aboutsummaryrefslogtreecommitdiffstats
path: root/docs/testing/user/userguide/opnfv_yardstick_tc056.rst
diff options
context:
space:
mode:
Diffstat (limited to 'docs/testing/user/userguide/opnfv_yardstick_tc056.rst')
-rw-r--r--docs/testing/user/userguide/opnfv_yardstick_tc056.rst149
1 files changed, 149 insertions, 0 deletions
diff --git a/docs/testing/user/userguide/opnfv_yardstick_tc056.rst b/docs/testing/user/userguide/opnfv_yardstick_tc056.rst
new file mode 100644
index 000000000..01aa99ac2
--- /dev/null
+++ b/docs/testing/user/userguide/opnfv_yardstick_tc056.rst
@@ -0,0 +1,149 @@
+.. This work is licensed under a Creative Commons Attribution 4.0 International
+.. License.
+.. http://creativecommons.org/licenses/by/4.0
+.. (c) OPNFV, Yin Kanglin and others.
+.. 14_ykl@tongji.edu.cn
+
+*************************************
+Yardstick Test Case Description TC056
+*************************************
+
++-----------------------------------------------------------------------------+
+|OpenStack Controller Messaging Queue Service High Availability |
++==============+==============================================================+
+|test case id | OPNFV_YARDSTICK_TC056:OpenStack Controller Messaging Queue |
+| | Service High Availability |
++--------------+--------------------------------------------------------------+
+|test purpose | This test case will verify the high availability of the |
+| | messaging queue service(RabbitMQ) that supports OpenStack on |
+| | controller node. When messaging queue service(which is |
+| | active) of a specified controller node is killed, the test |
+| | case will check whether messaging queue services(which are |
+| | standby) on other controller nodes will be switched active, |
+| | and whether the cluster manager on attacked the controller |
+| | node will restart the stopped messaging queue. |
++--------------+--------------------------------------------------------------+
+|test method | This test case kills the processes of messaging queue |
+| | service on a selected controller node, then checks whether |
+| | the request of the related Openstack command is OK and the |
+| | killed processes are recovered. |
++--------------+--------------------------------------------------------------+
+|attackers | In this test case, an attacker called "kill-process" is |
+| | needed. This attacker includes three parameters: |
+| | 1) fault_type: which is used for finding the attacker's |
+| | scripts. It should be always set to "kill-process" in this |
+| | test case. |
+| | 2) process_name: which is the process name of the specified |
+| | OpenStack service. If there are multiple processes use the |
+| | same name on the host, all of them are killed by this |
+| | attacker. |
+| | In this case, this parameter should always set to "rabbitmq".|
+| | 3) host: which is the name of a control node being attacked. |
+| | |
+| | e.g. |
+| | -fault_type: "kill-process" |
+| | -process_name: "rabbitmq-server" |
+| | -host: node1 |
+| | |
++--------------+--------------------------------------------------------------+
+|monitors | In this test case, two kinds of monitor are needed: |
+| | 1. the "openstack-cmd" monitor constantly request a specific |
+| | Openstack command, which needs two parameters: |
+| | 1) monitor_type: which is used for finding the monitor class |
+| | and related scritps. It should be always set to |
+| | "openstack-cmd" for this monitor. |
+| | 2) command_name: which is the command name used for request. |
+| | |
+| | 2. the "process" monitor check whether a process is running |
+| | on a specific node, which needs three parameters: |
+| | 1) monitor_type: which used for finding the monitor class |
+| | and related scripts. It should be always set to "process" |
+| | for this monitor. |
+| | 2) process_name: which is the process name for monitor |
+| | 3) host: which is the name of the node runing the process |
+| | In this case, the command_name of monitor1 should be |
+| | services that will use the messaging queue(current nova, |
+| | neutron, cinder ,heat and ceilometer are using RabbitMQ) |
+| | , and the process-name of monitor2 should be "rabbitmq", |
+| | for example: |
+| | |
+| | e.g. |
+| | monitor1-1: |
+| | -monitor_type: "openstack-cmd" |
+| | -command_name: "openstack image list" |
+| | monitor1-2: |
+| | -monitor_type: "openstack-cmd" |
+| | -command_name: "openstack network list" |
+| | monitor1-3: |
+| | -monitor_type: "openstack-cmd" |
+| | -command_name: "openstack volume list" |
+| | monitor2: |
+| | -monitor_type: "process" |
+| | -process_name: "rabbitmq" |
+| | -host: node1 |
+| | |
++--------------+--------------------------------------------------------------+
+|metrics | In this test case, there are two metrics: |
+| | 1)service_outage_time: which indicates the maximum outage |
+| | time (seconds) of the specified Openstack command request. |
+| | 2)process_recover_time: which indicates the maximum time |
+| | (seconds) from the process being killed to recovered |
+| | |
++--------------+--------------------------------------------------------------+
+|test tool | Developed by the project. Please see folder: |
+| | "yardstick/benchmark/scenarios/availability/ha_tools" |
+| | |
++--------------+--------------------------------------------------------------+
+|references | ETSI NFV REL001 |
+| | |
++--------------+--------------------------------------------------------------+
+|configuration | This test case needs two configuration files: |
+| | 1) test case file:opnfv_yardstick_tc056.yaml |
+| | -Attackers: see above "attackers" description |
+| | -waiting_time: which is the time (seconds) from the process |
+| | being killed to stoping monitors the monitors |
+| | -Monitors: see above "monitors" description |
+| | -SLA: see above "metrics" description |
+| | |
+| | 2)POD file: pod.yaml |
+| | The POD configuration should record on pod.yaml first. |
+| | the "host" item in this test case will use the node name in |
+| | the pod.yaml. |
+| | |
++--------------+--------------------------------------------------------------+
+|test sequence | description and expected result |
+| | |
++--------------+--------------------------------------------------------------+
+|step 1 | start monitors: |
+| | each monitor will run with independently process |
+| | |
+| | Result: The monitor info will be collected. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 2 | do attacker: connect the host through SSH, and then execute |
+| | the kill process script with param value specified by |
+| | "process_name" |
+| | |
+| | Result: Process will be killed. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 3 | stop monitors after a period of time specified by |
+| | "waiting_time" |
+| | |
+| | Result: The monitor info will be aggregated. |
+| | |
++--------------+--------------------------------------------------------------+
+|step 4 | verify the SLA |
+| | |
+| | Result: The test case is passed or not. |
+| | |
++--------------+--------------------------------------------------------------+
+|post-action | It is the action when the test cases exist. It will check |
+| | the status of the specified process on the host, and restart |
+| | the process if it is not running for next test cases. |
+| | |
++--------------+--------------------------------------------------------------+
+|test verdict | Fails only if SLA is not passed, or if there is a test case |
+| | execution problem. |
+| | |
++--------------+--------------------------------------------------------------+