backfill: sha1 prefix, cluster example

3 files changed, 19 insertions, 8 deletions

diff --git a/backfill/README.md b/backfill/README.md
index 90b4ba7..6af8f33 100644
--- a/backfill/README.md
+++ b/backfill/README.md
@@ -3,6 +3,10 @@ Run tests:
 
     pipenv run python -m pytest
 
-Run locally on a file:
+An example actually connecting to HBase from a local machine, with thrift
+running on a devbox:
+
+    ./backfill_hbase_from_cdx.py tests/files/example.cdx \
+        --hbase-table wbgrp-journal-extract-0-qa \
+        --hbase-host bnewbold-dev.us.archive.org
 
-    ./backfill_hbase_from_cdx.py tests/files/example.cdx
diff --git a/backfill/backfill_hbase_from_cdx.py b/backfill/backfill_hbase_from_cdx.py
index d14dd92..8008761 100755
--- a/backfill/backfill_hbase_from_cdx.py
+++ b/backfill/backfill_hbase_from_cdx.py
@@ -74,6 +74,8 @@ def transform_line(raw_cdx):
     if '-' in (surt, dt, url, mime, http_status, key, c_size, offset, warc):
         return None
 
+    key = "sha1:{}".format(key)
+
     info = dict(surt=surt, dt=dt, url=url, c_size=c_size, offset=offset,
         warc=warc)
 
@@ -92,7 +94,7 @@ def test_transform_line():
 
     raw = "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf 20170828233154 https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf application/pdf 200 WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G - - 210251 931661233 SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828231135742-00000-00009-wbgrp-svc284/SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828232253025-00005-3480~wbgrp-svc284.us.archive.org~8443.warc.gz"
     correct = {
-        'key': "WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G",
+        'key': "sha1:WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G",
         'file:mime': "application/pdf",
         'file:cdx': {
             'surt': "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf",
@@ -150,7 +152,9 @@ class MRCDXBackfillHBase(MRJob):
         if self.hb_table is None:
             try:
                 host = self.options.hbase_host
-                hb_conn = happybase.Connection(host=host)
+                # TODO: make these configs accessible from... mrconf.cfg?
+                hb_conn = happybase.Connection(host=host, transport="framed",
+                    protocol="compact")
             except Exception as err:
                 raise Exception("Couldn't connect to HBase using host: {}".format(host))
             self.hb_table = hb_conn.table(self.options.hbase_table)
diff --git a/backfill/tests/test_backfill_hbase_from_cdx.py b/backfill/tests/test_backfill_hbase_from_cdx.py
index d8277be..9af5b05 100644
--- a/backfill/tests/test_backfill_hbase_from_cdx.py
+++ b/backfill/tests/test_backfill_hbase_from_cdx.py
@@ -1,3 +1,6 @@
+"""
+TODO: could probably refactor to use unittest.mock.patch('happybase')
+"""
 
 import io
 import json
@@ -33,13 +36,13 @@ com,pbworks,educ333b)/robots.txt 20170705063311 http://educ333b.pbworks.com/robo
 
     assert job.hb_table.row(b'1') == {}
     # HTTP 301
-    assert job.hb_table.row(b'3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ') == {}
+    assert job.hb_table.row(b'sha1:3I42H3S6NNFQ2MSVX7XZKYAYSCX5QBYJ') == {}
     # valid
-    assert job.hb_table.row(b'MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J') != {}
+    assert job.hb_table.row(b'sha1:MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J') != {}
     # text/plain
-    assert job.hb_table.row(b'6VAUYENMOU2SK2OWNRPDD6WTQTECGZAD') == {}
+    assert job.hb_table.row(b'sha1:6VAUYENMOU2SK2OWNRPDD6WTQTECGZAD') == {}
 
-    row = job.hb_table.row(b'MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J')
+    row = job.hb_table.row(b'sha1:MPCXVWMUTRUGFP36SLPHKDLY6NGU4S3J')
     assert row[b'file:mime'] == b"application/pdf"
 
     file_cdx = json.loads(row[b'file:cdx'].decode('utf-8'))