bash background process modify global variable

问题:

I have a global var foo="some value" and a background process back_func, I want to the background process to access $foo and modify its value, which can be seen by the main process. It's something like the following:

#!/bin/bash
foo=0

function back_func {
     foo=$(($foo+1))
     echo "back $foo"
}

(back_func) &
echo "global $foo"

The result of the above script is

global 0
back 1

How could I get the result of global and back are both '1'?, i.e. the back ground process's modification can return back to the main process.

回答1:

Upgrade 2019

Playing with bash_ipc_demo adding completion and a graph generator.

Rendez-vous

If you wanna have two independant process which could communicate, you have to place a rendez-vous somewhere both process can reach.

This could be a simple file, a fifo pipe, a unix socket, a TCP socket or maybe else (Rexx port).

bash and other shell

Bash don't have a equivalent to rexx port, so there is a little sample, using a rendez-vous file, that work (on my Linux).

I'm using shared memory /dev/shm, to reduce disk load.

Simple counter sample

$ back_func() {
    while :;do
        echo $(($(</dev/shm/foo)+1)) >/dev/shm/foo;
        sleep .3;
      done;
}

Let play

$ echo 1 >/dev/shm/foo
$ back_func &

$ echo $(</dev/shm/foo)
4

$ echo $(</dev/shm/foo)
21

Than stop now:

$ fg
back_func
^C

or

$ kill $!
$
[1]+  Terminated              back_func

More than one variables

For having many vars, there could by a nice manner:

$ back_func() {
    declare -A MYGLOBAL
    local vars
    while :; do
        ((MYGLOBAL["counter"]++))
        IFS=\ / read -a vars <<< "$(</proc/uptime) $(</proc/loadavg)"
        MYGLOBAL["uptime"]=$vars
        MYGLOBAL["idle"]=${vars[1]}
        MYGLOBAL["l01m"]=${vars[2]}
        MYGLOBAL["l05m"]=${vars[3]}
        MYGLOBAL["l15m"]=${vars[4]}
        MYGLOBAL["active"]=${vars[5]}
        MYGLOBAL["procs"]=${vars[6]}
        MYGLOBAL["lpid"]=${vars[7]}
        MYGLOBAL["rand"]=$RANDOM
        MYGLOBAL["crt"]=$SECONDS
        declare -p MYGLOBAL > /dev/shm/foo
        sleep 1
    done
}

Then

$ back_func &
[1] 27429
$ . /dev/shm/foo
$ echo ${MYGLOBAL['counter']}
5
$ echo ${MYGLOBAL['lpid']}
27432

and from there, why not:

$ dumpMyGlobal() {
    . /dev/shm/foo
    printf "%8s " ${!MYGLOBAL[@]}
    echo
    printf "%8s " ${MYGLOBAL[@]}
    echo
}

$ dumpMyGlobal
    l15m   uptime      crt    procs     lpid   active     rand     idle     l05m
  counter     l01m 
    0.42 13815568.06       95      554      649        1    31135 21437004.95   
  0.38       73     0.50 
$ dumpMyGlobal
    l15m   uptime      crt    procs     lpid   active     rand     idle     l05m
  counter     l01m 
    0.41 13815593.29      120      553      727        2     3849 21437046.41   
  0.35       98     0.33 

or

$ dumpMyGlobal() {
    . /dev/shm/foo
    sort <(
        paste <(
            printf "%-12s\n" ${!MYGLOBAL[@]}
          ) <(printf "%s\n" ${MYGLOBAL[@]})
    )
}

$ dumpMyGlobal
active              1
counter             297
crt                 337
idle                21435798.86
l01m                0.40
l05m                0.44
l15m                0.45
lpid                30418
procs               553
rand                7328
uptime              13814820.80

Get variable with snapshot

and finally getMyGlobalVar function

$ declare -A MYGLOBALLOCK   # snapshot variable
$ getMyGlobalVar () { 
    local i sync=false
    [ "$1" == "--sync" ] && shift && sync=true
    if [ -z "${MYGLOBALLOCK[*]}" ] || $sync; then
        . /dev/shm/foo
        for i in ${!MYGLOBAL[@]}
        do
            MYGLOBALLOCK[$i]=${MYGLOBAL[$i]}
        done
    fi
    echo ${MYGLOBALLOCK[$1]}
}

will require --sync flag for re-reading rendez-vous in order to let you look about each fields from the same snapshot.

$ getMyGlobalVar --sync idle
362084.12

$ getMyGlobalVar idle
362084.12

$ getMyGlobalVar rand
1533

$ getMyGlobalVar rand
1533

$ getMyGlobalVar --sync rand
43256

$ getMyGlobalVar idle
362127.63

Full useable demo:

There is a full sample: bash_ipc_demo or bash_ipc_demo.shz

You could use by:

wget http://f-hauri.ch/vrac/bash_ipc_demo

source bash_ipc_demo
back_func help
Usage: back_func [-q] [start [-g N]|stop|restart|status|get|dump|help]
   -q    Quiet
   -g N  Start daemon, setting uptime_useGraph to N values

back_func status
Background loop function is not running.

back_func start -g 3600

back_func status
Background loop function (19939) is running.

From there, if you source bash_ipc_demo in another terminal, you could do the list into them.

You could even close the first terminal.

back_func dump
backFunc_count                     13
backFunc_now      2016-04-06 17:03:19
backFunc_pid                    19939
backFunc_running                  yes
backFunc_start    2016-04-06 17:03:07
cpu_numcores                        2
loadavg_15min                    0.44
loadavg_1min                     0.66
loadavg_5min                     0.54
loadavg_active                      1
loadavg_last_pid                20005
loadavg_process                   650
random                        3714432
uptime_graph_val                 3600
uptime_idle                 425499.43
uptime_up                   495423.53
uptime_usage1sec                 9.90
uptime_usage                    57.06
uptime_useGraph  57.06 8.91 7.50 6.93 12.00 9.41 7.84 9.90 7.50 11.88 7.92 9.31 
9.90 

Then, you could get one value

back_func get backFunc_pid newVar
echo $newVar 
19939

or build a quick cpu graph:

lastMinuteGraph -p -o /tmp/lastMinuteGraph.png -W 640 -H 220

This will render a 640x220 PNG graphic, with uptime_graph_val values. In this case, as back_func start was invoked with -g 3600 from more than one hour, graphic show 3600 peek on 640 columns and 0-100% on 220 lines:

(Nota: Command was originaly named lastMinuteGraph as 1st version of this just stored 60 values, now this use uptime_graph_val for number of values to store. As I've used -g 3600 argument, this command could by named lastHourGraph).

Then:

back_func stop  
back_func get backFunc_end
2019-01-02 16:35:00

回答2:

According to the Bash manual here,

If a command is terminated by the control operator ‘&’, the shell executes the command asynchronously in a subshell.

And since a process run in a subshell cannot modify the environment of the parent shell, I guess what you are trying to do is only possible via temp files / named pipes. Or you could rethink your approach.

回答3:

If the main process (let's call it main.sh) is another periodically running bash script then you could simply have the the other script (let's call it other.sh) write the value to a file (let's call this file value.sh).

other.sh

#! /bin/bash  
echo "SOME_VAR=42" > /tmp/value.sh

main.sh

#! /bin/bash  
. /tmp/value.sh  
# Now you can use SOME_VAR