Open akiross opened 7 years ago
I'm not sure if this is something related to tmux-cpu
or if this is something related to tpm
or tmux
in general.
Have you try the same thing with tmux-plugins/tmux-battery?
ctjhoa@cl-mackbook ~ % time .tmux/plugins/tmux-cpu/scripts/cpu_percentage.sh
8.3%
.tmux/plugins/tmux-cpu/scripts/cpu_percentage.sh 0.00s user 0.00s system 0% cpu 1.007 total
As you can see get cpu percentage is very slow because you can only get CPU from a time frame of 1 second (thanks to iostat
). So maybe it have impact on the status-interval
In theory, it should be something like this:
0 1 5 6 10
start-interval start-interval
|------------------------------------------------------------------------------->|------------------------------------------------------------------------------->|
start-cpu end-cpu start-cpu end-cpu
|----------------> |---------------->
cpu_percentage.sh
is slow here as well because it uses an interval of 1, but it can be omitted: iostat -c
is sufficient to get an immediate CPU usage statistic, instead of iostat -c 1 2
as it is now in the script.
In fact, replacing iostat -c 1 2
with iostat -c
in the cpu_percentage.sh
, the issue appears to be fixed and start-interval is honoured.
With the battery plugin it appears not to have this issue, at least using battery_remain.sh
... Not knowing how tmux-plugins is implemented, I can't tell, but I think it's something structural causing this issue (i.e. in tpm).
EDIT: just to clarify, I'm on linux.
From linux iostat
man page:
The first report generated by the iostat command provides statistics concerning the time since the system was booted
so I cannot use the first report
Same from BSD iostat
man page:
The first statistics that are printed are averaged over the system uptime.
Right... So, another way has to be found :)
FYI with the iostat in sysstat-11.5.4-2.fc26.x86_64
, you can do iostat -c -y 1 1
to get the current cpu utilization. The docs for -y
: Omit first report with statistics since system boot, if displaying multiple records at given interval.
. I also had to change it to tail -n 3
instead of 2, FWIW.
I have this issue as well, and I think it's related to https://github.com/tmux/tmux/issues/797#issuecomment-282552872. Maybe, as a way to fix this, the status-interval
is checked, and if that number of seconds since the last time have not passed, just return the last value to avoid a redraw. Not the nicest solution since we have to keep track of two variables between draws, but it should work
Is this still an issue in the latest bits, and what is the current recommended workaround (OP was a few years ago).
@lonix1 iostat versions are very different across OSes and even depending on the version used. If you really care about the update frequency of your status bar, an alternative could be to add an option to customize the command used to compute CPU.
@ctjhoa Since you know much more about this than I do, what approach do you take/recommend? (I'm on latest ubuntu BTW).
@lonix1 What I mean is, if we cannot guess what iostat
version is used and there is a faster way to compute cpu percentage, we could let the user specify the command specific to his system through a tmux-cpu option.
Using eval
or such should be straightforward but I have doubt that a faster command exist for the majority of OSes.
Is this still an issue? To be clear, this should not be the case:
For example, if I set
tmux set -g status-right "%H:%M:%S" tmux set -g status-interval 5
I can see the seconds updated every 5 seconds. On the other hand, if I add the tmux_cpu plugin:
tmux set -g status-right "%H:%M:%S #(/path/to/plugin/scripts/cpu_percentage.sh)" tmux set -g status-interval 5
I will see the seconds and the CPU percentage updated almost every second, disregarding the value of status-interval. I think status-interval should be taken into account, and I see this behaviour as buggy.
Seems like it was an upstream bug but isn't a thing any more? My local tests show status-interval
is loosely respected even with just tmux set -g status-right "%H:%M:%S"
.
While testing this, I think I've found an upstream problem (with tmux
or tpm
). Basically
set -g status-right "%H:%M:%S #(sleep 1)"
set -g status-interval 5
will ignore status-interval
and result in one update per second (!)
I have another solution. I know this is ugly because
But it's faster than running the iostat command directly and it can get the cpu percentage with longer interval (ex: 5s) Anyway you can consider to add this optional method if you want :)
my cronjob
* * * * * root mkdir -p /dev/shm/tmux-cpu && mpstat 1 59 > /dev/shm/tmux-cpu/mpstat
my tmux.conf
set -g @cpu-stat-file '/dev/shm/tmux-cpu/mpstat'
modified cpu_percentage.sh cpu_percentage.sh.txt
CURRENT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
source "$CURRENT_DIR/helpers.sh"
cpu_stat_file="${TMPDIR:-/tmp}/tmux-$EUID-cpu/iostat"
cpu_stat_interval=5
get_time() {
date +"%s.%N"
}
get_stat_stat() {
local stat_file=$1
local stat_stat="$stat_file.stat"
if [ -f "$stat_stat" ]; then
local now=$(get_time)
local time=""
while read line; do
if [ -z "$time" ]; then
time=$line
else
if (( $now - $time < 1 )); then
printf "%s" "$line"
break
fi
fi
done < $stat_stat
fi
}
print_cpu_stat() {
local stat_file=$1
local stat_stat="$stat_file.stat"
local value=$(get_stat_stat $stat_file)
if [ -z "$value" ]; then
local stat_interval=$(get_tmux_option "@cpu-stat-interval" "$cpu_stat_interval")
local stat_dir=$(dirname $stat_stat)
local time=$(get_time)
value=$(awk '$NF~/[0-9.]+/ {print 100-$NF}' $stat_file | tail -n $stat_interval | awk '{n+=1;sum+=$1} END{printf("%3.1f%%",n>0?sum/n:0)}')
[ ! -d "$stat_dir" ] && mkdir -p "$stat_dir" && chmod 0700 "$stat_dir"
printf "%s\n%s" "$time" "$value" > $stat_stat
fi
printf "%s" "$value"
}
print_cpu_percentage() {
stat_file=$(get_tmux_option "@cpu-stat-file" "$cpu_stat_file")
if [ -f $stat_file ]; then
print_cpu_stat $stat_file
elif command_exists "iostat"; then
if is_linux_iostat; then
iostat -c 1 2 | sed '/^\s*$/d' | tail -n 1 | awk '{usage=100-$NF} END {printf("%3.1f%%", usage)}' | sed 's/,/./'
elif is_osx; then
iostat -c 2 disk0 | sed '/^\s*$/d' | tail -n 1 | awk '{usage=100-$6} END {printf("%3.1f%%", usage)}' | sed 's/,/./'
elif is_freebsd || is_openbsd; then
iostat -c 2 | sed '/^\s*$/d' | tail -n 1 | awk '{usage=100-$NF} END {printf("%3.1f%%", usage)}' | sed 's/,/./'
else
echo "Unknown iostat version please create an issue"
fi
elif command_exists "sar"; then
sar -u 1 1 | sed '/^\s*$/d' | tail -n 1 | awk '{usage=100-$NF} END {printf("%3.1f%%", usage)}' | sed 's/,/./'
else
if is_cygwin; then
usage="$(WMIC cpu get LoadPercentage | grep -Eo '^[0-9]+')"
printf "%3.1f%%" $usage
else
load=`ps -aux | awk '{print $3}' | tail -n+2 | awk '{s+=$1} END {print s}'`
cpus=$(cpus_number)
echo "$load $cpus" | awk '{printf "%3.1f%%", $1/$2}'
fi
fi
}
main() {
print_cpu_percentage
}
main
So, after installing the plugin I noticed that my status bar was updated too frequently. Apparently, the presence of tmux-cpu influences the update and
status-interval
is ignored.For example, if I set
I can see the seconds updated every 5 seconds. On the other hand, if I add the tmux_cpu plugin:
I will see the seconds and the CPU percentage updated almost every second, disregarding the value of status-interval. I think status-interval should be taken into account, and I see this behaviour as buggy.
I can help with the debug, but not being familiar with tmux-plugins in any way I don't know where to start.