initial work on parquet support

[mediawiki_dump_tools.git] / wikiq

diff --git a/wikiq b/wikiq
index 6a1ada5cccee93edc9ae4c96f374542b1b4ba118..c2a2f124f7f5b8f86c43a1480fccb4c761043eae 100755 (executable)
--- a/wikiq
+++ b/wikiq
@@ -24,6 +24,13 @@ PERSISTENCE_RADIUS=7
 from deltas import SequenceMatcher
 from deltas import SegmentMatcher
 
 from deltas import SequenceMatcher
 from deltas import SegmentMatcher
 

+from dataclasses import dataclass
+import pandas as pd
+import pyarrow as pa
+import pyarrow.parquet as pq
+
+from typing import List
+

 class PersistMethod:
     none = 0
     sequence = 1
 class PersistMethod:
     none = 0
     sequence = 1


@@ -34,55 +41,6 @@ def calculate_persistence(tokens_added):
     return(sum([(len(x.revisions)-1) for x in tokens_added]),
            len(tokens_added))
 
     return(sum([(len(x.revisions)-1) for x in tokens_added]),
            len(tokens_added))
 

-def matchmake(scanned_content, rev_data, regex, label):
-    p = re.compile(regex)
-
-    temp_dict = {}
-    # if there are named capture groups in the regex
-    if bool(p.groupindex):
-        capture_groups = list(p.groupindex.keys())
-
-        # initialize the {capture_group_name:list} for each capture group
-        for cap_group in capture_groups:
-            temp_dict["{}_{}".format(label, cap_group)] = []
-
-        # if there are matches of some sort in this revision content, fill the lists for each cap_group
-        if p.search(scanned_content) is not None:
-            m = re.finditer(p,scanned_content)
-            matchobjects = list(m)
-
-            for cap_group in capture_groups:
-                temp_list = []
-                for match in matchobjects:
-                    # we only want to add the match for the capture group if the match is not None
-                    if match.group(cap_group) != None:
-                        temp_list.append(match.group(cap_group))
-
-                # if temp_list of matches is empty just make that column None
-                if len(temp_list)==0:
-                    temp_dict["{}_{}".format(label, cap_group)] = None
-                # else we put in the list we made in the for-loop above
-                else:
-                    temp_dict["{}_{}".format(label, cap_group)] = ', '.join(temp_list)
-        
-        # there are no matches at all in this revision content, we default values to None
-        else:
-            for cap_group in capture_groups:
-                temp_dict["{}_{}".format(label, cap_group)] = None
-
-    # there are no capture groups, we just search for all the matches of the regex
-    else:
-        #given that there are matches to be made
-        if p.search(scanned_content) is not None:
-            m = p.findall(scanned_content)
-            temp_dict[label] = ', '.join(m)
-        else:
-            temp_dict[label] = None    
-    # update rev_data with our new columns
-    rev_data.update(temp_dict)
-#    print(rev_data.keys())
-    return rev_data
-

 
 class WikiqIterator():
     def __init__(self, fh, collapse_user=False):
 
 class WikiqIterator():
     def __init__(self, fh, collapse_user=False):


@@ -176,31 +134,155 @@ class WikiqPage():
     def __next__(self):
         return next(self.__revisions)
 
     def __next__(self):
         return next(self.__revisions)
 

+
+class RegexPair(object):
+    def __init__(self, pattern, label):
+        self.pattern = re.compile(pattern)
+        self.label = label
+        self.has_groups = bool(self.pattern.groupindex)
+        if self.has_groups:
+            self.capture_groups = list(self.pattern.groupindex.keys())
+            
+    def _make_key(self, cap_group):
+        return ("{}_{}".format(self.label, cap_group))
+
+    def matchmake(self, content, rev_data):
+        
+        temp_dict = {}
+        # if there are named capture groups in the regex
+        if self.has_groups:
+
+            # if there are matches of some sort in this revision content, fill the lists for each cap_group
+            if self.pattern.search(content) is not None:
+                m = self.pattern.finditer(content)
+                matchobjects = list(m)
+
+                for cap_group in self.capture_groups:
+                    key = self._make_key(cap_group)
+                    temp_list = []
+                    for match in matchobjects:
+                        # we only want to add the match for the capture group if the match is not None
+                        if match.group(cap_group) != None:
+                            temp_list.append(match.group(cap_group))
+
+                    # if temp_list of matches is empty just make that column None
+                    if len(temp_list)==0:
+                        temp_dict[key] = None
+                    # else we put in the list we made in the for-loop above
+                    else:
+                        temp_dict[key] = ', '.join(temp_list)
+
+            # there are no matches at all in this revision content, we default values to None
+            else:
+                for cap_group in self.capture_groups:
+                    key = self._make_key(cap_group)
+                    temp_dict[key] = None
+
+        # there are no capture groups, we just search for all the matches of the regex
+        else:
+            #given that there are matches to be made
+            if type(content) in(str, bytes):
+                if self.pattern.search(content) is not None:
+                    m = self.pattern.findall(content)
+                    temp_dict[self.label] = ', '.join(m)
+                else:
+                    temp_dict[self.label] = None
+
+        # update rev_data with our new columns
+        rev_data.update(temp_dict)
+        return rev_data
+
+@dataclass
+class RevData():
+    revid: int
+    date_time: datetime
+    articleid: int
+    editorid: int
+    title: str
+    namespace: int
+    deleted: bool
+    text_chars: int
+    revert: bool
+    reverteds: list[bool]
+    sha1: str
+    text_chars: int
+    revert: bool
+    reverteds: list[int]
+    minor: bool
+    editor: str
+    anon: bool
+    collapsed_revs:int
+    token_revs:int
+    tokens_added:int
+    tokens_removed:int
+    tokens_window:int
+

 class WikiqParser():
 class WikiqParser():

-    def __init__(self, input_file, output_file, regex_match_revision, regex_match_comment, regex_revision_label, regex_comment_label, collapse_user=False, persist=None, urlencode=False, namespaces = None, revert_radius=15):
+    def __init__(self, input_file, output_file, regex_match_revision, regex_match_comment, regex_revision_label, regex_comment_label, collapse_user=False, persist=None, urlencode=False, namespaces = None, revert_radius=15, output_parquet=True, parquet_buffer_size=2000):

         """ 
         Parameters:
            persist : what persistence method to use. Takes a PersistMethod value
         """
         """ 
         Parameters:
            persist : what persistence method to use. Takes a PersistMethod value
         """

-

         self.input_file = input_file
         self.input_file = input_file

-        self.output_file = output_file
+

         self.collapse_user = collapse_user
         self.persist = persist
         self.printed_header = False
         self.namespaces = []
         self.urlencode = urlencode
         self.revert_radius = revert_radius
         self.collapse_user = collapse_user
         self.persist = persist
         self.printed_header = False
         self.namespaces = []
         self.urlencode = urlencode
         self.revert_radius = revert_radius

-        self.regex_match_revision = regex_match_revision
-        self.regex_revision_label = regex_revision_label
-        self.regex_match_comment = regex_match_comment
-        self.regex_comment_label = regex_comment_label
+        
+        self.parquet_buffer = []
+        self.parquet_buffer_size = parquet_buffer_size

 
         if namespaces is not None:
             self.namespace_filter = set(namespaces)
         else:
             self.namespace_filter = None
 
 
         if namespaces is not None:
             self.namespace_filter = set(namespaces)
         else:
             self.namespace_filter = None
 

+        self.regex_schemas = []
+        self.regex_revision_pairs = self.make_matchmake_pairs(regex_match_revision, regex_revision_label)
+        self.regex_comment_pairs = self.make_matchmake_pairs(regex_match_comment, regex_comment_label)
+
+
+        if output_parquet is True:
+            self.output_parquet = True
+            self.pq_writer = None
+            self.output_file = output_file
+        else:
+            self.output_file = open(output_file,'w')
+
+
+    def make_matchmake_pairs(self, patterns, labels):
+        if (patterns is not None and labels is not None) and \
+           (len(patterns) == len(labels)):
+            result = []
+            for pattern, label in zip(patterns, labels):
+                result.append(RegexPair(pattern, label))
+                self.regex_schemas.append(pa.field(label, pa.list_(pa.string())))
+
+            return result
+        elif (patterns is None and labels is None):
+            return []
+        else:
+            sys.exit('Each regular expression *must* come with a corresponding label and vice versa.')
+
+    def matchmake(self, rev, rev_data):
+        rev_data = self.matchmake_revision(rev.text, rev_data)
+        rev_data = self.matchmake_comment(rev.comment, rev_data)
+        return rev_data
+
+    def matchmake_revision(self, text, rev_data):
+        return self.matchmake_pairs(text, rev_data, self.regex_revision_pairs)
+
+    def matchmake_comment(self, comment, rev_data):
+        return self.matchmake_pairs(comment, rev_data, self.regex_comment_pairs)
+
+    def matchmake_pairs(self, text, rev_data, pairs):
+        for pair in pairs:
+            rev_data = pair.matchmake(text, rev_data)
+        return rev_data
+

     def __get_namespace_from_title(self, title):
         default_ns = None
 
     def __get_namespace_from_title(self, title):
         default_ns = None
 


@@ -243,10 +325,8 @@ class WikiqParser():
                 if namespace not in self.namespace_filter:
                     continue
 
                 if namespace not in self.namespace_filter:
                     continue
 

-#            print(self.revert_radius)

             rev_detector = mwreverts.Detector(radius = self.revert_radius)
 
             rev_detector = mwreverts.Detector(radius = self.revert_radius)
 

-

             if self.persist != PersistMethod.none:
                 window = deque(maxlen=PERSISTENCE_RADIUS)
 
             if self.persist != PersistMethod.none:
                 window = deque(maxlen=PERSISTENCE_RADIUS)
 


@@ -267,50 +347,17 @@ class WikiqParser():
             for rev in page:
                 
                 # initialize rev_data
             for rev in page:
                 
                 # initialize rev_data

-                rev_data = {}
-
-                # if the command line args only gave a label (and no regular expression is given)
-                if (self.regex_revision_label != None and self.regex_match_revision == None) or (self.regex_comment_label != None and self.regex_match_comment == None):
-                    sys.exit('The given regex label(s) has no corresponding regex to search for.')
-                
-                # if there's anything in the list of regex_match_revision
-                if self.regex_match_revision is not None:
-                    if (self.regex_revision_label == None) or (len(self.regex_match_revision) != len(self.regex_revision_label)):
-                        sys.exit('Each regular expression *must* come with a corresponding label and vice versa.')
-                    
-                    # initialize and construct the list of regex-label tuples
-                    pairs = []
-                    for i in range(0,len(self.regex_match_revision)):
-                        pairs.append((self.regex_match_revision[i], self.regex_revision_label[i]))
-
-                    # for each regex/label pair, we now run matchmake to check and output columns
-                    for pair in pairs:
-                        # pair[0] corresponds to the regex, pair[1] to the label
-                        rev_data = matchmake(rev.text, rev_data, pair[0], pair[1])
-                
-                # if there's anything in the list of regex_match_comment
-                if self.regex_match_comment is not None:
-                    if (self.regex_comment_label == None) or (len(self.regex_match_comment) != len(self.regex_comment_label)):
-                        sys.exit('Each regular expression *must* come with a corresponding label and vice versa.')
-                    
-                    # initialize and construct the list of regex-label tuples
-                    pairs = []
-                    for i in range(0,len(self.regex_match_comment)):
-                        pairs.append((self.regex_match_comment[i], self.regex_comment_label[i]))
-
-                    # for each regex/label pair, we now run matchmake to check and output columns
-                    for pair in pairs:
-                        # pair[0] corresponds to the regex, pair[1] to the label
-                        rev_data = matchmake(rev.comment, rev_data, pair[0], pair[1])
-
-                # we fill out the rest of the data structure now
-                rev_data['revid'] = rev.id
-                rev_data['date_time'] = rev.timestamp.strftime('%Y-%m-%d %H:%M:%S')
-                rev_data['articleid'] = page.id
-                rev_data['editor_id'] = "" if rev.deleted.user == True or rev.user.id is None else rev.user.id
-                rev_data['title'] = '"' + page.title + '"'
-                rev_data['namespace'] = namespace
-                rev_data['deleted'] = "TRUE" if rev.deleted.text else "FALSE"
+                rev_data = {
+                    'revid':rev.id,
+                    'date_time' : rev.timestamp.strftime('%Y-%m-%d %H:%M:%S'),
+                    'articleid' : page.id,
+                    'editorid' : "" if rev.deleted.user == True or rev.user.id is None else rev.user.id,
+                    'title' : '"' + page.title + '"',
+                    'namespace' : namespace,
+                    'deleted' : "TRUE" if rev.deleted.text else "FALSE"
+                }
+
+                rev_data = self.matchmake(rev, rev_data)

 
                 # if revisions are deleted, /many/ things will be missing
                 if rev.deleted.text:
 
                 # if revisions are deleted, /many/ things will be missing
                 if rev.deleted.text:


@@ -414,7 +461,6 @@ class WikiqParser():
                     rev_data["tokens_added"] = num_tokens
                     rev_data["tokens_removed"] = len(tokens_removed)
                     rev_data["tokens_window"] = len(window)-(i+1)
                     rev_data["tokens_added"] = num_tokens
                     rev_data["tokens_removed"] = len(tokens_removed)
                     rev_data["tokens_window"] = len(window)-(i+1)

-                    

                     self.print_rev_data(rev_data)
 
             page_count += 1
                     self.print_rev_data(rev_data)
 
             page_count += 1


@@ -422,22 +468,65 @@ class WikiqParser():
         print("Done: %s revisions and %s pages." % (rev_count, page_count),
               file=sys.stderr)
 
         print("Done: %s revisions and %s pages." % (rev_count, page_count),
               file=sys.stderr)
 

+        if self.output_parquet is True:
+            self.flush_parquet_buffer()
+            self.pq_writer.close()
+
+        else:
+            output_file.close()
+
+
+    def write_parquet_row(self, rev_data):
+        if 'deleted' in rev_data.keys():
+            rev_data['deleted'] = True if rev_data['deleted'] == "TRUE" else False
+
+        if 'minor' in rev_data.keys():
+            rev_data['minor'] = True if rev_data['minor'] == "TRUE" else False
+
+
+        if 'anon' in rev_data.keys():
+            rev_data['anon'] = True if rev_data['anon'] == "TRUE" else False
+
+
+        self.parquet_buffer.append(rev_data)
+
+        if len(self.parquet_buffer) >= self.parquet_buffer_size:
+            self.flush_parquet_buffer()
+
+    def flush_parquet_buffer(self):
+        outtable = pd.DataFrame.from_records(self.parquet_buffer) 
+        outtable = pa.Table.from_pandas(outtable)                        
+        if self.pq_writer is None:
+            schema = outtable.schema
+            for regex_schema in self.regex_schemas:
+                schema.append(regex_schema)
+
+            self.pq_writer = pq.ParquetWriter(self.output_file, schema, flavor='spark')
+
+        self.pq_writer.write_table(outtable)
+

     def print_rev_data(self, rev_data):
     def print_rev_data(self, rev_data):

+
+        if self.output_parquet is False:
+            printfunc = lambda rev_data: print("\t".join(rev_data), file=self.output_file)
+        else:
+            printfunc = self.write_parquet_row
+

         # if it's the first time through, print the header
         if self.urlencode:
             for field in TO_ENCODE:
                 rev_data[field] = quote(str(rev_data[field]))
 
         if not self.printed_header:
         # if it's the first time through, print the header
         if self.urlencode:
             for field in TO_ENCODE:
                 rev_data[field] = quote(str(rev_data[field]))
 
         if not self.printed_header:

-            print("\t".join([str(k) for k in sorted(rev_data.keys())]), file=self.output_file)
+            printfunc(rev_data)

             self.printed_header = True
         
             self.printed_header = True
         

-        print("\t".join([str(v) for k, v in sorted(rev_data.items())]), file=self.output_file)
+        printfunc(rev_data)

 
 
 def open_input_file(input_filename):
     if re.match(r'.*\.7z$', input_filename):
 
 
 def open_input_file(input_filename):
     if re.match(r'.*\.7z$', input_filename):

-        cmd = ["7za", "x", "-so", input_filename, '*'] 
+        cmd = ["7za", "x", "-so", input_filename, "*.xml"] 

     elif re.match(r'.*\.gz$', input_filename):
         cmd = ["zcat", input_filename] 
     elif re.match(r'.*\.bz2$', input_filename):
     elif re.match(r'.*\.gz$', input_filename):
         cmd = ["zcat", input_filename] 
     elif re.match(r'.*\.bz2$', input_filename):


@@ -450,13 +539,19 @@ def open_input_file(input_filename):
 
     return input_file
 
 
     return input_file
 

-def open_output_file(input_filename):
-    # create a regex that creates the output filename
+def get_output_filename(input_filename, parquet = False):

     output_filename = re.sub(r'\.(7z|gz|bz2)?$', '', input_filename)
     output_filename = re.sub(r'\.xml', '', output_filename)
     output_filename = re.sub(r'\.(7z|gz|bz2)?$', '', input_filename)
     output_filename = re.sub(r'\.xml', '', output_filename)

-    output_filename = output_filename + ".tsv"
-    output_file = open(output_filename, "w")
+    if parquet is False:
+        output_filename = output_filename + ".tsv"
+    else:
+        output_filename = output_filename + ".parquet"
+    return output_filename

 
 

+def open_output_file(input_filename):
+    # create a regex that creates the output filename
+    output_filename = get_output_filename(input_filename, parquet = False)
+    output_file = open(output_filename, "w")

     return output_file
 
 parser = argparse.ArgumentParser(description='Parse MediaWiki XML database dumps into tab delimitted data.')
     return output_file
 
 parser = argparse.ArgumentParser(description='Parse MediaWiki XML database dumps into tab delimitted data.')


@@ -466,7 +561,7 @@ parser.add_argument('dumpfiles', metavar="DUMPFILE", nargs="*", type=str,
                     help="Filename of the compressed or uncompressed XML database dump. If absent, we'll look for content on stdin and output on stdout.")
 
 parser.add_argument('-o', '--output-dir', metavar='DIR', dest='output_dir', type=str, nargs=1,
                     help="Filename of the compressed or uncompressed XML database dump. If absent, we'll look for content on stdin and output on stdout.")
 
 parser.add_argument('-o', '--output-dir', metavar='DIR', dest='output_dir', type=str, nargs=1,

-                    help="Directory for output files.")
+                    help="Directory for output files. If it ends with .parquet output will be in parquet format.")

 
 parser.add_argument('-s', '--stdout', dest="stdout", action="store_true",
                     help="Write output to standard out (do not create dump file)")
 
 parser.add_argument('-s', '--stdout', dest="stdout", action="store_true",
                     help="Write output to standard out (do not create dump file)")


@@ -505,6 +600,8 @@ parser.add_argument('-CPl', '--comment-pattern-label', dest="regex_comment_label
 
 args = parser.parse_args()
 
 
 args = parser.parse_args()
 

+
+

 # set persistence method
 
 if args.persist is None:
 # set persistence method
 
 if args.persist is None:


@@ -522,6 +619,7 @@ else:
     namespaces = None
 
 if len(args.dumpfiles) > 0:
     namespaces = None
 
 if len(args.dumpfiles) > 0:

+    output_parquet = False

     for filename in args.dumpfiles:
         input_file = open_input_file(filename)
 
     for filename in args.dumpfiles:
         input_file = open_input_file(filename)
 


@@ -531,13 +629,16 @@ if len(args.dumpfiles) > 0:
         else:
             output_dir = "."
 
         else:
             output_dir = "."
 

+        if output_dir.endswith(".parquet"):
+            output_parquet = True
+

         print("Processing file: %s" % filename, file=sys.stderr)
 
         if args.stdout:
             output_file = sys.stdout
         else:
             filename = os.path.join(output_dir, os.path.basename(filename))
         print("Processing file: %s" % filename, file=sys.stderr)
 
         if args.stdout:
             output_file = sys.stdout
         else:
             filename = os.path.join(output_dir, os.path.basename(filename))

-            output_file = open_output_file(filename)
+            output_file = get_output_filename(filename, parquet = output_parquet)

 
         wikiq = WikiqParser(input_file,
                             output_file,
 
         wikiq = WikiqParser(input_file,
                             output_file,


@@ -549,13 +650,15 @@ if len(args.dumpfiles) > 0:
                             regex_match_revision = args.regex_match_revision,
                             regex_revision_label = args.regex_revision_label,
                             regex_match_comment = args.regex_match_comment,
                             regex_match_revision = args.regex_match_revision,
                             regex_revision_label = args.regex_revision_label,
                             regex_match_comment = args.regex_match_comment,

-                            regex_comment_label = args.regex_comment_label)
+                            regex_comment_label = args.regex_comment_label,
+                            output_parquet=output_parquet)

 
 

+        print(wikiq.output_parquet)

         wikiq.process()
 
         # close things 
         input_file.close()
         wikiq.process()
 
         # close things 
         input_file.close()

-        output_file.close()
+

 else:
     wikiq = WikiqParser(sys.stdin,
                         sys.stdout,
 else:
     wikiq = WikiqParser(sys.stdin,
                         sys.stdout,