lib/oe/package_manager: Add utils function for multiprocess execution
PATCH 1/1 lib/oe/package_manager: Add utils function for multiprocess execution
Our usage of multitprocessing is problematic. In particular, there is a bug in python 2.7 multiprocessing where signals are not handled until command completion instead of immediately.
This factors the multiprocess code into a function which is enhanced with a workaround to ensure immediate signal handling and also better SIGINT handling which should happen in the parent, not the children to ensure clean exits. The workaround for the signals is being added to the core bb.utils function so it can benefit all users.
meta/lib/oe/package_manager.py +4 -28
@@ -7,6 +7,7 @@ import multiprocessing import re import bb import tempfile +import oe.utils # this can be used by all PM backends to create the index files in parallel
@@ -116,16 +117,7 @@ class RpmIndexer(Indexer): bb.note("There are no packages in %s" % self.deploy_dir) return - nproc = multiprocessing.cpu_count() - pool = bb.utils.multiprocessingpool(nproc) - results = list(pool.imap(create_index, index_cmds)) - pool.close() - pool.join() - - for result in results: - if result is not None: - return(result) - + oe.utils.multiprocess_exec(index_cmds, create_index) class OpkgIndexer(Indexer): def write_index(self):
@@ -161,15 +153,7 @@ class OpkgIndexer(Indexer): bb.note("There are no packages in %s!" % self.deploy_dir) return - nproc = multiprocessing.cpu_count() - pool = bb.utils.multiprocessingpool(nproc) - results = list(pool.imap(create_index, index_cmds)) - pool.close() - pool.join() - - for result in results: - if result is not None: - return(result) + oe.utils.multiprocess_exec(index_cmds, create_index) class DpkgIndexer(Indexer):
@@ -210,15 +194,7 @@ class DpkgIndexer(Indexer): bb.note("There are no packages in %s" % self.deploy_dir) return - nproc = multiprocessing.cpu_count() - pool = bb.utils.multiprocessingpool(nproc) - results = list(pool.imap(create_index, index_cmds)) - pool.close() - pool.join() - - for result in results: - if result is not None: - return(result) + oe.utils.multiprocess_exec(index_cmds, create_index) class PkgsList(object):
meta/lib/oe/utils.py +29 -0
@@ -151,3 +151,32 @@ def execute_pre_post_process(d, cmds): if cmd != '': bb.note("Executing %s ..." % cmd) bb.build.exec_func(cmd, d) + +def multiprocess_exec(commands, function): + import signal + import multiprocessing + + if not commands: + return [] + + def init_worker(): + signal.signal(signal.SIGINT, signal.SIG_IGN) + + nproc = min(multiprocessing.cpu_count(), len(commands)) + pool = bb.utils.multiprocessingpool(nproc, init_worker) + imap = pool.imap(function, commands) + + try: + results = list(imap) + pool.close() + pool.join() + results = [] + for result in results: + if result is not None: + results.append(result) + return results + + except KeyboardInterrupt: + pool.terminate() + pool.join() + raise
Quoted 3 lines · Richard Purdie
Our usage of multitprocessing is problematic. In particular, there is a bug in python 2.7 multiprocessing where signals are not handled until command completion instead of immediately.
This looks good, but the subject is misleading, in my opinion:
"lib/oe/package_manager:
Add utils function for multiprocess execution". It should probably either mention both utils and package_manager, or be split into two commits.
Signature · 5 lines
Christopher Larson clarson at kergoth dot com Founder - BitBake, OpenEmbedded, OpenZaurus Maintainer - Tslib Senior Software Engineer, Mentor Graphics
Quoted 8 lines · Christopher Larson
On Thu, Aug 21, 2014 at 1:46 PM, Richard Purdie <[email protected]> wrote:
Our usage of multitprocessing is problematic. In particular, there is a bug
in python 2.7 multiprocessing where signals are not handled until command
completion instead of immediately.
Quoted 4 lines
This looks good, but the subject is misleading, in my opinion:
"lib/oe/package_manager: Add utils function for multiprocess execution". It should probably either mention both utils andpackage_manager, or be split into two commits.
Good point, I tweaked before merging, thanks. It was the (bad) result of rearranging a few commits.
Cheers,
Richard