=============================================================================== Please See PEP# 241 for more details on Open Pegasus Stress Tests: =============================================================================== Open Pegasus stress tests are expected to exert excessive loads on the CIM server. The main intent of these tests are to see how fast, or whether, we can get the CIM server to crash or hang. These tests are designed to allow some flexibility in adjusting the desired stress level on the CIM server. =============================================================================== Table of Contents: =============================================================================== 1.0 Execution model assumptions 2.0 OpenPegasus Stress Test Overview 2.1 TestStressTestController 2.2 Stress Test Clients 2.3 The Stress Test Configuration File 3.0 Default Values 4.0 Running a Stress Test using default values 5.0 Running the new "stresstest" target 6.0 Stress Test Logs 7.0 Stress Test Errors 8.0 Other Considerations =============================================================================== ******************************************************************************* 1.0 Execution model assumptions: =============================================================================== 1.1 The proposed client-side functionality concerning stress test will be tested only on Windows and Linux. Note, that the CIM server that is being stressed can be running on any of the OpenPegasus supported operating system. 1.2 The CIM server that is being stress tested is expected but not required to be running on a different system than the stress clients. This is mainly to isolate the CIM server that is under stress. 1.3 ALL OF THE PROPOSED STRESS TEST CLIENTS INVOKED BY A GIVEN EXECUTION OF THE "TestStressTestController" ARE EXPECTED TO BE STRESS TESTING THE SAME CIMSERVER. ===================================================================i=========== DIAGNOSIS OF CIM SERVER HANGS OR CRASHES CAUSED BY THE STRESS TESTS ARE EXPECTED TO BE DONE USING ALREADY EXISTING DEBUGGING AND TRACING TECHNIQUES OR TOOLS. =============================================================================== ******************************************************************************* 2.0 OpenPegasus Stress Test Overview: =============================================================================== OpenPegasus Stress Test will be comprised of three major components, as follows: 1. TestStressTestController Command that invokes one or more of the Stress Test Clients. 2. Stress Test Clients. Client Programs that may contain one or more of the CIM Operation(s). 3. Stress Test Configuration. A flat file containing test configuration information. The "stressTestController" will read the configuration file and then, invoke the "Stress Test Client" instances specified in the "Stress Test Configuration". Each of the client instances will be a separate process. These Clients will in turn connect to the same CIM Server and perform their respective operations concurrently. 2.1 TestStressTestController: ===================== This command can be run using the following options: (Usage information may be displayed using "--help" option) TestStressTestController [ -s ] [ -h hostname ] [ -p portnumber ] [ -u username ] [ -w password ] [ --help ] [ --version ] [ --verbose ] [] Options : -h - Connect to CIM Server on specified hostname --help - Display this help message -p - Connect to CIM Server on specified portnumber -s - Use SSL protocol between 'stressTestController' client and the CIM Server -u - Connect to CIM Server using the specified username --version - Display CIM Server version number --verbose - Display verbose information -w - Connect to CIM Server using the specified password Operands : - Specifies the name of the configuration file that is to be used for the tests. 2.1.1 The user may specify the name of the configuration file that is to be used for the tests. When a configuration file is not specified at command line, then the TeststressTestController command will use the default configuration file if it exists, otherwise take the default values hard coded in the TestStressTestcontroller. 2.1.2 This command will also verify the validity of the configuration values specified in the configuration file. 2.1.3 This command will invoke one or more of the Stress Test Clients specified below. 2.1.4 This command will end for any of the following reasons: - When "Duration" as specified in the Stress Test configuration file has elapsed. - If the specified configuration file cannot be found in the configuration directory. - If the configuration file has any invalid parameter or value. - A persistent failure in one (or more) of the Stress Test Clients. - The configuration can be setup to allow some margin of client failures. By default, any client failure will be reported and result in termination of the stress test (this corresponds to a ToleranceLevel of 0%). 2.1.5 When the command ends, it will as part of its cleanup also end all the stress clients that it had originally started. 2.1.6 All the options specified here on the command line will have precedence over similar parameters specified in the configuration file. 2.1.7 Log files for the controller can be found in $PEGASUS_HOME/test/StressTestController/log/ and will be named as "_stressTest_.log " where, "DD" denotes the day. "MM" denotes the month. "YY" denotes the year. "hh" denotes the hour. "mm" denotes the minutes. "ss" denotes the seconds. "controller-pid" denotes the pid of the stressTestController 2.1.8 Log file for the clients intiated by a controller can be found in $PEGASUS_HOME/test/StressTestController/log/ and will be named as _StressTestClients.log where, "controller-pid" denotes the pid of the stressTestController 2.2 Stress Test Clients: ==================== The following table lists the Stress Test Clients and the associated CIM Operations that each of these clients will perform on the CIM Server. These clients once started will run repeatedly until stopped by the stressTestController command. The expected stress factor imposed on the CIM server is directly proportional to the number of client instances running concurrently against the same CIM Server. STRESS TEST CLIENTS DESCRIPTION TestModelWalkStressClient For all the detectable namespaces this client will perform an enumeration of all Classes, Instances, References, Associators & Qualifiers. TestWrapperStressClient This is a wrapper client that will allow existing clients or programs to be existing clients or programs to be executed. The intended client program that is expected to be executed must be specified in the configuration file when using this client. The client program must ensure that it connect to the correct cimserver. For example, this client can be used to run pegasus CLI's to stress test the CIM server. Also note that the "CLI" command uses -l option to set the hostname where the CIMServer is running. Please refer to section 2.3.8(A Sample Configuration File) for an example. Note: The above list of stress test Clients can grow as needed. Please see PEP# 241 for possible future "Stress Test Clients": 2.2.1 Stress Test Clients will Connect to the CIM Server running on the specified hostname. However, if the hostname is not specified then the client will attempt to connect with the CIM server on the same system the client is invoked on. 2.2.2 "Stress Test Clients" on completion or failure of its operation as specified above will continue to repeat the operations until stopped by the "TestStressTestController". 2.3 The Stress Test Configuration File: =================================== 2.3.1 The default configuration filename for the stress test will be named as "default_stresstest.conf". 2.3.2 The user may specify the name of the configuration file that is to be used for the tests. 2.3.3 Default configuration file will be kept in $PEGASSUS_ROOT/test/StressTestController/ directory. 2.3.4 The Configuration file can contain the following common values that would in turn provide the factors required for the stress tests: 1. Duration of Tests. - Duration in minutes all the .Stress Test Client. are to be run. 2. HostName - Host name or IP address of the system where the CIM server is expected to be running. 3. Port. - Specifies Port number. The default port is 5988. 4. Number of Clients. - Number of concurrent clients. 5. Name Space. - Name space that will be utilized by the Stress Test Clients. 6. Class Name. - Name of the existing class that will be utilized by the Stress Test Clients. 7. SSL. - Use SSL protocol between Stress Client and the CIM Server. 8. UserName. - User name that the Stress Test Clients will use to connect to the server. 9. Password. - Password that is associated to the user name. 10. ToleranceLevel. - This represents the percentage of the started "Stress Test client" that can be allowed to fail in every one minute interval without resulting in the termination of the stress tests. NOTE: Any reported client failure in a one minute interval corresponds to a "failure status" for that interval. The default is 0%, which means no failures are allowed, or any failures will result in termination of the stress test. 11. ClientDuration. - Specifies duration in minutes for the corresponding stress client before it should be stopped. 12. ClientWait. - Specifies waiting period in minutes before the corresponding stress client is re-started again. "ClientDuration" when used without "ClientWait" will not be re-run. "ClientWait" will only apply when used with "ClientDuration". 13. ClientName. - The name of the Client program that will be used by a TestWrapperStressClient instance. 14. Options. - The options that are desired for the corresponding Client program that will be used by a TestWrapperStressClient. 2.3.5 Instances of the Clients may also have unique configurable values. Unique values can be specified for each instance in the configuration file as shown in the Sample Configuration File below. If any of the above parameters are repeated for a client instance, then that particular instance(s) will utilize the new value. 2.3.6 Parameters "ClientName" and "Options" are only used for "TestWrapperStressClient". 2.3.7 The parameter "Hostname" will have to be the same for all the Clients. NOTE: - Options or values specific for each client or instance must be placed after the corresponding client or instance. Otherwise, the order of the configuration variables in the configuration file is not important. - When using a Client program like a "CLI" or any other "clients" with the "TestWrapperStressClient", ensure that options associated to the Client program will connect to the same CIM Server. Example: [clientname="CLI",options="niall -l "] - Running without a failure implies that the client is communicating with the CIMServer and has recieved responses other than an exception or a CIMError. 2.3.8 A Sample Configuration File: =========================== #Machine Host of Server to be tested #If left blank then default is the local host. Hostname=narten # Total Duration of the Stress Client Test set as 300 minutes. Duration=300 # Tolerance level in percentage ToleranceLevel=0 # Number of Client instances # The format for running a client program: # = # [option1=, option2=, option3=, ..... ] # # specify all the options for a Client instance or instances within # a square bracket "[ .... ]". # # # TestWrapperStressClient can only have the parameters "ClientName" # and "Options". All other options will be ignored. # [ClientName=,Options=] ModelWalkStressClient=1 [NameSpace=/root/cimv2,ClassName=CIM_door,ClientDuration=60,ClientWait=10] ModelWalkStressClient=5 WrapperStressClient=2 [ClientName=CLI,Options="-l narten niall"] WrapperStressClient=10 [ClientName=CLI,Options="-l narten ei pg_computersystem"] WrapperStressClient=5 [ClientName=osinfo,Options="-h narten",ClientDuration=120] # Description: # The above config values sets up the stress tests for the following # stress test clients to be run concurrently: # 1. One ModelWalkStressClient which will use the Name space # "/root/cimv2" and ClassName "CIM_door" overriding any # default value. # This client will be run for 60 minutes then stopped for 10 # minutes before it is re-started again. This process will be # repeated for the duration of the stress tests. # 2. Five ModelWalkStressClient's using common values. # 3. Two WrapperStressClients will run the commands # "CLI -l narten niall" concurrently. # 4. Ten WrapperStressCleints will concurrently run the command # "CLI -l narten ei pg_computersystem". # 5. Five WrapperStressCleints will run the command "osinfo -h # narten" concurrently for 120 minutes and then stop. # # Note: # The 17 wrapper clients instances are also connecting to the # same hostname as the other clients.(Please refer to items 3, 4 # & 5 under Descriptions) # Please note that the "CLI" command uses -l option while the # "osinfo" command uses -h option to set the hostname respectively. ******************************************************************************* 3.0 Default Values: =============================================================================== 3.1 Default Hostname will be the local system where the stress tests are run. 3.2 Default port will be 5988. 3.3 Default "Duration" will be set as 180 minutes. 3.4 Default number of instances for each of the known clients is "5". 3.5 Default NameSpace when needed will be "root/cimv2". 3.6 Default UserName and Password will be blank. 3.7 Default will not use SSL. 3.8 Default percentage for "ToleranceLevel" will be set as 0%. 3.9 Default ClientWait & ClientDuration will be blank. 3.10 Default Clients for the stress tests will be set as follows: ModelWalkStressClient=5 WrapperStressClient=5 [ClientName=CLI, Options=niall] Note: The above default sets up the stress tests for the following test clients to be run concurrently: Five ModelWalkStressClient instances using default values. Five WrapperStressClient instances running the command "CLI niall". ***************************************************************************** 4.0 Running a Stress Test using default values: ============================================================================= Assumptions: CIMServer and Stress Test successfully build. $PEGASUS_ROOT, $PEGASUS_HOME and other pegasus environments are appropriately set. 4.1 Start the CIM Server on a specific hostname. 4.2 Run the following command : (Ensure that the hostname specified below matches where the above CIM Server resides.) TestStressTestController -h NOTE: If the is local host run the command without the "-h" option as follows: TestStressTestController 4.3 The "stressTestController" will search for the default file in the configuration directory. 4.4 Assuming that we are using default values but connecting to the specified host, the "stressController" will then invoke the Five ModelWalkStressClient instances using default values & the Five WrapperStressClient instances running the command "CLI niall -l " . (See section 3.0 (Default Values) above for reference) 4.5 The "TestStressTestController" will log status and other information in its log file. 4.6 Each instance of the invoked stressClients(ModelWalkStressClient & Wrapper StressClient) will connect to the specified hostname via the default port and run its operations. 4.7 Note that the stress clients will repeat these operations until it unexpectedly ends or is stopped by the "TestStressTestController". 4.8 TestStressTestController will monitor the overall status of these stress clients for every One minute interval. - If the overall stress client status is less than or equal to the tolerance level then all is well and all the client instances are allowed to chug along. - if the overall stress client status is greater than the tolerance level, then the stressTestController will terminate all the client instances and subsequently end the stress tests while reporting & loging this failure. 4.9 If all is well then the tests will continue till the end of the duration of the stress tests, which for default will be 180 minutes. ***************************************************************************** 5.0 Running the new "stresstest" target: ============================================================================= Assumptions: CIMServer successfully build. $PEGASUS_ROOT, $PEGASUS_HOME and other pegasus environments are appropriately set. 5.1 Change directory to $PEGASUS_ROOT 5.2 Run the following command. "make stresstests" - The new target "stresstests" will run the stress Tests ("stressTestController") using default configuration. The stress test programs are expected to be build already and ready to run. - Any failure, will be reported accordingly and end the tests. 5.3 Target "tests" can also be run from $PEGASUS_ROOT/test/StressTests/tests These tests will run some usage tests for the "TestStressTestController" Any failure, will be reported accordingly and end the tests. (These tests will be run as part of the nightly test.) To invoke this target: - Change directory to $PEGASUS_ROOT/test/StressTests/tests/ - Run the following command "make tests" ***************************************************************************** 6.0 Stress Test Logs: ============================================================================= Stress Test logs will contain general information about the stress tests. Most of the informations concerns configuration properties, client status, controller status, actual tolerance status, started/re-started clients, stopped clients and other general information concerning the tests. A little more detailed logs and status reports are kept when the verbose option is selected. 6.1 Log files for the Stress Tests will be saved in the following directory: $PEGASUS_HOME/test/StressTestController/log/ 6.2 Each test will generate two log files.(One for the controller and one for all the clients executed by this test.) 6.3 Log files for the controller can be found in above log directory: and will be named as "_stressTest_.log " where, "DD" denotes the day. "MM" denotes the month. "YY" denotes the year. "hh" denotes the hour. "mm" denotes the minutes. "ss" denotes the seconds. 6.4 Log file for all the clients intiated by a controller can also be found in the above log directory and will be named as _StressTestClients.log where, "controller-pid" denotes the pid of the TestStressTestController ******************************************************************************* 7.0 Stress Test Errors: =============================================================================== There are several sources of errors in the stress test. They are 7.1 Stress Test Controller errors (These failures will end the tests) 7.1.1 Invalid command options or operands. 7.1.2 Non-Existing Configuration file. 7.1.3 Syntax Error in Configuration file. 7.1.4 Invalid config property in config file. 7.1.5 Non-Existing Client PID file.(Failed to Communicate with Client) 7.1.6 Error reading Client PID File. 7.1.7 Not within tolerance Error. 7.2 Stress Test Client errors 7.2.1 Time out waiting to connect This is an exception thrown by the client. It will be logged and viewed by the controller as a failure. 7.2.2 Time out waiting for response This is an exception thrown by the client. It will be logged and viewed by the controller as a failure. 7.2.2 Failed getInstance operation for some classes(See Bugs 4957 & 4268). 7.3 Server Response Errors 7.3.1 Varied error response returned by the server. This kind of error will be logged and viewed by the controller as a failure. ******************************************************************************* 8.0 Other Considerations: =============================================================================== Compilation of the "Stress Test Clients" and the associated files will be included in the nightly build. However, testing of the "Stress Test Clients" and the associated files will NOT be included in the nightly tests.