1 # Copyright 1999-2014 Gentoo Foundation
2 # Distributed under the terms of the GNU General Public License v2
5 # @ECLASS: multiprocessing.eclass
7 # base-system@gentoo.org
9 # Brian Harring <ferringb@gentoo.org>
10 # Mike Frysinger <vapier@gentoo.org>
11 # @BLURB: parallelization with bash (wtf?)
13 # The multiprocessing eclass contains a suite of functions that allow ebuilds
14 # to quickly run things in parallel using shell code.
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.
24 # # First initialize things:
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}
32 # # Then wait for all the children to finish:
36 if [[ -z ${_MULTIPROCESSING_ECLASS} ]]; then
37 _MULTIPROCESSING_ECLASS=1
41 # Return the process id of the current sub shell. This is to support bash
42 # versions older than 4.0 that lack $BASHPID support natively. Simply do:
43 # echo ${BASHPID:-$(bashpid)}
45 # Note: Using this func in any other way than the one above is not supported.
47 # Running bashpid plainly will return incorrect results. This func must
48 # be run in a subshell of the current subshell to get the right pid.
49 # i.e. This will show the wrong value:
51 # But this will show the right value:
56 # @FUNCTION: get_nproc
57 # @USAGE: [${fallback:-1}]
59 # Attempt to figure out the number of processing units available.
60 # If the value can not be determined, prints the provided fallback
61 # instead. If no fallback is provided, defaults to 1.
66 if type -P nproc &>/dev/null; then
71 if [[ -z ${nproc} ]] && type -P sysctl &>/dev/null; then
72 nproc=$(sysctl -n hw.ncpu 2>/dev/null)
75 # fallback to python2.6+
76 # note: this may fail (raise NotImplementedError)
77 if [[ -z ${nproc} ]] && type -P python &>/dev/null; then
78 nproc=$(python -c 'import multiprocessing; print(multiprocessing.cpu_count());' 2>/dev/null)
81 if [[ -n ${nproc} ]]; then
88 # @FUNCTION: makeopts_jobs
89 # @USAGE: [${MAKEOPTS}] [${inf:-999}]
91 # Searches the arguments (defaults to ${MAKEOPTS}) and extracts the jobs number
92 # specified therein. Useful for running non-make tools in parallel too.
93 # i.e. if the user has MAKEOPTS=-j9, this will echo "9" -- we can't return the
94 # number as bash normalizes it to [0, 255]. If the flags haven't specified a
95 # -j flag, then "1" is shown as that is the default `make` uses. Since there's
96 # no way to represent infinity, we return ${inf} (defaults to 999) if the user
97 # has -j without a number.
99 [[ $# -eq 0 ]] && set -- ${MAKEOPTS}
100 # This assumes the first .* will be more greedy than the second .*
101 # since POSIX doesn't specify a non-greedy match (i.e. ".*?").
102 local jobs=$(echo " $* " | sed -r -n \
103 -e 's:.*[[:space:]](-[a-z]*j|--jobs[=[:space:]])[[:space:]]*([0-9]+).*:\2:p' \
104 -e "s:.*[[:space:]](-[a-z]*j|--jobs)[[:space:]].*:${2:-999}:p")
108 # @FUNCTION: makeopts_loadavg
109 # @USAGE: [${MAKEOPTS}] [${inf:-999}]
111 # Searches the arguments (defaults to ${MAKEOPTS}) and extracts the value set
112 # for load-average. For make and ninja based builds this will mean new jobs are
113 # not only limited by the jobs-value, but also by the current load - which might
114 # get excessive due to I/O and not just due to CPU load.
115 # Be aware that the returned number might be a floating-point number. Test
116 # whether your software supports that.
117 # If no limit is specified or --load-average is used without a number, ${inf}
118 # (defaults to 999) is returned.
120 [[ $# -eq 0 ]] && set -- ${MAKEOPTS}
121 # This assumes the first .* will be more greedy than the second .*
122 # since POSIX doesn't specify a non-greedy match (i.e. ".*?").
123 local lavg=$(echo " $* " | sed -r -n \
124 -e 's:.*[[:space:]](-[a-z]*l|--(load-average|max-load)[=[:space:]])[[:space:]]*([0-9]+|[0-9]+\.[0-9]+).*:\3:p' \
125 -e "s:.*[[:space:]](-[a-z]*l|--(load-average|max-load))[[:space:]].*:${2:-999}:p")
126 # Default to ${inf} since the default is to not use a load limit.
127 echo ${lavg:-${2:-999}}
130 # @FUNCTION: multijob_init
131 # @USAGE: [${MAKEOPTS}]
133 # Setup the environment for executing code in parallel.
134 # You must call this before any other multijob function.
136 # When something goes wrong, try to wait for all the children so we
137 # don't leave any zombies around.
138 has wait ${EBUILD_DEATH_HOOKS} || EBUILD_DEATH_HOOKS+=" wait "
140 # Setup a pipe for children to write their pids to when they finish.
141 # We have to allocate two fd's because POSIX has undefined behavior
142 # when you open a FIFO for simultaneous read/write. #487056
143 local pipe="${T}/multijob.pipe"
144 mkfifo -m 600 "${pipe}"
145 redirect_alloc_fd mj_write_fd "${pipe}"
146 redirect_alloc_fd mj_read_fd "${pipe}"
149 # See how many children we can fork based on the user's settings.
150 mj_max_jobs=$(makeopts_jobs "$@")
154 # @FUNCTION: multijob_child_init
155 # @USAGE: [--pre|--post] [command to run in background]
157 # This function has two forms. You can use it to execute a simple command
158 # in the background (and it takes care of everything else), or you must
159 # call this first thing in your forked child process.
161 # The --pre/--post options allow you to select the child generation mode.
164 # # 1st form: pass the command line as arguments:
165 # multijob_child_init ls /dev
166 # # Or if you want to use pre/post fork modes:
167 # multijob_child_init --pre ls /dev
168 # multijob_child_init --post ls /dev
170 # # 2nd form: execute multiple stuff in the background (post fork):
172 # multijob_child_init
174 # if echo "${out}" | grep foo ; then
180 # # 2nd form: execute multiple stuff in the background (pre fork):
183 # multijob_child_init
185 # if echo "${out}" | grep foo ; then
190 multijob_child_init() {
193 --pre) mode="pre" ; shift ;;
194 --post) mode="post"; shift ;;
197 if [[ $# -eq 0 ]] ; then
198 trap 'echo ${BASHPID:-$(bashpid)} $? >&'${mj_write_fd} EXIT
199 trap 'exit 1' INT TERM
202 [[ ${mode} == "pre" ]] && { multijob_pre_fork; ret=$?; }
203 ( multijob_child_init ; "$@" ) &
204 [[ ${mode} == "post" ]] && { multijob_post_fork; ret=$?; }
209 # @FUNCTION: _multijob_fork
212 # Do the actual book keeping.
214 [[ $# -eq 1 ]] || die "incorrect number of arguments"
217 [[ $1 == "post" ]] && : $(( ++mj_num_jobs ))
218 if [[ ${mj_num_jobs} -ge ${mj_max_jobs} ]] ; then
222 [[ $1 == "pre" ]] && : $(( ++mj_num_jobs ))
226 # @FUNCTION: multijob_pre_fork
228 # You must call this in the parent process before forking a child process.
229 # If the parallel limit has been hit, it will wait for one child to finish
230 # and return its exit status.
231 multijob_pre_fork() { _multijob_fork pre "$@" ; }
233 # @FUNCTION: multijob_post_fork
235 # You must call this in the parent process after forking a child process.
236 # If the parallel limit has been hit, it will wait for one child to finish
237 # and return its exit status.
238 multijob_post_fork() { _multijob_fork post "$@" ; }
240 # @FUNCTION: multijob_finish_one
242 # Wait for a single process to exit and return its exit code.
243 multijob_finish_one() {
244 [[ $# -eq 0 ]] || die "${FUNCNAME} takes no arguments"
247 read -r -u ${mj_read_fd} pid ret || die
248 : $(( --mj_num_jobs ))
252 # @FUNCTION: multijob_finish
254 # Wait for all pending processes to exit and return the bitwise or
255 # of all their exit codes.
258 while [[ ${mj_num_jobs} -gt 0 ]] ; do
262 # Let bash clean up its internal child tracking state.
265 # Do this after reaping all the children.
266 [[ $# -eq 0 ]] || die "${FUNCNAME} takes no arguments"
268 # No need to hook anymore.
269 EBUILD_DEATH_HOOKS=${EBUILD_DEATH_HOOKS/ wait / }
274 # @FUNCTION: redirect_alloc_fd
275 # @USAGE: <var> <file> [redirection]
277 # Find a free fd and redirect the specified file via it. Store the new
278 # fd in the specified variable. Useful for the cases where we don't care
279 # about the exact fd #.
280 redirect_alloc_fd() {
281 local var=$1 file=$2 redir=${3:-"<>"}
283 # Make sure /dev/fd is sane on Linux hosts. #479656
284 if [[ ! -L /dev/fd && ${CBUILD} == *linux* ]] ; then
285 eerror "You're missing a /dev/fd symlink to /proc/self/fd."
286 eerror "Please fix the symlink and check your boot scripts (udev/etc...)."
287 die "/dev/fd is broken"
290 if [[ $(( (BASH_VERSINFO[0] << 8) + BASH_VERSINFO[1] )) -ge $(( (4 << 8) + 1 )) ]] ; then
291 # Newer bash provides this functionality.
292 eval "exec {${var}}${redir}'${file}'"
294 # Need to provide the functionality ourselves.
297 # Make sure the fd isn't open. It could be a char device,
298 # or a symlink (possibly broken) to something else.
299 if [[ ! -e /dev/fd/${fd} ]] && [[ ! -L /dev/fd/${fd} ]] ; then
300 eval "exec ${fd}${redir}'${file}'" && break
302 [[ ${fd} -gt 1024 ]] && die 'could not locate a free temp fd !?'