• date +"%T.%N" returns the current time with nanoseconds.

  06:46:41.431857000
  

• date +"%T.%6N" returns the current time with nanoseconds rounded to the first 6 digits, which is microseconds.

  06:47:07.183172
  

• date +"%T.%3N" returns the current time with nanoseconds rounded to the first 3 digits, which is milliseconds.

  06:47:42.773
  

In general, every field of the date command's format can be given an optional field width.

Nano is 10⁻⁹ and milli 10⁻³. Hence, we can use the three first characters of nanoseconds to get the milliseconds:

date +%s%3N

From man date:

%N nanoseconds (000000000..999999999)

%s seconds since 1970-01-01 00:00:00 UTC

Source: Server Fault's How do I get the current Unix time in milliseconds in Bash?.

On OS X, where date does not support the %N flag, I recommend installing coreutils using Homebrew. This will give you access to a command called gdate that will behave as date does on Linux systems.

brew install coreutils

For a more "native" experience, you can always add this to your .bash_aliases:

alias date='gdate'

Then execute

$ date +%s%N

Here is a somehow portable hack for Linux for getting time in milliseconds:

#!/bin/sh
read up rest </proc/uptime; t1="${up%.*}${up#*.}"
sleep 3    # your command
read up rest </proc/uptime; t2="${up%.*}${up#*.}"

millisec=$(( 10*(t2-t1) ))
echo $millisec

The output is:

3010

This is a very cheap operation, which works with shell internals and procfs.

date command didnt provide milli seconds on OS X, so used an alias from python

millis(){  python -c "import time; print(int(time.time()*1000))"; }

OR

alias millis='python -c "import time; print(int(time.time()*1000))"'

EDIT: following the comment from @CharlesDuffy. Forking any child process takes extra time.

$ time date +%s%N
1597103627N
date +%s%N  0.00s user 0.00s system 63% cpu 0.006 total

Python is still improving it's VM start time, and it is not as fast as ahead-of-time compiled code (such as date).

On my machine, it took about 30ms - 60ms (that is 5x-10x of 6ms taken by date)

$ time python -c "import time; print(int(time.time()*1000))"
1597103899460
python -c "import time; print(int(time.time()*1000))"  0.03s user 0.01s system 83% cpu 0.053 total

I figured awk is lightweight than python, so awk takes in the range of 6ms to 12ms (i.e. 1x to 2x of date):

$ time awk '@load "time"; BEGIN{print int(1000 * gettimeofday())}'
1597103729525
awk '@load "time"; BEGIN{print int(1000 * gettimeofday())}'  0.00s user 0.00s system 74% cpu 0.010 total

Pure bash solution

Since bash 5.0 (released on 7 Jan 2019) you can use the built-in variable EPOCHREALTIME which contains the seconds since the epoch, including decimal places down to micro-second (μs) precision (echo $EPOCHREALTIME prints something like 1547624774.371215). By removing the . and the last three places we get milliseconds:

Either use

(( t = ${EPOCHREALTIME/./} / 1000 ))

or something like

t=${EPOCHREALTIME/./}  # remove the dot (s → µs)
t=${t%???}             # remove the last three digits (µs → ms)

Either way t will be something like 1547624774371.

The other answers are probably sufficient in most cases but I thought I'd add my two cents as I ran into a problem on a BusyBox system.

The system in question did not support the %N format option and doesn't have no Python or Perl interpreter.

After much head scratching, we (thanks Dave!) came up with this:

adjtimex | awk '/(time.tv_sec|time.tv_usec):/ { printf("%06d", $2) }'

It extracts the seconds and microseconds from the output of adjtimex (normally used to set options for the system clock) and prints them without new lines (so they get glued together). Note that the microseconds field has to be pre-padded with zeros, but this doesn't affect the seconds field which is longer than six digits anyway. From this it should be trivial to convert microseconds to milliseconds.

If you need a trailing new line (maybe because it looks better) then try

adjtimex | awk '/(time.tv_sec|time.tv_usec):/ { printf("%06d", $2) }' && printf "\n"

Also note that this requires adjtimex and awk to be available. If not then with BusyBox you can point to them locally with:

ln -s /bin/busybox ./adjtimex
ln -s /bin/busybox ./awk

And then call the above as

./adjtimex | ./awk '/(time.tv_sec|time.tv_usec):/ { printf("%06d", $2) }'

Or of course you could put them in your PATH

EDIT:

The above worked on my BusyBox device. On Ubuntu I tried the same thing and realised that adjtimex has different versions. On Ubuntu this worked to output the time in seconds with decimal places to microseconds (including a trailing new line)

sudo apt-get install adjtimex
adjtimex -p | awk '/raw time:/ { print $6 }'

I wouldn't do this on Ubuntu though. I would use date +%s%N

To show date with time and time-zone

date +"%d-%m-%Y %T.%N %Z"

Output : 22-04-2020 18:01:35.970289239 IST

I want to generate value from bash and use that value in Java code to convert back to date(java.util).

Following command works for me to generate the value in bash file:

date +%s000

I just wanted to add to Alper's answer what I had to do to get this stuff working:

On Mac, you'll need brew install coreutils, so we can use gdate. Otherwise on Linux, it's just date. And this function will help you time commands without having to create temporary files or anything:

function timeit() {
    start=`gdate +%s%N`
    bash -c $1
    end=`gdate +%s%N`
    runtime=$(((end-start)/1000000000.0))
    echo " seconds"
}

And you can use it with a string:

timeit 'tsc --noEmit'

When you use GNU AWK since version 4.1, you can load the time library and do:

$ awk '@load "time"; BEGIN{printf "%.6f", gettimeofday()}'

This will print the current time in seconds since 1970-01-01T00:00:00 in sub second accuracy.

the_time = gettimeofday() Return the time in seconds that has elapsed since 1970-01-01 UTC as a floating-point value. If the time is unavailable on this platform, return -1 and set ERRNO. The returned time should have sub-second precision, but the actual precision may vary based on the platform. If the standard C gettimeofday() system call is available on this platform, then it simply returns the value. Otherwise, if on MS-Windows, it tries to use GetSystemTimeAsFileTime().

_{source: GNU awk manual}

On Linux systems, the standard C function getimeofday() returns the time in microsecond accuracy.

Perl can be used for this, even on exotic platforms like AIX. Example:

#!/usr/bin/perl -w

use strict;
use Time::HiRes qw(gettimeofday);

my ($t_sec, $usec) = gettimeofday ();
my $msec= int ($usec/1000);

my ($sec,$min,$hour,$mday,$mon,$year,$wday,$yday,$isdst) =
    localtime ($t_sec);

printf "%04d-%02d-%02d %02d:%02d:%02d %03d\n",
    1900+$year, 1+$mon, $mday, $hour, $min, $sec, $msec;

A Python script like this:

import time
cur_time = int(time.time()*1000)

To print decimal seconds:

start=$(($(date +%s%N)/1000000)) \
    && sleep 2 \
    && end=$(($(date +%s%N)/1000000)) \
    && runtime=$((end - start))

divisor=1000 \
    && foo=$(printf "%s.%s" $(( runtime / divisor )) $(( runtime % divisor ))) \
    && printf "runtime %s\n" $foo # in bash integer cannot cast to float

Output: runtime 2.3