garabik / polynice Goto Github PK

written by Radovan Garabík
http://kassiopeia.juls.savba.sk/~garabik/software/polynice.html

polynice is a nice(1)-like utility to throttle long running processes
beyond what can be achieved by nice(1), by repeatedly suspending and
resuming the process.

It can also lower scheduler and I/O priorities, emulating nice(1) and
ionice(1), and terminate the processes after given time, emulating
timeout(1).


Requirements
------------

polynice is written in python3 and runs mainly on Linux systems, but has
been also tested on Mac OS X (Darwin).

If psutil is installed, polynice will take advantage of it and children
detection will be multiplatform. This will also enable nice(1) and ionice(1)
functionality. If psutil is not installed, polynice will try to use nice(1) and
ionice(1) to achieve the same functionality.

There is some python2 compatibility, but not all functionality is guaranteed.

Usage
-----

You can use polynice either to run a command (in this case it mimics nice(1)
usage), or to specify a PID of a running process.

Throttling is done by periodically suspending (via SIGSTOP) and resuming (via
SIGCONT) the process being throttled.

The parameter -n A/B is used to specify the throttling ratio. The process will
be suspended for A seconds, then it will run for B seconds, then again
suspended for A seconds etc... Floating point values can be used, e.g. -n 0.4/2
will make the process run 2 seconds and suspend for 0.4 seconds out of 2.4
second time interval.

If a single number is specified, B is taken to be 1, e.g. -n 3 will suspend the
process for 3 seconds and run it 1 second.

If 0 is used, do not throttle (useful in combination with -a or --term-after
options).

The default is 1/4, i.e. run the process 80% of the time, suspend it every 4th
second for one second.

Using the parameter --mode=random changes the semantics of throttling -- the
process will be watched in B sized time interval, and for each of the
intervals, it will be suspended with probability p=A/(A+B) (and run with
probability 1-p). Statistically, the stop/run ratio will approximate the
default (--mode=fixed) ratio if run sufficiently long.

polynice is able to replace to a certain extent nice, renice and ionice
commands. It accepts following compatibility options:

-c IOCLASS, --class IOCLASS
Sets  process I/O scheduling class (the same as ionice): 0: none,
1: realtime, 2: best-effort, 3: idle. The  same  as  the  ionice
parameter -c.

--ion N, --classdata N
Specifies  the  I/O  scheduling class data. The same as the ionice
parameter -n.

--nn N
Specifies the scheduling priority of the process. The same as  the
renice parameter -n.


polynice can replace somewhat the timeout(1) command and accepts these options:

--term-after DURATION
If the controlled program is running after DURATION seconds, terminate
it (with TERM signal).

--kill-after DURATION
Also send a KILL signal if the program is still running this long after
the --term-after time expired and the TERM signal has been sent.

Other options:

--mode {random,fixed}
Throttling mode. If fixed (the default), stop and re-run processes in fixed
time intervals. If random, stop or run the processes with probability
corresponding to the niceness.

-a
In addition to throttling, set maximum possible nice values (equivalent
to ionice -c 3 nice -n 20)

You can use the -v parameter to increase verbosity - print to stderr $ when
process is stopped and = when run.

Use -vv to be more verbose, print timestamp for each stop/run action
to stderr. Use mostly with the -p option, otherwise it
will clobber screen output.

Use -vvv to be even more verbose. Useful for debugging.


Caveats
-------

polynice uses only pid's to identify processes, which can lead to a race
condition - if the child's pid is reused in the time between creation of list
of children and action (stopping or resuming), an incorrect process might be
throttled (but at most for one control period).


Examples
--------

$ polynice -n 2/0.5 scp server:bigfile local/

will copy `bigfile' from the remote server, repeatedly suspending the scp
command for 2 seconds, running it for 0.5 seconds, thus reducing the overall
download speed to one fifths and keeping the network connection responsive.


$ polynice -vv -n 3 -p 12345

will run the process (and all its children) with pid 12345 for 1 second,
suspend it for 3 seconds, i.e. keeping CPU load to one fourth, thus reducing
its power consumption and overheating.

$ polynice program

just use the default, run the program 80% of the time in 1 second intervals.


$ polynice -a program

will use the defaults as well, run the program 80% of the time in 1
second intervals, but in addition it will give it the lowest possible
scheduler priority (equivalent to running `nice 20 program') and the
lowest possible I/O priority (equivalent to running `ionice -c 3
program').