Is there any option in perf to look into processes running on a particular cpu /core, and how much percentage of that core is taken by each process.
Reference links would be helpful.
perf is intended to do a profiling which is not good fit for your case. You may try to do sampling /proc/sched_debug (if it is compiled in your kernel). For example you may check which process is currently running on CPU:
egrep '^R|cpu#' /proc/sched_debug
cpu#0, 917.276 MHz
R egrep 2614 37730.177313 ...
cpu#1, 917.276 MHz
R bash 2023 218715.010833 ...
By using his PID as a key, you may check how many CPU time in milliseconds it consumed:
grep se.sum_exec_runtime /proc/2023/sched
se.sum_exec_runtime : 279346.058986
However, as #BrenoLeitão mentioned, SystemTap is quite useful for your script. Here is script for your task.
global cputimes;
global cmdline;
global oncpu;
global NS_PER_SEC = 1000000000;
probe scheduler.cpu_on {
oncpu[pid()] = local_clock_ns();
}
probe scheduler.cpu_off {
if(oncpu[pid()] == 0)
next;
cmdline[pid()] = cmdline_str();
cputimes[pid(), cpu()] <<< local_clock_ns() - oncpu[pid()];
delete oncpu[pid()];
}
probe timer.s(1) {
printf("%6s %3s %6s %s\n", "PID", "CPU", "PCT", "CMDLINE");
foreach([pid+, cpu] in cputimes) {
cpupct = #sum(cputimes[pid, cpu]) * 10000 / NS_PER_SEC;
printf("%6d %3d %3d.%02d %s\n", pid, cpu,
cpupct / 100, cpupct % 100, cmdline[pid]);
}
delete cputimes;
}
It traces moments when process is running on CPU and stops execution on that (due to migration or sleeping) by attaching to scheduler.cpu_on and scheduler.cpu_off probes. Second probe calculates time difference between these events and saves it to cputimes aggregation along with process command line arguments.
timer.s(1) fires once per second -- it walks over aggregation and calculates percentage. Here is sample output for Centos 7 with bash running infinite loop:
0 0 100.16
30 1 0.00
51 0 0.00
380 0 0.02 /usr/bin/python -Es /usr/sbin/tuned -l -P
2016 0 0.08 sshd: root#pts/0 "" "" "" ""
2023 1 100.11 -bash
2630 0 0.04 /usr/libexec/systemtap/stapio -R stap_3020c9e7ba76838179be68cd2390a10c_2630 -F3
I understand that perf is not the proper way to do it, although you can limit perf per CPU, as using perf record -C <cpulist> or even perf stat -c <cpulist>.
The close you are going to see is the context-switch event, but, this is not going to provide you the application names at all.
I think you are going to need something more powerful, as systemtap.
Related
Problem:
I'm writing program in golang on linux that needs to execute long running process so that:
I redirect stdout of running process to file.
I control the user of process.
Process doesn't die when my program exits.
The process doesn't become a zombie when it crashes.
I get PID of running process.
I'm running my program with root permissions.
Attempted solution:
func Run(pathToBin string, args []string, uid uint32, stdLogFile *os.File) (int, error) {
cmd := exec.Command(pathToBin, args...)
cmd.SysProcAttr = &syscall.SysProcAttr{
Credential: &syscall.Credential{
Uid: uid,
},
}
cmd.Stdout = stdLogFile
if err := cmd.Start(); err != nil {
return -1, err
}
go func() {
cmd.Wait() //Wait is necessary so cmd doesn't become a zombie
}()
return cmd.Process.Pid, nil
}
This solution seems to satisfy almost all of my requirements except that when I send SIGTERM/SIGKILL to my program the underlying process crashes. In fact I want my background process to be as separate as possible: it has different parent pid, group pid etc. from my program. I want to run it as daemon.
Other solutions on stackoverflow suggested to use cmd.Process.Release() for similar use cases, but it doesn't seem to work.
Solutions which are not applicable in my case:
I have no control over code of process I'm running. My solution has to work for any process.
I can't use external commands to run it, just pure go. So using systemd or something similar is not applicable.
I can in fact use library that is easily importable using import from github etc.
TLDR;
Just use https://github.com/hashicorp/go-reap
There is a great Russian expression which reads "don't try to give birth to a bicycle" and it means don't reinvent the wheel and keep it simple. I think it applies here. If I were you, I'd reconsider using one of:
https://github.com/krallin/tini
https://busybox.net/
https://software.clapper.org/daemonize/
https://wiki.gentoo.org/wiki/OpenRC
https://www.freedesktop.org/wiki/Software/systemd/
This issue has already been solved ;)
Your question is imprecise or you are asking for non-standard features.
In fact I want my background process to be as separate as possible: it has different parent pid, group pid etc. from my program. I want to run it as daemon.
That is not how process inheritance works. You can not have process A start Process B and somehow change the parent of B to C. To the best of my knowledge this is not possible in Linux.
In other words, if process A (pid 55) starts process B (100), then B must have parent pid 55.
The only way to avoid that is have something else start the B process such as atd, crond, or something else - which is not what you are asking for.
If parent 55 dies, then PID 1 will be the parent of 100, not some arbitrary process.
Your statement "it has different parent pid" does not makes sense.
I want to run it as daemon.
That's excellent. However, in a GNU / Linux system, all daemons have a parent pid and those parents have a parent pid going all the way up to pid 1, strictly according to the parent -> child rule.
when I send SIGTERM/SIGKILL to my program the underlying process crashes.
I can not reproduce that behavior. See case8 and case7 from the proof-of-concept repo.
make case8
export NOSIGN=1; make build case7
unset NOSIGN; make build case7
$ make case8
{ sleep 6 && killall -s KILL zignal; } &
./bin/ctrl-c &
sleep 2; killall -s TERM ctrl-c
kill with:
{ pidof ctrl-c; pidof signal ; } | xargs -r -t kill -9
main() 2476074
bashed 2476083 (2476081)
bashed 2476084 (2476081)
bashed 2476085 (2476081)
zignal 2476088 (2476090)
go main() got 23 urgent I/O condition
go main() got 23 urgent I/O condition
zignal 2476098 (2476097)
go main() got 23 urgent I/O condition
zignal 2476108 (2476099)
main() wait...
p 2476088
p 2476098
p 2476108
p 2476088
go main() got 15 terminated
sleep 1; killall -s TERM ctrl-c
p 2476098
p 2476108
p 2476088
go main() got 15 terminated
sleep 1; killall -s TERM ctrl-c
p 2476098
p 2476108
p 2476088
Bash c 2476085 EXITs ELAPSED 4
go main() got 17 child exited
go main() got 23 urgent I/O condition
main() children done: 1 %!s(<nil>)
main() wait...
go main() got 15 terminated
go main() got 23 urgent I/O condition
sleep 1; killall -s KILL ctrl-c
p 2476098
p 2476108
p 2476088
balmora: ~/src/my/go/doodles/sub-process [main]
$ p 2476098
p 2476108
Bash _ 2476083 EXITs ELAPSED 6
Bash q 2476084 EXITs ELAPSED 8
The bash processes keep running after the parent is killed.
killall -s KILL ctrl-c;
All 3 "zignal" sub-processes are running until killed by
killall -s KILL zignal;
In both cases the sub-processes continue to run despite main process being signaled with TERM, HUP, INT. This behavior is different in a shell environment because of convenience reasons. See the related questions about signals. This particular answer illustrates a key difference for SIGINT. Note that SIGSTOP and SIGKILL cannot be caught by an application.
It was necessary to clarify the above before proceeding with the other parts of the question.
So far you have already solved the following:
redirect stdout of sub-process to a file
set owner UID of sub-process
sub-process survives death of parent (my program exits)
the PID of sub-process can be seen by the main program
The next one depends on whether the children are "attached" to a shell or not
sub-process survives the parent being killed
The last one is hard to reproduce, but I have heard about this problem in the docker world, so the rest of this answer is focused on addressing this issue.
sub-process survives if the parent crashes and does not become a zombie
As you have noted, the Cmd.Wait() is necessary to avoid creating zombies. After some experimentation I was able to consistency produce zombies in a docker environment using an intentionally simple replacement for /bin/sh. This "shell" implemented in go will only run a single command and not much else in terms of reaping children. You can study the code over at github.
The zombie solution
the simple wrapper which causes zombies
package main
func main() {
Sh()
}
The reaper wrapper
package main
import (
"fmt"
"sync"
"github.com/fatih/color"
"github.com/hashicorp/go-reap"
)
func main() {
if reap.IsSupported() {
done := make(chan struct{})
var reapLock sync.RWMutex
pids := make(reap.PidCh, 1)
errors := make(reap.ErrorCh, 1)
go reap.ReapChildren(pids, errors, done, &reapLock)
go report(pids, errors, done)
Sh()
close(done)
} else {
fmt.Println("Sorry, go-reap isn't supported on your platform.")
}
}
func report(pids reap.PidCh, errors reap.ErrorCh, done chan struct{}) {
sprintf := color.New(color.FgWhite, color.Bold).SprintfFunc()
for ;; {
select {
case pid := <-pids:
println(sprintf("raeper pid %d", pid))
case err := <-errors:
println(sprintf("raeper er %s", err))
case <-done:
return
}
}
}
The init / sh (pid 1) process which runs other commands
package main
import (
"os"
"os/exec"
"strings"
"time"
"github.com/google/shlex"
"github.com/tox2ik/go-poc-reaper/fn"
)
func Sh() {
args := os.Args[1:]
script := args[0:0]
if len(args) >= 1 {
if args[0] == "-c" {
script = args[1:]
}
}
if len(script) == 0 {
fn.CyanBold("cmd: expecting sh -c 'foobar'")
os.Exit(111)
}
var cmd *exec.Cmd
parts, _ := shlex.Split(strings.Join(script, " "))
if len(parts) >= 2 {
cmd = fn.Merge(exec.Command(parts[0], parts[1:]...), nil)
}
if len(parts) == 1 {
cmd = fn.Merge(exec.Command(parts[0]), nil)
}
if fn.IfEnv("HANG") {
fn.CyanBold("cmd: %v\n start", parts)
ex := cmd.Start()
if ex != nil {
fn.CyanBold("cmd %v err: %s", parts, ex)
}
go func() {
time.Sleep(time.Millisecond * 100)
errw := cmd.Wait()
if errw != nil {
fn.CyanBold("cmd %v err: %s", parts, errw)
} else {
fn.CyanBold("cmd %v all done.", parts)
}
}()
fn.CyanBold("cmd: %v\n dispatched, hanging forever (i.e. to keep docker running)", parts)
for {
time.Sleep(time.Millisecond * time.Duration(fn.EnvInt("HANG", 2888)))
fn.SystemCyan("/bin/ps", "-e", "-o", "stat,comm,user,etime,pid,ppid")
}
} else {
if fn.IfEnv("NOWAIT") {
ex := cmd.Start()
if ex != nil {
fn.CyanBold("cmd %v start err: %s", parts, ex)
}
} else {
ex := cmd.Run()
if ex != nil {
fn.CyanBold("cmd %v run err: %s", parts, ex)
}
}
fn.CyanBold("cmd %v\n dispatched, exit docker.", parts)
}
}
The Dockerfile
FROM scratch
# for sh.go
ENV HANG ""
# for sub-process.go
ENV ABORT ""
ENV CRASH ""
ENV KILL ""
# for ctrl-c.go, signal.go
ENV NOSIGN ""
COPY bin/sh /bin/sh ## <---- wrapped or simple /bin/sh or "init"
COPY bin/sub-process /bin/sub-process
COPY bin/zleep /bin/zleep
COPY bin/fork-if /bin/fork-if
COPY --from=busybox:latest /bin/find /bin/find
COPY --from=busybox:latest /bin/ls /bin/ls
COPY --from=busybox:latest /bin/ps /bin/ps
COPY --from=busybox:latest /bin/killall /bin/killall
Remaining code / setup can be seen here:
https://github.com/tox2ik/go-poc-reaper
Case 5 (simple /bin/sh)
The gist of it is we start two sub-processes from go, using the "parent" sub-process binary. The first child is zleep and the second fork-if. The second one starts a "daemon" that runs a forever-loop in addition to a few short-lived threads. After a while, we kill the sub-procss parent, forcing sh to take over the parenting for these children.
Since this simple implementation of sh does not know how to deal with abandoned children, the children become zombies.
This is standard behavior. To avoid this, init systems are usually responsible for cleaning up any such children.
Check out this repo and run the cases:
$ make prep build
$ make prep build2
The first one will use the simple /bin/sh in the docker container, and the socond one will use the same code wrapped in a reaper.
With zombies:
$ make prep build case5
(…)
main() Daemon away! 16 (/bin/zleep)
main() Daemon away! 22 (/bin/fork-if)
(…)
main() CRASH imminent
panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x1 addr=0x0 pc=0x49e45c]
goroutine 1 [running]:
main.main()
/home/jaroslav/src/my/go/doodles/sub-process/sub-process.go:137 +0xfc
cmd [/bin/sub-process /log/case5 3 /bin/zleep 111 2 -- /dev/stderr 3 /bin/fork-if --] err: exit status 2
Child '1' done
thread done
STAT COMMAND USER ELAPSED PID PPID
R sh 0 0:02 1 0
S zleep 3 0:02 16 1
Z fork-if 3 0:02 22 1
R fork-child-A 3 0:02 25 1
R fork-child-B 3 0:02 26 25
S fork-child-C 3 0:02 27 26
S fork-daemon 3 0:02 28 27
R ps 0 0:01 30 1
Child '2' done
thread done
daemon
(…)
STAT COMMAND USER ELAPSED PID PPID
R sh 0 0:04 1 0
Z zleep 3 0:04 16 1
Z fork-if 3 0:04 22 1
Z fork-child-A 3 0:04 25 1
R fork-child-B 3 0:04 26 1
S fork-child-C 3 0:04 27 26
S fork-daemon 3 0:04 28 27
R ps 0 0:01 33 1
(…)
With reaper:
$ make -C ~/src/my/go/doodles/sub-process case5
(…)
main() CRASH imminent
(…)
Child '1' done
thread done
raeper pid 24
STAT COMMAND USER ELAPSED PID PPID
S sh 0 0:02 1 0
S zleep 3 0:01 18 1
R fork-child-A 3 0:01 27 1
R fork-child-B 3 0:01 28 27
S fork-child-C 3 0:01 30 28
S fork-daemon 3 0:01 31 30
R ps 0 0:01 32 1
Child '2' done
thread done
raeper pid 27
daemon
STAT COMMAND USER ELAPSED PID PPID
S sh 0 0:03 1 0
S zleep 3 0:02 18 1
R fork-child-B 3 0:02 28 1
S fork-child-C 3 0:02 30 28
S fork-daemon 3 0:02 31 30
R ps 0 0:01 33 1
STAT COMMAND USER ELAPSED PID PPID
S sh 0 0:03 1 0
S zleep 3 0:02 18 1
R fork-child-B 3 0:02 28 1
S fork-child-C 3 0:02 30 28
S fork-daemon 3 0:02 31 30
R ps 0 0:01 34 1
raeper pid 18
daemon
STAT COMMAND USER ELAPSED PID PPID
S sh 0 0:04 1 0
R fork-child-B 3 0:03 28 1
S fork-child-C 3 0:03 30 28
S fork-daemon 3 0:03 31 30
R ps 0 0:01 35 1
(…)
Here is a picture of the same output, which may be less confusing to read.
Zombies
Reaper
How to run the cases in the poc repo
Get the code
git clone https://github.com/tox2ik/go-poc-reaper.git
One terminal:
make tail-cases
Another terminal
make prep
make build
or make build2
make case0 case1
...
Related questions:
go
How to create a daemon process in Golang?
How to start a Go program as a daemon in Ubuntu?
how to keep subprocess running after program exit in golang?
Prevent Ctrl+C from interrupting exec.Command in Golang
signals
https://unix.stackexchange.com/questions/386999/what-terminal-related-signals-are-sent-to-the-child-processes-of-the-shell-direc
https://unix.stackexchange.com/questions/6332/what-causes-various-signals-to-be-sent
https://en.wikipedia.org/wiki/Signal_(IPC)#List_of_signals
Related discussions:
https://github.com/golang/go/issues/227
https://blog.phusion.nl/2015/01/20/docker-and-the-pid-1-zombie-reaping-problem/
Relevant projects:
http://software.clapper.org/daemonize/ (what I would use)
https://github.com/hashicorp/go-reap (if you must have run go on pid 1)
https://github.com/sevlyar/go-daemon (mimics posix fork)
Relevant prose:
A zombie process is a process whose execution is completed but it still has an entry in the process table. Zombie processes usually occur for child processes, as the parent process still needs to read its child’s exit status. Once this is done using the wait system call, the zombie process is eliminated from the process table. This is known as reaping the zombie process.
from https://www.tutorialspoint.com/what-is-zombie-process-in-linux
I want to use systemtap for extracting details of my linux production server. my systemtap script is
global bt;
global quit = 0
probe begin {
printf("start profiling...\n")
}
probe timer.profile {
if (pid() == target()) {
if (!quit)
{
bt[backtrace(), ubacktrace()] <<< 1
}
else
{
foreach ([sys, usr] in bt- limit 1000)
{
print_stack(sys)
print_ustack(usr)
printf("\t%d\n", #count(bt[sys, usr]))
}
exit()
}
}
}
probe timer.s(20) {
quit = 1
}
When I start run this script with command
sudo stap --ldd -d $program_name --all-modules \
-D MAXMAPENTRIES=10240 -D MAXACTION=20000 -D MAXTRACE=40 \
-D MAXSTRINGLEN=4096 -D MAXBACKTRACE=40 -x $program_pid \
profile.stp --vp 00001 > profile.out
It fails, and prints following error:
ERROR: error allocating hash
ERROR: global variable 'bt' allocation failed
WARNING: /usr/bin/staprun exited with status: 1
my production server memory info is
total used free shared buffers cached
Mem: 16008 15639 368 0 80 3090
-/+ buffers/cache: 12468 3539
I think it is enough, because in my test server, there is only 2G memory, and the systemtap script runs well for another server
Unfortunately, this is intended behavior, see my discussion here: https://sourceware.org/ml/systemtap/2015-q1/msg00033.html
The problem is that SystemTap allocates associative arrays at once (to prevent allocation failures in future) and on per-cpu basis (to prevent locking), which means that bt will require (2 * MAXSTRINGLEN + sizeof(statistic)) * MAXMAPENTRIES * NR_CPU =~ 2 Gb if NR_CPU == 128.
Reduce MAXSTRINGLEN (which is set to 4k in your case) or size of bt array:
global bt[128];
This happens if parent crashes after cloning child process, but before sending the unblocking byte with SendContinueSignalToChild(). In this case pipe file handle remains opened and child stays infinitely blocked on read(...) within WaitForContinueSignal(). After the crash, child is adopted by init process.
Steps to reproduce:
l. Simulate parent crash in google_breakpad::ExceptionHandler::GenerateDump(CrashContext *context):
...
const pid_t child = sys_clone(
ThreadEntry, stack, CLONE_FILES | CLONE_FS | CLONE_UNTRACED, &thread_arg, NULL, NULL, NULL);
int r, status;
// Allow the child to ptrace us
sys_prctl(PR_SET_PTRACER, child, 0, 0, 0);
int *ptr = 0;
*ptr = 42; // <------- Crash here
SendContinueSignalToChild();
...
Send one of the handled signal to the parent (e.g. SIGSEGV), so that the above GenerateDump(...) method is envoked.
Observe that parent exits but child still exists, blocked on WaitForContinueSignal().
Output for the above steps:
dmytro#db:~$ ./test &
[1] 25050
dmytro#db:~$ Test: started
dmytro#db:~$ ps aflxw | grep test
0 1000 25050 18923 20 0 40712 2680 - R pts/37 0:13 | | \_ ./test
0 1000 25054 18923 20 0 6136 856 pipe_w S+ pts/37 0:00 | | \_ grep --color=auto test
dmytro#db:~$ kill -11 25050
[1]+ Segmentation fault (core dumped) ./test
dmytro#db:~$ ps aflxw | grep test
0 1000 25058 18923 20 0 6136 852 pipe_w S+ pts/37 0:00 | | \_ grep --color=auto test
1 1000 25055 1687 20 0 40732 356 pipe_w S pts/37 0:00 \_ ./test
1687 is the init pid.
In the real world the crash happens in a thread parallel to the one that handles signal.
NOTE: the issue can also happen because of normal program termination (i.e. exit(0) is called in a parallel thread).
Tested on Linux 3.3.8-2.2., mips and i686 platforms.
So, my 2 questions:
Is it the expected behavior for the breakpad library to keep child alive? My expectation is that child should exit immediately after parent crashes/exits.
If it is not expected behavior, what is the best solution to finish client after parent crash/exit?
Thanks in advance!
Any clue on possible solution?
This can also happen during shutdown crash, if the crashed thread is not main, and the parent process exits from main() in exactly this time slot, so apparently it's not that unlikely to happen as it seems at a first glance.
At this moment, I think this is happening because of CLONE_FILES flag of clone() function. This leads to the situation where read() on pipe in child is not returning EOF if parent process quits.
I have not yet done the examination if we can safely get rid of this flag in clone() call.
I have a process that spawns some other processes,
I want to use the time command on a specific process and get the same output as the time command.
Is that possible and how?
I want to use the time command on a specific process and get the same output as the time command.
Probably it is enough just to use pidstat to get user and sys time:
$ pidstat -p 30122 1 4
Linux 2.6.32-431.el6.x86_64 (hostname) 05/15/2014 _x86_64_ (8 CPU)
04:42:28 PM PID %usr %system %guest %CPU CPU Command
04:42:29 PM 30122 706.00 16.00 0.00 722.00 3 has_serverd
04:42:30 PM 30122 714.00 12.00 0.00 726.00 3 has_serverd
04:42:31 PM 30122 714.00 14.00 0.00 728.00 3 has_serverd
04:42:32 PM 30122 708.00 16.00 0.00 724.00 3 has_serverd
Average: 30122 710.50 14.50 0.00 725.00 - has_serverd
If not then according to strace time uses wait4 system call (http://linux.die.net/man/2/wait4) to get information about a process from the kernel. The same info returns getrusage but you cannot call it for an arbitrary process according to its documentation (http://linux.die.net/man/2/getrusage).
So, I do not know any command that will give the same output. However it is feasible to create a bash script that gets PID of the specific process and outputs something like time outpus then
This script does these steps:
1) Get the number of clock ticks per second
getconf CLK_TCK
I assume it is 100 and 1 tick is equal to 10 milliseconds.
2) Then in loop do the same sequence of commands while exists the directory /proc/YOUR-PID:
while [ -e "/proc/YOUR-PID" ];
do
read USER_TIME SYS_TIME REAL_TIME <<< $(cat /proc/PID/stat | awk '{print $14, $15, $22;}')
sleep 0.1
end loop
Some explanation - according to man proc :
user time: ($14) - utime - Amount of time that this process has been scheduled in user mode, measured in clock ticks
sys time: ($15) - stime - Amount of time that this process has been scheduled in kernel mode, measured in clock ticks
starttime ($22) - The time in jiffies the process started after system boot.
3) When the process is finished get finish time
read FINISH_TIME <<< $(cat '/proc/self/stat' | awk '{print $22;}')
And then output:
the real time = ($FINISH_TIME-$REAL_TIME) * 10 - in milliseconds
user time: ($USER_TIME/(getconf CLK_TCK)) * 10 - in milliseconds
sys time: ($SYS_TIME/(getconf CLK_TCK)) * 10 - in milliseconds
I think it should give roughly the same result as time. One possible problem I see is if the process exists for a very short period of time.
This is my implementation of time:
#!/bin/bash
# Uses herestrings
print_res_jeffies()
{
let "TIME_M=$2/60000"
let "TIME_S=($2-$TIME_M*60000)/1000"
let "TIME_MS=$2-$TIME_M*60000-$TIME_S*1000"
printf "%s\t%dm%d.%03dms\n" $1 $TIME_M $TIME_S $TIME_MS
}
print_res_ticks()
{
let "TIME_M=$2/6000"
let "TIME_S=($2-$TIME_M*6000)/100"
let "TIME_MS=($2-$TIME_M*6000-$TIME_S*100)*10"
printf "%s\t%dm%d.%03dms\n" $1 $TIME_M $TIME_S $TIME_MS
}
if [ $(getconf CLK_TCK) != 100 ]; then
exit 1;
fi
if [ $# != 1 ]; then
exit 1;
fi
PROC_DIR="/proc/"$1
if [ ! -e $PROC_DIR ]; then
exit 1
fi
USER_TIME=0
SYS_TIME=0
START_TIME=0
while [ -e $PROC_DIR ]; do
read TMP_USER_TIME TMP_SYS_TIME TMP_START_TIME <<< $(cat $PROC_DIR/stat | awk '{print $14, $15, $22;}')
if [ -e $PROC_DIR ]; then
USER_TIME=$TMP_USER_TIME
SYS_TIME=$TMP_SYS_TIME
START_TIME=$TMP_START_TIME
sleep 0.1
else
break
fi
done
read FINISH_TIME <<< $(cat '/proc/self/stat' | awk '{print $22;}')
let "REAL_TIME=($FINISH_TIME - $START_TIME)*10"
print_res_jeffies 'real' $REAL_TIME
print_res_ticks 'user' $USER_TIME
print_res_ticks 'sys' $SYS_TIME
And this is an example that compares my implementation of time and real time:
>time ./sys_intensive > /dev/null
Alarm clock
real 0m10.004s
user 0m9.883s
sys 0m0.034s
In another terminal window I run my_time.sh and give it PID:
>./my_time.sh `pidof sys_intensive`
real 0m10.010ms
user 0m9.780ms
sys 0m0.030ms
In Linux, supposing a thread's pid is [pid], from the directory /proc/[pid] we can get many useful information. For example, these proc files, /proc/[pid]/status,/proc/[pid]/stat and /proc/[pid]/schedstat are all useful. But how can I get the CPU core number that a thread is running in? If a thread is in sleep state, how can I know which core it will run after it is scheduled again?
BTW, is there a way to dump the process(thread) list of running and sleeping tasks for each CPU core?
The "top" command may help towards this, it does not have CPU-grouped list of threads but rather you can see the list of threads (probably for a single process) and which CPU cores the threads are running on by
top -H -p {PROC_ID}
then pressing f to go into field selection, j to enable the CPU core column, and Enter to display.
The answer below is no longer accurate as of 2014
Tasks don't sleep in any particular core. And the scheduler won't know ahead of time which core it will run a thread on because that will depend on future usage of those cores.
To get the information you want, look in /proc/<pid>/task/<tid>/status. The third field will be an 'R' if the thread is running. The sixth from the last field will be the core the thread is currently running on, or the core it last ran on (or was migrated to) if it's not currently running.
31466 (bc) S 31348 31466 31348 34819 31466 4202496 2557 0 0 0 5006 16 0 0 20 0 1 0 10196934 121827328 1091 18446744073709551615 4194304 4271839 140737264235072 140737264232056 217976807456 0 0 0 137912326 18446744071581662243 0 0 17 3 0 0 0 0 0
Not currently running. Last ran on core 3.
31466 (bc) R 31348 31466 31348 34819 31466 4202496 2557 0 0 0 3818 12 0 0 20 0 1 0 10196934 121827328 1091 18446744073709551615 4194304 4271839 140737264235072 140737264231824 4235516 0 0 0 2 0 0 0 17 2 0 0 0 0 0
Currently running on core 2.
To see what the rest of the fields mean, have a look at the Linux kernel source -- specifically the do_task_stat function in fs/proc/array.c or Documentation/filesystems/stat.txt.
Note that all of this information may be obsolete by the time you get it. It was true at some point between when you made the open call on the file in proc and when that call returned.
You can also use ps, something like this:
ps -mo pid,tid,%cpu,psr -p `pgrep BINARY-NAME`
The threads are not necessary to bound one particular Core (if you did not pin it). Therefore to see the continuous switching of the core you can use (a modified answer of Dmitry):
watch -tdn0.5 ps -mo pid,tid,%cpu,psr -p \`pgrep BINARY-NAME\`
For example:
watch -tdn0.5 ps -mo pid,tid,%cpu,psr -p \`pgrep firefox\`
This can be done with top command. The default top command output does not show these details. To view this detail you will have to press f key while on top command interface and then press j(press Enter key after you pressed j). Now the output will show you details regarding a process and which processor its running. A sample output is shown below.
top - 04:24:03 up 96 days, 13:41, 1 user, load average: 0.11, 0.14, 0.15
Tasks: 173 total, 1 running, 172 sleeping, 0 stopped, 0 zombie
Cpu(s): 7.1%us, 0.2%sy, 0.0%ni, 88.4%id, 0.1%wa, 0.0%hi, 0.0%si, 4.2%st
Mem: 1011048k total, 950984k used, 60064k free, 9320k buffers
Swap: 524284k total, 113160k used, 411124k free, 96420k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ P COMMAND
12426 nginx 20 0 345m 47m 29m S 77.6 4.8 40:24.92 7 php-fpm
6685 mysql 20 0 3633m 34m 2932 S 4.3 3.5 63:12.91 4 mysqld
19014 root 20 0 15084 1188 856 R 1.3 0.1 0:01.20 4 top
9 root 20 0 0 0 0 S 1.0 0.0 129:42.53 1 rcu_sched
6349 memcache 20 0 355m 12m 224 S 0.3 1.2 9:34.82 6 memcached
1 root 20 0 19404 212 36 S 0.0 0.0 0:20.64 3 init
2 root 20 0 0 0 0 S 0.0 0.0 0:30.02 4 kthreadd
3 root 20 0 0 0 0 S 0.0 0.0 0:12.45 0 ksoftirqd/0
The P column in the output shows the processor core number where the process is currently being executed. Monitoring this for a few minutes will make you understand that a pid is switching processor cores in between. You can also verify whether your pid for which you have set affinity is running on that particular core only
top f navigation screen ( a live system example ) :
Fields Management for window 1:Def, whose current sort field is forest view
Navigate with Up/Dn, Right selects for move then <Enter> or Left commits,
'd' or <Space> toggles display, 's' sets sort. Use 'q' or <Esc> to end!
* PID = Process Id
* USER = Effective User Name
* PR = Priority
* NI = Nice Value
* VIRT = Virtual Image (KiB)
* RES = Resident Size (KiB)
* SHR = Shared Memory (KiB)
* S = Process Status
* %CPU = CPU Usage
* %MEM = Memory Usage (RES)
* TIME+ = CPU Time, hundredths
* COMMAND = Command Name/Line
PPID = Parent Process pid
UID = Effective User Id
RUID = Real User Id
RUSER = Real User Name
SUID = Saved User Id
SUSER = Saved User Name
GID = Group Id
GROUP = Group Name
PGRP = Process Group Id
TTY = Controlling Tty
TPGID = Tty Process Grp Id
SID = Session Id
nTH = Number of Threads
* P = Last Used Cpu (SMP)
TIME = CPU Time
SWAP = Swapped Size (KiB)
CODE = Code Size (KiB)
DATA = Data+Stack (KiB)
nMaj = Major Page Faults
nMin = Minor Page Faults
nDRT = Dirty Pages Count
WCHAN = Sleeping in Function
Flags = Task Flags <sched.h>
CGROUPS = Control Groups
SUPGIDS = Supp Groups IDs
SUPGRPS = Supp Groups Names
TGID = Thread Group Id
ENVIRON = Environment vars
vMj = Major Faults delta
vMn = Minor Faults delta
USED = Res+Swap Size (KiB)
nsIPC = IPC namespace Inode
nsMNT = MNT namespace Inode
nsNET = NET namespace Inode
nsPID = PID namespace Inode
nsUSER = USER namespace Inode
nsUTS = UTS namespace Inode
Accepted answer is not accurate. Here are the ways to find out which CPU is running the thread (or was the last one to run) at the moment of inquiry:
Directly read /proc/<pid>/task/<tid>/stat. Before doing so, make sure format didn't change with latest kernel. Documentation is not always up to date, but at least you can try https://www.kernel.org/doc/Documentation/filesystems/proc.txt. As of this writing, it will be the 14th value from the end.
Use ps. Either give it -F switch, or use output modifiers and add code PSR.
Use top with Last Used Cpu column (hitting f gets you to column selection)
Use htop with PROCESSOR column (hitting F2 gets you to setup screen)
To see the threads of a process :
ps -T -p PID
To see the thread run info
ps -mo pid,tid,%cpu,psr -p PID
Example :
/tmp # ps -T -p 3725
PID SPID TTY TIME CMD
3725 3725 ? 00:00:00 Apps
3725 3732 ? 00:00:10 t9xz1d920
3725 3738 ? 00:00:00 XTimer
3725 3739 ? 00:00:05 Japps
3725 4017 ? 00:00:00 QTask
3725 4024 ? 00:00:00 Kapps
3725 4025 ? 00:00:17 PTimer
3725 4026 ? 00:01:17 PTask
3725 4027 ? 00:00:00 RTask
3725 4028 ? 00:00:00 Recv
3725 4029 ? 00:00:00 QTimer
3725 4033 ? 00:00:01 STask
3725 4034 ? 00:00:02 XTask
3725 4035 ? 00:00:01 QTimer
3725 4036 ? 00:00:00 RTimer
3725 4145 ? 00:00:00 t9xz1d920
3725 4147 ? 00:00:02 t9xz1d920
3725 4148 ? 00:00:00 t9xz1d920
3725 4149 ? 00:00:00 t9xz1d920
3725 4150 ? 00:00:00 t9xz1d920
3725 4865 ? 00:00:02 STimer
/tmp #
/tmp #
/tmp # ps -mo pid,tid,%cpu,psr -p 3725
PID TID %CPU PSR
3725 - 1.1 -
- 3725 0.0 2
- 3732 0.1 0
- 3738 0.0 0
- 3739 0.0 0
- 4017 0.0 6
- 4024 0.0 3
- 4025 0.1 0
- 4026 0.7 0
- 4027 0.0 3
- 4028 0.0 7
- 4029 0.0 0
- 4033 0.0 4
- 4034 0.0 1
- 4035 0.0 0
- 4036 0.0 2
- 4145 0.0 2
- 4147 0.0 0
- 4148 0.0 5
- 4149 0.0 2
- 4150 0.0 7
- 4865 0.0 0
/tmp #