From: Mike O'Connor Date: Mon, 13 Apr 2009 20:01:56 +0000 (-0400) Subject: possibly working multithread contents writer that has udebs disabled X-Git-Url: https://git.decadent.org.uk/gitweb/?p=dak.git;a=commitdiff_plain;h=ed6728af1e7acc51129d9fd8b3f1d91697928872 possibly working multithread contents writer that has udebs disabled --- diff --git a/dak/contents.py b/dak/contents.py index b2e9816f..f89d050c 100755 --- a/dak/contents.py +++ b/dak/contents.py @@ -39,6 +39,8 @@ import os import logging import math import gzip +import threading +import Queue import apt_pkg from daklib import utils from daklib.binary import Binary @@ -71,9 +73,6 @@ OPTIONS -s, --suite={stable,testing,unstable,...} only operate on a single suite - - -a, --arch={i386,amd64} - only operate on a single architecture """ sys.exit(exit_code) @@ -111,18 +110,19 @@ olddeb_q = """PREPARE olddeb_q(int) as LIMIT 1""" # find me all of the contents for a given .deb -contents_q = """PREPARE contents_q(int,int,int,int) as +contents_q = """PREPARE contents_q(int,int,character varying(4)) as SELECT (p.path||'/'||n.file) AS fn, - comma_separated_list(s.section||'/'||b.package) + s.section, + b.package, + b.architecture from content_associations c join content_file_paths p ON (c.filepath=p.id) JOIN content_file_names n ON (c.filename=n.id) JOIN binaries b ON (b.id=c.binary_pkg) JOIN override o ON (o.package=b.package) JOIN section s ON (s.id=o.section) WHERE o.suite = $1 AND o.type = $2 - AND b.id in (SELECT ba.bin from bin_associations ba join binaries b on b.id=ba.bin where (b.architecture=$3 or b.architecture=$4)and ba.suite=$1 and b.type='deb') - GROUP BY fn - ORDER BY fn;""" + and b.type='$3' + ORDER BY fn""" # find me all of the contents for a given .udeb udeb_contents_q = """PREPARE udeb_contents_q(int,int,int,int,int) as @@ -160,18 +160,55 @@ remove_filepath_cruft_q = """DELETE FROM content_file_paths LEFT JOIN content_associations ca ON ca.filepath=cfn.id WHERE ca.id IS NULL)""" -class Contents(object): - """ - Class capable of generating Contents-$arch.gz files - Usage GenerateContents().generateContents( ["main","contrib","non-free"] ) - """ +class EndOfContents(object): + pass - def __init__(self): - self.header = None +class GzippedContentWriter(object): + def __init__(self, suite, arch): + self.queue = Queue.Queue() + self.current_file = None + self.first_package = True + self.output = self.open_file("dists/%s/Contents-%s.gz" % (suite, arch)) + self.thread = threading.Thread(target=self.write_thread, + name='Contents-%s writer'%arch) - def reject(self, message): - log.error("E: %s" % message) + self.thread.start() + + def open_file(self, filename): + filepath = Config()["Contents::Root"] + filename + filedir = os.path.dirname(filepath) + if not os.path.isdir(filedir): + os.makedirs(filedir) + return gzip.open(filepath, "w") + + def write(self, filename, section, package): + self.queue.put((file,section,package)) + + + def write_thread(self): + while True: + print("hi, i'm a thread" ); + next = self.queue.get() + print("GOT SOMEHTING FROM THE QUEUE" ); + if isinstance(next, EndOfContents): + self.output.write('\n') + self.output.close() + break + + (filename,section,package)=next + if next != current_file: + # this is the first file, so write the header first + if not current_file: + self.output.write(self._getHeader) + + self.output.write('\n%s\t' % filename) + self.first_package = True + if not first_package: + self.output.write(',') + else: + self.first_package=False + self.output.write('%s/%s' % (section,package)) def _getHeader(self): """ @@ -196,8 +233,6 @@ class Contents(object): return self.header - # goal column for section column - _goal_column = 54 def _write_content_file(self, cursor, filename): """ @@ -225,6 +260,25 @@ class Contents(object): finally: f.close() + + + +class Contents(object): + """ + Class capable of generating Contents-$arch.gz files + + Usage GenerateContents().generateContents( ["main","contrib","non-free"] ) + """ + + def __init__(self): + self.header = None + + def reject(self, message): + log.error("E: %s" % message) + + # goal column for section column + _goal_column = 54 + def cruft(self): """ remove files/paths from the DB which are no longer referenced @@ -292,6 +346,7 @@ class Contents(object): suites = self._suites() + # Get our suites, and the architectures for suite in [i.lower() for i in suites]: suite_id = DBConn().get_suite_id(suite) @@ -299,21 +354,63 @@ class Contents(object): arch_all_id = DBConn().get_architecture_id("all") + file_writers = {} + for arch_id in arch_list: - cursor.execute("EXECUTE contents_q(%d,%d,%d,%d)" % (suite_id, debtype_id, arch_all_id, arch_id[0] )) - self._write_content_file(cursor, "dists/%s/Contents-%s.gz" % (suite, arch_id[1])) + file_writers[arch_id] = GzippedContentWriter(suite, arch_id[1]) + + + print("EXECUTE contents_q(%d,%d,'%s')" % (suite_id, debtype_id, 'deb')) +# cursor.execute("EXECUTE contents_q(%d,%d,'%s');" % (suite_id, debtype_id, 'deb')) + cursor.execute("""SELECT (p.path||'/'||n.file) AS fn, + s.section, + b.package, + b.architecture + from content_associations c join content_file_paths p ON (c.filepath=p.id) + JOIN content_file_names n ON (c.filename=n.id) + JOIN binaries b ON (b.id=c.binary_pkg) + JOIN override o ON (o.package=b.package) + JOIN section s ON (s.id=o.section) + WHERE o.suite = %d AND o.type = %d + and b.type='deb' + ORDER BY fn""" % (suite_id, debtype_id)) + + while True: + r = cursor.fetchone() + print( "got contents: %s" % r ) + if not r: + print( "STU:END" ); + break + + print( "STU:NOT END" ); + filename, section, package, arch = r + + if arch == arch_all_id: + ## its arch all, so all contents files get it + for writer in file_writers.values(): + writer.write(filename, section, package) + + else: + file_writers[arch].write(filename, section, package) + + # close all the files + for writer in file_writers.values(): + writer.close() + + +# self._write_content_file(cursor, "dists/%s/Contents-%s.gz" % (suite, arch_id[1])) # The MORE fun part. Ok, udebs need their own contents files, udeb, and udeb-nf (not-free) # This is HORRIBLY debian specific :-/ - for section, fn_pattern in [("debian-installer","dists/%s/Contents-udeb-%s.gz"), - ("non-free/debian-installer", "dists/%s/Contents-udeb-nf-%s.gz")]: +# for section, fn_pattern in [("debian-installer","dists/%s/Contents-udeb-%s.gz"), +# ("non-free/debian-installer", "dists/%s/Contents-udeb-nf-%s.gz")]: - for arch_id in arch_list: - section_id = DBConn().get_section_id(section) # all udebs should be here) - if section_id != -1: - cursor.execute("EXECUTE udeb_contents_q(%d,%d,%d,%d,%d)" % (suite_id, udebtype_id, section_id, arch_id[0], arch_all_id)) +# for arch_id in arch_list: +# section_id = DBConn().get_section_id(section) # all udebs should be here) +# if section_id != -1: +# cursor.execute("EXECUTE udeb_contents_q(%d,%d,%d,%d,%d)" % (suite_id, udebtype_id, section_id, arch_id[0], arch_all_id)) - self._write_content_file(cursor, fn_pattern % (suite, arch_id[1])) +# self._write_content_file(cursor, fn_pattern % (suite, arch_id[1])) ################################################################################ @@ -333,20 +430,15 @@ class Contents(object): """ return a list of archs to operate on """ - arch_list = [ ] - if Config().has_key( "%s::%s" %(options_prefix,"Arch")): - archs = utils.split_args(Config()[ "%s::%s" %(options_prefix,"Arch")]) - for arch_name in archs: - arch_list.append((DBConn().get_architecture_id(arch_name), arch_name)) - else: - cursor.execute("EXECUTE arches_q(%d)" % (suite)) - while True: - r = cursor.fetchone() - if not r: - break + arch_list = [] + cursor.execute("EXECUTE arches_q(%d)" % (suite)) + while True: + r = cursor.fetchone() + if not r: + break - if r[1] != "source" and r[1] != "all": - arch_list.append((r[0], r[1])) + if r[1] != "source" and r[1] != "all": + arch_list.append((r[0], r[1])) return arch_list @@ -360,7 +452,6 @@ def main(): ('s',"suite", "%s::%s" % (options_prefix,"Suite"),"HasArg"), ('q',"quiet", "%s::%s" % (options_prefix,"Quiet")), ('v',"verbose", "%s::%s" % (options_prefix,"Verbose")), - ('a',"arch", "%s::%s" % (options_prefix,"Arch"),"HasArg"), ] commands = {'generate' : Contents.generate,