/[gentoo-x86]/eclass/multiprocessing.eclass
Gentoo

Contents of /eclass/multiprocessing.eclass

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1.5 - (show annotations) (download)
Tue Dec 3 08:15:14 2013 UTC (12 months, 3 weeks ago) by vapier
Branch: MAIN
Changes since 1.4: +8 -1 lines
add a /dev/fd sanity check #479656

1 # Copyright 1999-2013 Gentoo Foundation
2 # Distributed under the terms of the GNU General Public License v2
3 # $Header: /var/cvsroot/gentoo-x86/eclass/multiprocessing.eclass,v 1.4 2013/11/28 20:49:14 vapier Exp $
4
5 # @ECLASS: multiprocessing.eclass
6 # @MAINTAINER:
7 # base-system@gentoo.org
8 # @AUTHOR:
9 # Brian Harring <ferringb@gentoo.org>
10 # Mike Frysinger <vapier@gentoo.org>
11 # @BLURB: parallelization with bash (wtf?)
12 # @DESCRIPTION:
13 # The multiprocessing eclass contains a suite of functions that allow ebuilds
14 # to quickly run things in parallel using shell code.
15 #
16 # It has two modes: pre-fork and post-fork. If you don't want to dive into any
17 # more nuts & bolts, just use the pre-fork mode. For main threads that mostly
18 # spawn children and then wait for them to finish, use the pre-fork mode. For
19 # main threads that do a bit of processing themselves, use the post-fork mode.
20 # You may mix & match them for longer computation loops.
21 # @EXAMPLE:
22 #
23 # @CODE
24 # # First initialize things:
25 # multijob_init
26 #
27 # # Then hash a bunch of files in parallel:
28 # for n in {0..20} ; do
29 # multijob_child_init md5sum data.${n} > data.${n}
30 # done
31 #
32 # # Then wait for all the children to finish:
33 # multijob_finish
34 # @CODE
35
36 if [[ ${___ECLASS_ONCE_MULTIPROCESSING} != "recur -_+^+_- spank" ]] ; then
37 ___ECLASS_ONCE_MULTIPROCESSING="recur -_+^+_- spank"
38
39 # @FUNCTION: makeopts_jobs
40 # @USAGE: [${MAKEOPTS}]
41 # @DESCRIPTION:
42 # Searches the arguments (defaults to ${MAKEOPTS}) and extracts the jobs number
43 # specified therein. Useful for running non-make tools in parallel too.
44 # i.e. if the user has MAKEOPTS=-j9, this will echo "9" -- we can't return the
45 # number as bash normalizes it to [0, 255]. If the flags haven't specified a
46 # -j flag, then "1" is shown as that is the default `make` uses. Since there's
47 # no way to represent infinity, we return 999 if the user has -j without a number.
48 makeopts_jobs() {
49 [[ $# -eq 0 ]] && set -- ${MAKEOPTS}
50 # This assumes the first .* will be more greedy than the second .*
51 # since POSIX doesn't specify a non-greedy match (i.e. ".*?").
52 local jobs=$(echo " $* " | sed -r -n \
53 -e 's:.*[[:space:]](-j|--jobs[=[:space:]])[[:space:]]*([0-9]+).*:\2:p' \
54 -e 's:.*[[:space:]](-j|--jobs)[[:space:]].*:999:p')
55 echo ${jobs:-1}
56 }
57
58 # @FUNCTION: makeopts_loadavg
59 # @USAGE: [${MAKEOPTS}]
60 # @DESCRIPTION:
61 # Searches the arguments (defaults to ${MAKEOPTS}) and extracts the value set
62 # for load-average. For make and ninja based builds this will mean new jobs are
63 # not only limited by the jobs-value, but also by the current load - which might
64 # get excessive due to I/O and not just due to CPU load.
65 # Be aware that the returned number might be a floating-point number. Test
66 # whether your software supports that.
67 makeopts_loadavg() {
68 [[ $# -eq 0 ]] && set -- ${MAKEOPTS}
69 # This assumes the first .* will be more greedy than the second .*
70 # since POSIX doesn't specify a non-greedy match (i.e. ".*?").
71 local lavg=$(echo " $* " | sed -r -n \
72 -e 's:.*[[:space:]](-l|--load-average[=[:space:]])[[:space:]]*([0-9]+|[0-9]+\.[0-9]+)[^0-9.]*:\2:p' \
73 -e 's:.*[[:space:]](-l|--load-average)[[:space:]].*:999:p')
74 echo ${lavg:-1}
75 }
76
77 # @FUNCTION: multijob_init
78 # @USAGE: [${MAKEOPTS}]
79 # @DESCRIPTION:
80 # Setup the environment for executing code in parallel.
81 # You must call this before any other multijob function.
82 multijob_init() {
83 # When something goes wrong, try to wait for all the children so we
84 # don't leave any zombies around.
85 has wait ${EBUILD_DEATH_HOOKS} || EBUILD_DEATH_HOOKS+=" wait "
86
87 # Setup a pipe for children to write their pids to when they finish.
88 # We have to allocate two fd's because POSIX has undefined behavior
89 # when you open a FIFO for simultaneous read/write. #487056
90 local pipe="${T}/multijob.pipe"
91 mkfifo -m 600 "${pipe}"
92 redirect_alloc_fd mj_write_fd "${pipe}"
93 redirect_alloc_fd mj_read_fd "${pipe}"
94 rm -f "${pipe}"
95
96 # See how many children we can fork based on the user's settings.
97 mj_max_jobs=$(makeopts_jobs "$@")
98 mj_num_jobs=0
99 }
100
101 # @FUNCTION: multijob_child_init
102 # @USAGE: [--pre|--post] [command to run in background]
103 # @DESCRIPTION:
104 # This function has two forms. You can use it to execute a simple command
105 # in the background (and it takes care of everything else), or you must
106 # call this first thing in your forked child process.
107 #
108 # The --pre/--post options allow you to select the child generation mode.
109 #
110 # @CODE
111 # # 1st form: pass the command line as arguments:
112 # multijob_child_init ls /dev
113 # # Or if you want to use pre/post fork modes:
114 # multijob_child_init --pre ls /dev
115 # multijob_child_init --post ls /dev
116 #
117 # # 2nd form: execute multiple stuff in the background (post fork):
118 # (
119 # multijob_child_init
120 # out=`ls`
121 # if echo "${out}" | grep foo ; then
122 # echo "YEAH"
123 # fi
124 # ) &
125 # multijob_post_fork
126 #
127 # # 2nd form: execute multiple stuff in the background (pre fork):
128 # multijob_pre_fork
129 # (
130 # multijob_child_init
131 # out=`ls`
132 # if echo "${out}" | grep foo ; then
133 # echo "YEAH"
134 # fi
135 # ) &
136 # @CODE
137 multijob_child_init() {
138 local mode="pre"
139 case $1 in
140 --pre) mode="pre" ; shift ;;
141 --post) mode="post"; shift ;;
142 esac
143
144 if [[ $# -eq 0 ]] ; then
145 trap 'echo ${BASHPID} $? >&'${mj_write_fd} EXIT
146 trap 'exit 1' INT TERM
147 else
148 local ret
149 [[ ${mode} == "pre" ]] && { multijob_pre_fork; ret=$?; }
150 ( multijob_child_init ; "$@" ) &
151 [[ ${mode} == "post" ]] && { multijob_post_fork; ret=$?; }
152 return ${ret}
153 fi
154 }
155
156 # @FUNCTION: _multijob_fork
157 # @INTERNAL
158 # @DESCRIPTION:
159 # Do the actual book keeping.
160 _multijob_fork() {
161 [[ $# -eq 1 ]] || die "incorrect number of arguments"
162
163 local ret=0
164 [[ $1 == "post" ]] && : $(( ++mj_num_jobs ))
165 if [[ ${mj_num_jobs} -ge ${mj_max_jobs} ]] ; then
166 multijob_finish_one
167 ret=$?
168 fi
169 [[ $1 == "pre" ]] && : $(( ++mj_num_jobs ))
170 return ${ret}
171 }
172
173 # @FUNCTION: multijob_pre_fork
174 # @DESCRIPTION:
175 # You must call this in the parent process before forking a child process.
176 # If the parallel limit has been hit, it will wait for one child to finish
177 # and return its exit status.
178 multijob_pre_fork() { _multijob_fork pre "$@" ; }
179
180 # @FUNCTION: multijob_post_fork
181 # @DESCRIPTION:
182 # You must call this in the parent process after forking a child process.
183 # If the parallel limit has been hit, it will wait for one child to finish
184 # and return its exit status.
185 multijob_post_fork() { _multijob_fork post "$@" ; }
186
187 # @FUNCTION: multijob_finish_one
188 # @DESCRIPTION:
189 # Wait for a single process to exit and return its exit code.
190 multijob_finish_one() {
191 [[ $# -eq 0 ]] || die "${FUNCNAME} takes no arguments"
192
193 local pid ret
194 read -r -u ${mj_read_fd} pid ret || die
195 : $(( --mj_num_jobs ))
196 return ${ret}
197 }
198
199 # @FUNCTION: multijob_finish
200 # @DESCRIPTION:
201 # Wait for all pending processes to exit and return the bitwise or
202 # of all their exit codes.
203 multijob_finish() {
204 local ret=0
205 while [[ ${mj_num_jobs} -gt 0 ]] ; do
206 multijob_finish_one
207 : $(( ret |= $? ))
208 done
209 # Let bash clean up its internal child tracking state.
210 wait
211
212 # Do this after reaping all the children.
213 [[ $# -eq 0 ]] || die "${FUNCNAME} takes no arguments"
214
215 # No need to hook anymore.
216 EBUILD_DEATH_HOOKS=${EBUILD_DEATH_HOOKS/ wait / }
217
218 return ${ret}
219 }
220
221 # @FUNCTION: redirect_alloc_fd
222 # @USAGE: <var> <file> [redirection]
223 # @DESCRIPTION:
224 # Find a free fd and redirect the specified file via it. Store the new
225 # fd in the specified variable. Useful for the cases where we don't care
226 # about the exact fd #.
227 redirect_alloc_fd() {
228 local var=$1 file=$2 redir=${3:-"<>"}
229
230 # Make sure /dev/fd is sane. #479656
231 if [[ ! -L /dev/fd ]] ; then
232 eerror "You're missing a /dev/fd symlink to /proc/self/fd."
233 eerror "Please fix the symlink and check your boot scripts (udev/etc...)."
234 die "/dev/fd is broken"
235 fi
236
237 if [[ $(( (BASH_VERSINFO[0] << 8) + BASH_VERSINFO[1] )) -ge $(( (4 << 8) + 1 )) ]] ; then
238 # Newer bash provides this functionality.
239 eval "exec {${var}}${redir}'${file}'"
240 else
241 # Need to provide the functionality ourselves.
242 local fd=10
243 while :; do
244 # Make sure the fd isn't open. It could be a char device,
245 # or a symlink (possibly broken) to something else.
246 if [[ ! -e /dev/fd/${fd} ]] && [[ ! -L /dev/fd/${fd} ]] ; then
247 eval "exec ${fd}${redir}'${file}'" && break
248 fi
249 [[ ${fd} -gt 1024 ]] && die 'could not locate a free temp fd !?'
250 : $(( ++fd ))
251 done
252 : $(( ${var} = fd ))
253 fi
254 }
255
256 fi

  ViewVC Help
Powered by ViewVC 1.1.20