PYLOGEN

PYthon LOad GENerator

Function

A quick experimental load injector

Based on python multiprocessing module to avoid GIL effect.

A run launches a main, that launches subprocess (runners, workers) for CUT processing

A CUT is a subclass of ClassUnderTest.

A run may run as

closed model processing : driver : predefined workers count A runner runs loops of calls to the CUT (see parameters loop, duration ...). The --proc parameter controls the count of runners.
opened model processing : driver : events arrivals rate dynamically creates wand adjusts workers count

A CUT executes requests that may be embedded in transactions.

Stats are displayed on STDIN.

The stats are processed thru a single queue. One line per event (request or transaction), with local data (ex : CUT process throughput, request time) and global datas (global throughput).

A summary with high water marks is displayed at end of run If queue length highwatermark is high (hint : > 4 * cores as a rule of thumb), records could not be poped as quickly as required .. : throughput biased

Important :

It's python : may be sufficient for Kms testing. Beware of cpu
Global stat may be biased if queure reader process is too low. Local not !
Have a look at the sum of local throughputs vs global one : they should be consistent : global ~ process * local
global count and throughput count only requests, not transactions
a transaction and its requests have the same transaction id (None if no transaction)

Installation

Python venv

python3 -m venv venv
. venv/bin/activate
pip3 install --upgrade pip
pip3 install -r requrements.txt

Then

. venv/bin/activate

Usage

General Pylogen

Please see Pylogen.py section for new (but not fully tested) functionnalities

usage: Pylogen.py {action} [-v] [--process PROCESS] [--postpone POSTPONE] [--rampup RAMPUP] [--duration DURATION] [--loops LOOPS] [--lengths LENGTHS] [--extra EXTRA] [--pauseloop PAUSELOOP] [--pauselen PAUSELEN] [--summary SUMMARY] [--outformat OUTFORMAT]

action : either

name of a CUT
keyword "scenario"

Note : if "scenario", none of the following are usable, except --file

-v : verbosity -vv _vvv ....
--postpone POSTPONE : postpones the start of the run (permits to shape a load)
--rampup RAMPUP : time between each process start
--lengths LENGTHS : array of lengths (for exemple). Loop on length in a loop execution
--extra EXTRA : parameters for plugins
--pauselen PAUSELEN : pause between lengths
--summary SUMMARY : time between 2 summary computations
--outformat OUTFORMAT : csv (short) of anything (long)
--id ID : an arbitrary ID

--file FILE : name of scenario file

Closed model

By default, closed model : a predefined number of worker process running cuts is given

--process PROCESS : number of process (to set in accordance with the expected load and the capacities of the server
--duration DURATION : duration of the process (vs loops). Prioritary
--loops LOOPS : how many loops per process
--pauseloop PAUSELOOP : pause between loops

Opened model

Opened model : a generator generates events at a given frequency. These events are processed by workers A controller adjusts the number of worker to process as quick as possible

--openedmodel : use openedmodel --controllerDelay : frequency for the controller to adjust number of process --generatorDelay : ? --sechedule : schedule : example 30@10,20@15 : during 30 seconds generate 10 runs/second then during 20 seconds 15 runs/second --trigger : if more than trigger events in the jobqueue, start a new worker --prefork : number of workers to start before generation begins --decrease : number of consecutive times the controller found no job in queue : then kills a worker to spare resources --burst : non-linear : at a given time a burst of events will be generated

The max size of a burst can be controlled --- burst = 0 : burst has random size, max equals to mean requested throughput --- burst = n : burst has random size, max equals to n --- burst = f : burst has fixed size

burst = f1 is the same as burst = 1 !

After a burst of size n, the n-1 following triggers won't generate any events

Scenario

A file with 1 line per CUT. Useful to mix operations at the same time. May be used to shape a load profile with postpone starts and durations

Sample : python3 Pylogen.py scenario --file Scenario.txt

Samples Pylogen

Hereunder Dummy and KmsEncrytSymmetric are class names (in .py files)

python3 Pylogen.py cuts.Dummy --le 384 --pauselen 0 --pauselo 0.001 --duration 10 --ram 0 --proc 2
python3 Pylogen.py cuts.KmsEncrytSymmetric --defaults myprofile.json --le 384,512,1024 --pauselen 0 --pauseloop 0 --duration 10 --proc 1 --extra '{"file":"KmsDefault.json","key":"symmetric1"}'
python3 Pylogen.py cuts.Dummy -vvv --open --pre 2 --sche "5@1,10@0.1,5@1,30@0.1" --len 0 --out x --contr 0.5 --extra '{"sleep":1}'
python3 Pylogen.py cuts.Sbgc01 -vvv --open --pre 2 --sche "5@1" --len 0 --out x --contr 0.5 --extra '{"sleep":1}'
python3 Pylogen.py cuts.Sbgc01 -vvv --open --pre 2 --sche "5@1" --len 0 --out x --contr 0.5 --extra '{"sleep":1}'

Beware : use --defaults or --def , not -d nor --defa nor --defau ...

CUTS

Integrate

Cuts are imported dynamically (importlib). Put them in cuts directory !!!

Design

A CUT inherits ClassUnderTest
A CUT contains requests
A CUT may contain a transaction (requestsManager) that groups the requests
be careful, give relevant names (id, requestsManager and requests names) to interpret output
CUT should be written with @Executor.exec decorator

cuts options : extra

Options are a json string in --extra with 2 keys :

file : name of a json file containing keys definition. Optionnal : default KmsDefault.json
key : the name of the key in file Sample :
--extra '{"key":"symmetric1"}'
--extra '{"file":"MyKmsDefault.json","key":"symmetric1"}'

GaspardZeBear/pylogen