MqPerf
A benchmark of message queues with data replication and at-least-once delivery guarantees.
Source code for the mqperf article at SoftwareMill's blog: Evaluating persistent, replicated message queues
Setting up the environment
Tools
Tests have been run with the following prerequisites:
- python 3.9.5 (
via pyenv
) - ansible 2.9.5 (
pip install 'ansible==2.9.5'
) - boto3 1.17.96 (
pip install boto3
)
AWS Credentials
Message queues and test servers are automatically provisioned using Ansible on AWS. You will need to have the
AWS_ACCESS_KEY_ID
and AWS_SECRET_ACCESS_KEY
present in the environment for things to work properly, as well
as Ansible and Boto installed.
See Creating AWS access key for details.
Message generation notes
- By default, each message has length of 100 characters (this is configurable)
- For each test, we generate a pool of 10000 random messages
- Each batch of messages is constructed using messages from that pool
- Each message in the batch is modified: 13 characters from the end are replaced with stringified timestamp (TS value is used for measurement on the receiver end)
Please consider the above when configuring message size parameter in test configuration: "msg_size": 100
.
If message is too short, then majority of its content will be the TS information. For that reason, we suggest
configuring message length at 50+ characters.
Configuring tests
Test configurations are located under ansible/tests
. Each configuration has a number of parameters
that may influence the test execution and its results.
Running tests
Note: all commands should be run in the ansible
directory
Provision broker nodes with relevant script
ansible-playbook install_and_setup_YourQueueName.yml
Note: since AWS SQS is a serverless offering, you don't need to setup anything for it. For SQS, you can skip this step.
Note: you can select EC2 instance type for your tests by setting ec2_instance_type
in the group_vars/all.yml
file
Provision sender and receiver nodes
ansible-playbook provision_mqperf_nodes.yml
Note: you can adjust the number of these EC2 instances for your own tests.
WARNING: after each code change, you'll need to remove the fat-jars from the target/scala-2.12
directory and re-run
provision_mqperf_nodes.yml
.
Provision Prometheus and Grafana nodes
ansible-playbook install_and_setup_prometheus.yml
WARNING: this must be done each time after provisioning new sender / receiver nodes (previous step) so that Prometheus is properly configured to scrape the new servers for metrics
Monitoring tests
Metrics are gathered using Prometheus and visualized using Grafana.
Accessing monitoring dashboard:
- Lookup the public IP address of the EC2 node where metric tools have been deployed.
- Open
IP:3000/dashboards
in your browser - Login with
admin/pass
credentials - Select
MQPerf Dashboard
Execute test
- Choose your test configuration from the
tests
directory - Use the file name as the
test_name
in therun_tests.yml
file - Run the command
ansible-playbook run_tests.yml
Cleaning up
There are few commands dedicated to cleaning up the cloud resources after the tests execution.
- Stopping sender and receiver processing
ansible-playbook stop.yml
- Terminating EC2 instances
ansible-playbook shutdown_ec2_instances.yml
- Removing all MQPerf-related resources on AWS
ansible-playbook remove_aws_resources.yml
Utilities
- Checking receiver/sender status
ansible-playbook check_status.yml
- Running sender nodes only
ansible-playbook sender_only.yml
- Running receiver nodes only
ansible-playbook receiver_only.yml
Implementation-specific notes
Kafka
Before running the tests, create the Kafka topics by running ansible-playbook kafka_create_topic.yml
Redpanda [vectorized.io]
Redpanda requires xfs filesystem, to configure it update storage_fs_type: xfs
in all.yml
file.
Before running the tests, create the Redpanda topics by running ansible-playbook redpanda_create_topic.yml
.
Default partition number in a topic creation script is 64, if you need to adjust it update --partitions 64
param in redpanda_create_topic.yml
script.
##Redis Streams
Before running the tests, create the required streams and consumer groups by running ansible-playbook redistreams_create_streams.yml
.
This script creates streams named stream0, stream1... stream100. If you need more streams please edit the loop counter.
If you'd like to rerun tests without cluster redeployment use ansible-playbook redistreams_trim_streams.yml
to flush the streams.
To manipulate streams count use streamCount property in test JSON.
Notes. Cluster create command (last step) sometimes fails randomly. It's sometimes easier to run it directly from ec2.
Pulsar
The ack property is set on the Bookkeeper level via the CLI or REST or a startup parameter.
Go to the docs for more details.
Currently, this is not implemented, hence the mq.ack
attribute is ignored.
RabbitMQ
- when installing Rabbit MQ, you need to specify the Erlang cookie, e.g.:
ansible-playbook install_and_setup_rabbitmq.yml -e erlang_cookie=1234
- management console is available on port 15672 (
guest
/guest
) - if you'd like to SSH to the broker servers the user is
centos
- queues starting with
ha.
will be mirrored
ActiveMQ
- management console is available on port 8161 (
admin
/admin
)
ActiveMQ Artemis
- note that for the client code, we are using the same one as for ActiveMQ (
ActiveMq.scala
) - there is no dedicated management console for ActiveMQ Artemis, however monitoring is possible via exposed Jolokia web app. Jolokia web application is deployed along ActiveMQ Artemis by default. To view broker's data:
- Navigate to:
http://<AWS_EC2_PUBLIC_IP>:8161/jolokia/list
- plain JSON content should be visible - to verify if it works. - To view instance's state navigate to e.g.:
http://<AWS_EC2_PUBLIC_IP>:8161/jolokia/read/org.apache.activemq.artemis:address="mq",broker="<BROKER_NAME>",component=addresses
, where:org.apache.activemq.artemis:address="mq",broker="<BROKER_NAME>",component=addresses
is the key ("
signs are obligatory). To know other keys refer to the previous step. <BROKER_NAME>
typically resolves to AWS_EC2_PRIVATE_IP with.
replaced with_
.
- Navigate to:
- configuration changes: bumped Xmx, bumped global-max-size
EventStore
- configuration changes: see the
EventStoreMq
implementation
Oracle AQ support
-
to build the oracleaq module, first install the required dependencies available in your Oracle DB installation
- aqapi.jar (oracle/product/11.2.0/dbhome_1/rdbms/jlib/aqapi.jar)
- ojdbc6.jar (oracle/product/11.2.0/dbhome_1/jdbc/lib/ojdbc6.jar)
-
to install a dependency in your local repository, create a build.sbt file:
organization := "com.oracle"
name := "ojdbc6"
version := "1.0.0"
scalaVersion := "2.11.6"
packageBin in Compile := file(s"${name.value}.jar")
Now you can publish the file. It should be available in ~/.ivy2/local/com.oracle/
$ sbt publishLocal
Ansible notes
Zookeeper installation contains an ugly workaround for a bug in Cloudera's RPM repositories
(http://community.cloudera.com/t5/Cloudera-Manager-Installation/cloudera-manager-installer-fails-on-centos-7-3-vanilla/td-p/55086/highlight/true).
See ansible/roles/zookeeper/tasks/main.yml
. This should be removed in the future when the bug is fixed by Cloudera.
FAQ
- I'm getting: skipping: no hosts matched, why? Probably you are running ansible from project root.
Instead
cd
toansible/
(whereansible.cfg
is located) and try to run playbook from this location.
Local test
To run locally execute the Sender and Receiver classes with following:
- parameters:
-Dconfig.file=/tmp/test-config.json
- environment variables:
RUN_ID=1;HOST_ID=1