A853F?sequence%3D1 application/PDF 200 MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J - - 854156 328850624 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz com,pbworks,educ333b)/robots.txt 20170705063311 http://educ333b.pbworks.com/robots.txt text/plain 200 6VAUYENMOU2SK2OWNRPDD6WTQTECGZAD - - 638 398190140 CITESEERX-CRAWL-2017-06-20-20170705062707827-00049-00058-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705063158203-00053-31209~wbgrp-svc284.us.archive.org~8443.warc.gz """) output = io.BytesIO() job.sandbox(stdin=raw, stdout=output) job.run_mapper() # for debugging tests #print(output.getvalue().decode('utf-8')) #print(list(job.hb_table.scan())) # wayback gets FETCH 1x times mock_fetch.assert_called_once_with( "CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz", 328850624, 854156) # grobid gets POST 1x times assert len(responses.calls) == 1 # HBase assert job.hb_table.row(b'1') == {} # HTTP 301 assert job.hb_table.row(b'sha1:3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ') == {} # valid assert job.hb_table.row(b'sha1:MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J') != {} # text/plain assert job.hb_table.row(b'sha1:6VAUYENMOU2SK2OWNRPDD6WTQTECGZAD') == {} # Saved extraction info row = job.hb_table.row(b'sha1:MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J') assert struct.unpack("!q", row[b'file:size'])[0] == len(FAKE_PDF_BYTES) assert row[b'file:mime'] == b"application/pdf" assert struct.unpack("!q", row[b'grobid0:status_code'])[0] == 200 # TODO: assert row[b'grobid0:quality'] == None status = json.loads(row[b'grobid0:status'].decode('utf-8')) assert type(status) == type(dict()) assert row[b'grobid0:tei_xml'].decode('utf-8') == real_tei_xml tei_json = json.loads(row[b'grobid0:tei_json'].decode('utf-8')) metadata = json.loads(row[b'grobid0:metadata'].decode('utf-8')) assert tei_json['title'] == metadata['title'] assert 'body' in tei_json assert 'body' not in metadata def test_parse_cdx_invalid(job): print("valid") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert status is None print("space-prefixed line") raw = " com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert info is None assert status['status'] == "invalid" assert 'prefix' in status['reason'] print("commented line") raw = "#com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert info is None assert status['status'] == "invalid" assert 'prefix' in status['reason'] print("wrong column count") raw = "a b c d" info, status = job.parse_line(raw) assert info is None assert status['status'] == "invalid" assert 'parse' in status['reason'] print("missing mimetype") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf - 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert info is None print(status) assert status['status'] == "invalid" assert 'parse' in status['reason'] print("HTTP status") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 501 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert info is None assert status['status'] == "invalid" print("datetime") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 501 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.parse_line(raw) assert info is None assert status['status'] == "invalid" def test_parse_cdx_skip(job): job.mapper_init() print("warc format") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.mapper(None, raw).__next__() assert info is None assert status['status'] == "skip" assert 'WARC' in status['reason'] print("mimetype") raw = "com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf text/html 200 3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz" info, status = job.mapper(None, raw).__next__() assert info is None assert status['status'] == "skip" assert 'mimetype' in status['reason'] @mock.patch('extraction_cdx_grobid.MRExtractCdxGrobid.fetch_warc_content', return_value=(FAKE_PDF_BYTES, None)) @responses.activate def test_grobid_503(mock_fetch, job): status = b'{"status": "done broke due to 503"}' responses.add(responses.POST, 'http://localhost:8070/api/processFulltextDocument', status=503, body=status) raw = io.BytesIO(b"""com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz""") output = io.BytesIO() job.sandbox(stdin=raw, stdout=output) job.run_mapper() row = job.hb_table.row(b'sha1:ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ') status = json.loads(row[b'grobid0:status'].decode('utf-8')) assert json.loads(row[b'grobid0:status'].decode('utf-8')) == status @mock.patch('extraction_cdx_grobid.MRExtractCdxGrobid.fetch_warc_content', return_value=(FAKE_PDF_BYTES, None)) @responses.activate def test_grobid_not_xml(mock_fetch, job): payload = b'this is not XML' responses.add(responses.POST, 'http://localhost:8070/api/processFulltextDocument', status=200, body=payload) raw = io.BytesIO(b"""com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz""") output = io.BytesIO() job.sandbox(stdin=raw, stdout=output) job.run_mapper() row = job.hb_table.row(b'sha1:ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ') assert struct.unpack("!q", row[b'grobid0:status_code'])[0] == 200 assert row[b'grobid0:tei_xml'] == payload assert b'grobid0:tei_json' not in row @mock.patch('extraction_cdx_grobid.MRExtractCdxGrobid.fetch_warc_content', return_value=(FAKE_PDF_BYTES, None)) @responses.activate def test_grobid_invalid_connection(mock_fetch, job): status = b'{"status": "done broke"}' job.options.grobid_uri = 'http://host.invalid:8070/api/processFulltextDocument' raw = io.BytesIO(b"""com,sagepub,cep)/content/28/9/960.full.pdf 20170705062200 http://cep.sagepub.com/content/28/9/960.full.pdf application/pdf 200 ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ - - 401 313356621 CITESEERX-CRAWL-2017-06-20-20170705061647307-00039-00048-wbgrp-svc284/CITESEERX-CRAWL-2017-06-20-20170705062052659-00043-31209~wbgrp-svc284.us.archive.org~8443.warc.gz""") output = io.BytesIO() job.sandbox(stdin=raw, stdout=output) #with pytest.raises... job.run_mapper() assert job.hb_table.row(b'sha1:ABCDEF12345Q2MSVX7XZKYAYSCX5QBYJ') == {} # TODO: failure to fetch from wayback