Usage¶
Configuration¶
For OpenStack scenarios the connection is configured using standard
openrc
file (refer to Set environment variables using the OpenStack RC file
on how to retrieve it).
The config can be passed to Shaker rather by sourcing into system env source openrc
or via set of CLI parameters --os-project-name
, --os-username
, --os-password
,
--os-auth-url
and --os-region-name
.
Connection to SSL endpoints is configured by parameters --os-cacert
and --os-insecure
(to disable certificate verification). Configuration can also be specified in
config file, refer to Shaker config parameters. Config file name can be passed by parameter --config-file
.
Note
Shaker is better run under user with admin privileges. However, it’s possible to run under ordinary user too - refer to Running Shaker by non-admin user
Common Parameters¶
The following parameters are applicable for both OpenStack mode (shaker) and spot mode (shaker-spot).
- Run the scenario with defaults and generate interactive report into file report.html:
shaker --scenario <scenario> --report report.html
- Run the scenario and store raw result:
shaker --scenario <scenario> --output output.json
- Run the scenario and store SLA verification results in subunit stream file:
shaker --scenario <scenario> --subunit report.subunit
- Generate report from the raw data:
shaker-report --input output.json --output report.html
Scenario Explained¶
Shaker scenario is file in YAML format. It describes how agents are deployed (at OpenStack instances or statically) and sequence of tests to execute. When agents are deployed at OpenStack instances a reference to Heat template is provided.
description:
This scenario launches pairs of VMs in the same private network. Every VM is
hosted on a separate compute node.
deployment:
template: l2.hot
accommodation: [pair, single_room]
execution:
progression: quadratic
tests:
-
title: Iperf TCP
class: iperf_graph
time: 60
Deployment¶
By default Shaker spawns instances on every available compute node. The distribution
of instances is configured by parameter accommodation
. There are several instructions
that allow control the scheduling precisely:
pair
- instances are grouped in pairs, meaning that one can be used as source of traffic and the other as a consumer (needed for networking tests)single_room
- 1 instance per compute nodedouble_room
- 2 instances per compute nodedensity: N
- the multiplier for number of instances per compute nodecompute_nodes: N
- how many compute nodes should be used (by default Shaker use all of them *see note below)zones: [Z1, Z2]
- list of Nova availability zones to usebest_effort
- proceed even if the number of available compute nodes is less than what was requested
Examples:
As result of deployment the set of agents is produced. For networking testing this set contains
agents in primary
and minion
roles. Primary agents are controlled by shaker
tool and execute commands.
Minions are used as back-ends and do not receive any commands directly.
*If a flavor is chosen, which has aggregate_instance_extra_specs metadata set to match a host aggregate, Shaker will only use matching computes for compute_nodes calculations. If no aggregate_instance_extra_specs is set on a flavor Shaker will use all computes by default.
For example if we have 10 computes in a host aggregate with metadata special_hardware=true and use a flavor with aggregate_instance_extra_specs:special_hardware=true Shaker will only take into account the 10 matching computes, and by default try to use all of them
Execution¶
The execution part of scenario contains a list of tests that are executed one by one. By default Shaker runs the test
simultaneously on all available agents. The level of concurrency can be controlled by option progression
. There are
3 values available:
- no value specified - all agents are involved;
linear
- the execution starts with 1 agent and increases by 1 until all agents are involved;quadratic
- the execution starts with 1 agent (or 1 pair) and doubles until all agents are involved.
Tests are executed in order of definition. The exact action is defined by option class
, additional attributes are provided
by respective parameters. The following classes are available:
iperf3
- runsiperf3
tool and shows chart and statisticsflent
- runsflent
(http://flent.org) and shows chart and statisticsiperf
- runsiperf
tool and shows plain outputnetperf
- runsnetpers
tool and shows plain outputshell
- runs any shell command or process and shows plain outputiperf_graph
- runsiperf
tool and shows chart and statistics (deprecated)
Test classes¶
Tools are configured via key-value attributes in test definition. For all networking tools Shaker offers unified parameters, that are translated automatically.
iperf3, iperf, iperf_graph:¶
time
- time in seconds to transmit for, defaults to 60udp
- use UDP instead of TCP, defaults to TCPinterval
- seconds between periodic bandwidth reports, defaults to 1 sbandwidth
- for UDP, bandwidth to send at in bits/sec, defaults to 1 Mbit/sthreads
- number of parallel client threads to runhost
- the address of destination host to run the tool against, defaults to IP address of minion agentdatagram_size
- the size of UDP datagramsmss
- set TCP maximum segment size
flent:¶
time
- time in seconds to transmit for, defaults to 60interval
- seconds between periodic bandwidth reports, defaults to 1method
- which flent scenario to use, see https://github.com/tohojo/flent/tree/master/flent/tests for the whole list, defaults to tcp_downloadhost
- the address of destination host to run the tool against, defaults to IP address of minion agent
netperf:¶
time
- time in seconds to transmit for, defaults to 60method
- one of built-in test names, see http://linux.die.net/man/1/netperf for the whole list, defaults to TCP_STREAMhost
- the address of destination host to run the tool against, defaults to IP address of minion agent
shell:¶
program
- run single programscript
- run bash script
SLA validation¶
Test case can contain SLA rules that are calculated upon test completion. Every rule has 2 parts: record selector and condition. The record selector allows to filter only subset of all records, e.g. of type agent to filter records produced by a single agent. The condition applies to particular statistics.
- SLA examples:
[type == 'agent'] >> (stats.bandwidth.min > 1000)
- require min bandwidth on every agent be at least 1000 Mbit[type == 'agent'] >> (stderr == '')
- require stderr to be empty
Results of SLA validation can be obtained by generating output in subunit format. To do this a file name should be provided via –subunit parameter.