This is a network tester playbook, which uses HP Netperf and Flent under the hood. The test is based on a mixed TCP/UDP/ICMP workload that is executed over three different client-server set configurations (1xM, Nx1, NxM). This software requires ansible >= 2.4 and python 3.6. The playbook has tested on the following OS platforms: RHEL/CentOS 7.x, Ubuntu 18.04, Rocky 9.
Install ansible packages:
$ ansible-galaxy collection install -r requirements.yaml
All further dependencies will be installed by the install role of the playbook. Be aware that this includes adding package repositories and installing software from them. See the network_diag_install tasks and defaults for more details.
Before you run the test make sure you can access all hosts via SSH. The user needs to be able to do password-less sudo.
We recommend using SSH keys for authentication, although passwords also can be used. You may rely on your current SSH client configuration or provide authentication data explicitly in the file group_vars/netdiag_hosts.yaml. Refer to Ansible documentation for further information. Log in once to each host to add hosts to trusted
hosts.
Make sure that the firewall configuration of all hosts allows pings and TCP connections on ports in the range of 5000-65536.
Host are divided into two sets: clients and servers. The sets may overlap, but since this playbook is about network testing disjoint set configuration is recommended. Populate inventory file accordingly. For a typical Quobyte installation, you would run tests between actual client and server machines, and second test between one half of the servers as clients, and the other half as servers.
To verify configuration is correct run the following:
$ ansible -i inventory_file -m ping -b all
The command must produce no errors.
The playbook includes a Netperf RPM package for RHEL/CentOS 7.x for your convenience. If you would like to provide your own binaries, bear in mind --enable-demo option must be supported by Netperf. In that case set netperf_install: False.
Due to extensive use of sockets a user on behalf which tests are executed should be granted very high or unlimited file descriptor limits on target hosts. This is typically done by the ansible command execution automatically.
To install dependencies, run the test, and clean up aftewards in one go, run:
$ ansible-playbook -i inventory_file netdiag.yaml
You can also split that into three phases:
$ ansible-playbook -i inventory_file netdiag.yaml -t install
$ ansible-playbook -i inventory_file netdiag.yaml -t test
$ ansible-playbook -i inventory_file netdiag.yaml -t cleanup
Test results are stored on an ansible host at /tmp/quobyte_netdiag_data.
Current and prospective Quobyte customers may create a ticket and upload test results for further analysis.
A stripped down version of the netdiag playbook can be used for host stack monitoring under production load. The name of the relevant playbook is get.kernel.data.yaml.
$ ansible-playbook -i inventory_file netdiag.yaml -t install
$ ansible-playbook -i inventory_file get.kernel.data.yaml -e stats_interval=120
$ ansible-playbook -i inventory_file netdiag.yaml -t cleanup
The variable stats_interval defines period between two consecutive diagnostic probes launched on target hosts. By default it is 60 seconds.
Product names and brands found in the repository are property of their respective owners. The playbook is distributed under Apache License 2.0.