PAEXEC distributes performing the given tasks across several CPUs or machines in a network.
Be the first to post a text review of Parallel executor. Rate and review a project by clicking thumbs up or thumbs down in the right column.
2009-03-07 Aleksey Cheusov <vle@gmx.net> * NEWS: date * paexec.c: fixes for -Z * NEWS: fixes * Makefile.version, NEWS: version -> 0.13.0 Release notes in NEWS * paexec.pod: rewording 2009-03-04 Aleksey Cheusov <vle@gmx.net> * Makefile: CLEANFILES target is used instead of hand-made rules 2009-02-28 Aleksey Cheusov <vle@gmx.net> * paexec.c: fixed: compilation bug with Intel CC * paexec.c: FIXED: In the new code for cycles detection, one array was not cleanly zeroed, seen on NetBSD-2.1 2009-02-27 Aleksey Cheusov <vle@gmx.net> * TODO: restarting command on failed nodes is done * paexec.c, tests/test.out, tests/test.sh: Floyd's algorithm with hash table for accessing "square" table seems very slow. New algorithm is implemented (based on depth-first search in graph). It works much faster and doesn't require square matrix at all, i.e. doesn't allocate lots of memory. 2009-02-26 Aleksey Cheusov <vle@gmx.net> * paexec.c: When 'paexec -s' retrieves ~5000 tasks, its cycles_detection phrase allocated about ~400Mb of memory. This happens because this phase allocated a square table [tasks_count x tasks_count] of int elements. Now I've fixed this by using libmaa's hash tables instead of square table. As a result cycle detection phase comes a little slower but doesn't require hundreds of megabytes of memory anymore. * paexec.c: SIGALRM is sent once a second, not per persistent_timeout. This allows to use several different timeouts (for future changes). * paexec.c: FIXED: cycles detection phase was too slow. I just forgot to optimize Floyd' algorithm. Now it is fixed. More debugging output * Makefile: fixes in comments * tests/test.out, tests/test.sh: tests compatible with NetBSD, Linux and FreeBSD 2009-02-25 Aleksey Cheusov <vle@gmx.net> * examples/broken_echo/transport_broken_echo2, tests/test.out: more complex test, more comments 2009-02-18 Aleksey Cheusov <vle@gmx.net> * paexec.c: SA_RESTART is not needed anymore * Makefile, nonblock_helpers.c, nonblock_helpers.h, wrappers.c, wrappers.h: unused/dead code has been removed * examples/broken_echo/transport_broken_echo2, paexec.c, paexec.pod, tests/test.out, tests/test.sh: ADDED: -Z<timeout> =item B<-Z> I<timeout> When I<-z> applied, if a I<command> fails, appropriate node is marked as broken and is excluded from the following task distribution. If B<-Z> applied, every I<timeout> seconds it is made an attempt to rerun a failed comand on failed node. I<-Z> implies I<-z>. 2009-02-17 Aleksey Cheusov <vle@gmx.net> * nonblock_helpers.c, nonblock_helpers.h, paexec.c, wrappers.c, wrappers.h: no EINTR wrappers (iread, xread etc.). SIGCHLD and SIGALRM are blocked most of the time * tests/test.sh: fix in test * tests/test.sh: fix: one number per line * TODO: one task minus
2008-12-23 Aleksey Cheusov <vle@gmx.net> * paexec.c, tests/test.out, tests/test.sh: fix for -zi: task is output to stdout before the keyword "fatal" in case of fatal failure. * NEWS: rewording * Makefile.version, NEWS: version -> 0.12.1 release notes * Makefile: fix: FreeBSD make doesn't support .PARSEDIR * paexec.c: grammar fix * paexec.c: indent * tests/test.sh: more FATALs * paexec.c: failed_nodes_*** renamed to failed_taskids_*** * paexec.c: line_num renamed to taskid line_nums renamed to node2taskid * tests/test.out, tests/test.sh, tests/transport_broken_rnd: yet another real life regression test * tests/test.out, tests/test.sh, tests/transport_broken, tests/transport_broken_rnd: more complex regression test * paexec.c: fixed: sometimes -lz shows incorrect task IDs 2008-12-22 Aleksey Cheusov <vle@gmx.net> * paexec.c, tests/test.out, tests/test.sh, tests/transport_broken: FIX: Actually -z doesn't work at all :-( Existing regression tests do not show this :-/ Now, I'm trying to fix it. SIGCHLD handler is almost empty. Actual work is done ONLY when read/write fail. Trivial regression test for this.
2008-12-22 Aleksey Cheusov <vle@gmx.net> * NEWS: release time * Makefile: "SRCROOT?=..." and ".PATH:..." removed * Makefile.version, NEWS: version -> 0.12.0, release time change * Makefile, NEWS: CPPFLAGS -> CFLAGS * paexec.c: fgets -> xfgetln * Makefile: minor clean-ups * paexec.pod: minor fixes and clean-ups * NEWS, paexec.pod: fixes * NEWS, paexec.pod: release notes and manual page updates (-z) 2008-12-21 Aleksey Cheusov <vle@gmx.net> * Makefile: clean-ups * tests/test.sh: make test: temporary files are created in .OBJDIR 2008-12-20 Aleksey Cheusov <vle@gmx.net> * Makefile: make clean: additional clean-ups * Makefile, tests/test.out, tests/test.sh: support for 'env MAKEOBJDIR=tratata bmake test' 2008-12-19 Aleksey Cheusov <vle@gmx.net> * paexec.c: typo fix * Makefile, NEWS, examples/broken_echo/transport_closed_stdin.c, paexec.c, tests/test.out, tests/test.sh: Now 'paexec -z' handles correctly write(2) errors. For this to work sigpipe signal is ignored. New regression test for this. transport_closed_stdin.c - test tool for regression test 2008-12-01 Aleksey Cheusov <vle@gmx.net> * TODO: some tasks done * paexec.c: SA_RESTART is set to 0 if it is not defined 2008-11-22 Aleksey Cheusov <vle@gmx.net> * tests/test.out, tests/test.sh: new regression test * examples/broken_echo/transport_broken_echo, paexec.c, tests/test.out: -z: resistance to write(2) error * examples/broken_echo/transport_broken_echo, paexec.c, tests/test.out, tests/test.sh: new regression tests for -z option (failure resistance) -z: minor fixes and improvements 2008-11-21 Aleksey Cheusov <vle@gmx.net> * paexec.c: SA_RESTART for SIGCHLD signal handler 2008-11-19 Aleksey Cheusov <vle@gmx.net> * paexec.c: <errno.h> 2008-11-18 Aleksey Cheusov <vle@gmx.net> * examples/broken_echo/transport_broken_echo, paexec.c, tests/test.out, tests/test.sh: Further support for -z * paexec.c: node2task [n] keep a task sent to node `n' 2008-11-17 Aleksey Cheusov <vle@gmx.net> * tests/test.out, tests/test.sh, examples/broken_echo/transport_broken_echo, paexec.c, wrappers.c, wrappers.h: initial support for -z and -Z regression test for the case when transport exits unexpectedly 2008-11-16 Aleksey Cheusov <vle@gmx.net> * TODO: rhomb deps bug already fixed * paexec.c, wrappers.c, wrappers.h: added: alive_nodes_count variable added: iclose function added: mark_node_as_dead function minor clean-ups * paexec.c: Signal handler for SIGCHLD * paexec.c, wrappers.c, wrappers.h: homemade fgetln is used for reading tasks from stdin * paexec.c: code clean-ups, select(2) do not listen on fd=0 (stdin) Instead, fgets(..., stdin) is used 2008-11-15 Aleksey Cheusov <vle@gmx.net> * examples/make_package/make_package_cmd__all_failed, examples/make_package/make_package_cmd__autoconf, examples/make_package/make_package_cmd__byacc, examples/make_package/make_package_cmd__dict-client, examples/make_package/make_package_cmd__dict-server, examples/make_package/make_package_cmd__dictem, examples/make_package/make_package_cmd__flex, examples/make_package/make_package_cmd__flex_byacc, examples/make_package/make_package_cmd__glib2, examples/make_package/make_package_cmd__gmake, examples/make_package/make_package_cmd__gmake_autoconf, examples/make_package/make_package_cmd__libjudy, examples/make_package/make_package_cmd__libmaa, examples/make_package/make_package_cmd__m4, examples/make_package/make_package_cmd__xxx_failed, tests/test.out, tests/test.sh: clean-ups in regression tests, lots of scripts are replaced by one program.
2008-10-25 Aleksey Cheusov <vle at gmx.net} * tests/test.sh: fix for regression test * examples/make_package/make_package_tasks_cycle: a part of new regression test * Makefile.version, NEWS: version -> 0.11.0 release notes in NEWS * tests/test.sh: no "set -e" * paexec.pod: minor fix * Makefile: WARNS=4 ---> WARNS?=4 * tests/test.out, tests/test.sh: Regression test for cycle detection * paexec.c: Child processes are run at the very end of paexec initialization * paexec.c: paexec -s: before entering main 'loop', input graph is checked for cycles. If it found paexec exits with error.
2008-10-24 Aleksey Cheusov <vle@gmx.net> * Makefile.version, NEWS: version -> 0.10.1 release notes * Makefile.version: version -> 0.10.1beta1 * paexec.c, tests/test.out: minor fix for rhomb like dependencies * paexec.c: code clean-ups, enum instead of -1|0|1 constants * Makefile: remove .sinclude * tests/test.sh: remove gawk appeared by mistake diff -U ---> standard diff -C10 2008-09-27 Aleksey Cheusov <vle@gmx.net> * paexec.pod: fix: accept(2) -> select(2) * paexec.pod: AUTHOR section is at the very end end of man page * README, paexec.pod: By default, long options are not available on systems that don't provide getopt_long(3) * portabhacks.h, tests/test.out, tests/test.sh: regression tests use short options only (under Interix long options/get_opt_long is not supported) fixes in code for disabling use of getopt_long 2008-09-26 Aleksey Cheusov <vle@gmx.net> * paexec.c: minor clean-ups * portabhacks.h: getopt_long is used on the following "platforms" (macro): __NetBSD__, __FreeBSD__, __OpenBSD__, __DragonFly__ and __APPLE__. On others getopt_long defaults to the standard getopt(3) 2008-09-02 Aleksey Cheusov <vle@gmx.net> * tests/test.sh: fix: diff -H20 ===> diff -U 20 * TODO, examples/make_package/make_package_tasks2, tests/test.out, tests/test.sh: new regression test (currently fails) and TODO for rhomb-line dependancy 2008-08-31 Aleksey Cheusov <vle@gmx.net> * examples/make_package/make_package_cmd__all_failed: new script for regression tests
Be the first person to add a text review.
Copyright © 2009 Geeknet, Inc. All rights reserved. Terms of Use
Thanks for your rating!
Would you also like to write a review?