Simulation tools
It is sometimes necessary create an test environment and incur artificial load to observe how well load managershandle the load. The load simulation controller, the load simulation client, and the broker monitor were created as aneffort to make create this load and observe the effects on the managers more easily.
Simulation Client
The simulation client is a machine which will create and subscribe to topics with configurable message rates and sizes.Because it is sometimes necessary in simulating large load to use multiple client machines, the user does not interactwith the simulation client directly, but instead delegates their requests to the simulation controller, which will thensend signals to clients to start incurring load. The client implementation is in the classorg.apache.pulsar.testclient.LoadSimulationClient
.
Usage
To Start a simulation client, use the pulsar-perf
script with the command simulation-client
as follows:
pulsar-perf simulation-client --port <listen port> --service-url <pulsar service url>
The client will then be ready to receive controller commands.
Simulation Controller
The simulation controller send signals to the simulation clients, requesting them to create new topics, stop oldtopics, change the load incurred by topics, as well as several other tasks. It is implemented in the classorg.apache.pulsar.testclient.LoadSimulationController
and presents a shell to the user as an interface to sendcommand with.
Usage
To start a simulation controller, use the pulsar-perf
script with the command simulation-controller
as follows:
pulsar-perf simulation-controller --cluster <cluster to simulate on> --client-port <listen port for clients>
--clients <comma-separated list of client host names>
The clients should already be started before the controller is started. You will then be presented with a simple prompt,where you can issue commands to simulation clients. Arguments often refer to tenant names, namespace names, and topicnames. In all cases, the BASE name of the tenants, namespaces, and topics are used. For example, for the topicpersistent://my_tenant/my_cluster/my_namespace/my_topic
, the tenant name is my_tenant
, the namespace name ismy_namespace
, and the topic name is my_topic
. The controller can perform the following actions:
- Create a topic with a producer and a consumer
trade <tenant> <namespace> <topic> [—rate <message rate per second>] [—rand-rate <lower bound>,<upper bound>] [—size <message size in bytes>]
- Create a group of topics with a producer and a consumer
trade_group <tenant> <group> <num_namespaces> [—rate <message rate per second>] [—rand-rate <lower bound>,<upper bound>] [—separation <separation between creating topics in ms>] [—size <message size in bytes>] [—topics-per-namespace <number of topics to create per namespace>]
- Change the configuration of an existing topic
change <tenant> <namespace> <topic> [—rate <message rate per second>] [—rand-rate <lower bound>,<upper bound>] [—size <message size in bytes>]
- Change the configuration of a group of topics
change_group <tenant> <group> [—rate <message rate per second>] [—rand-rate <lower bound>,<upper bound>] [—size <message size in bytes>] [—topics-per-namespace <number of topics to create per namespace>]
- Shutdown a previously created topic
stop <tenant> <namespace> <topic>
- Shutdown a previously created group of topics
stop_group <tenant> <group>
- Copy the historical data from one ZooKeeper to another and simulate based on the message rates and sizes in thathistory
copy <tenant> <source zookeeper> <target zookeeper> [—rate-multiplier value]
- Simulate the load of the historical data on the current ZooKeeper (should be same ZooKeeper being simulated on)
simulate <tenant> <zookeeper> [—rate-multiplier value]
- Stream the latest data from the given active ZooKeeper to simulate the real-time load of that ZooKeeper.
stream <tenant> <zookeeper> [—rate-multiplier value]
The "group" arguments in these commands allow the user to create or affect multiple topics at once. Groups are createdwhen calling thetrade_group
command, and all topics from these groups may be subsequently modified or stoppedwith thechange_group
andstop_group
commands respectively. All ZooKeeper arguments are of the formzookeeper_host:port
.
Difference Between Copy, Simulate, and Stream
The commands copy
, simulate
, and stream
are very similar but have significant differences. copy
is used whenyou want to simulate the load of a static, external ZooKeeper on the ZooKeeper you are simulating on. Thus,source zookeeper
should be the ZooKeeper you want to copy and target zookeeper
should be the ZooKeeper you aresimulating on, and then it will get the full benefit of the historical data of the source in both load managerimplementations. simulate
on the other hand takes in only one ZooKeeper, the one you are simulating on. It assumesthat you are simulating on a ZooKeeper that has historical data for SimpleLoadManagerImpl
and creates equivalenthistorical data for ModularLoadManagerImpl
. Then, the load according to the historical data is simulated by theclients. Finally, stream
takes in an active ZooKeeper different than the ZooKeeper being simulated on and streamsload data from it and simulates the real-time load. In all cases, the optional rate-multiplier
argument allows theuser to simulate some proportion of the load. For instance, using —rate-multiplier 0.05
will cause messages tobe sent at only 5%
of the rate of the load that is being simulated.
Broker Monitor
To observe the behavior of the load manager in these simulations, one may utilize the broker monitor, which isimplemented in org.apache.pulsar.testclient.BrokerMonitor
. The broker monitor will print tabular load data to theconsole as it is updated using watchers.
Usage
To start a broker monitor, use the monitor-brokers
command in the pulsar-perf
script:
pulsar-perf monitor-brokers --connect-string <zookeeper host:port>
The console will then continuously print load data until it is interrupted.