Testing Service Oriented Architecture Based Web Applications

By Bernard Stepien, Liam Peyton, Pulei Xiong, Pierre Seguin

SOA based web applications

Catalog service User 1 User n WEB application invoice service

Web formatting service

Testing approaches
1. Black box testing from a users perspective 2. Black box testing from a services perspective (a kind of unit testing) 3. Black box testing from a user and services perspectives. 4. Grey box testing from an integration testing perspective (in-process or logbased)
3

Test purposes
For a given set of web requests, receive the correct web responses with appropriate quality of service. If the above fails, in a SOA based system the cause could be located:
In the web application logic:
Wrong processing within the web application. Web application sent the wrong request to the service.

In the service logic:

Wrong processing within the service

In the SOA infrastructure

Unable to respond with appropriate quality of service
4

Black box testing from a users perspective

The SUT is observed only via the web application. The services are not visible to the tester

TTCN-3 test suite

SUT WEB application

Black box testing from a services perspective

The SUT is the collection of services The ATS simulates the web application
SUT Catalog service

TTCN-3 test suite

invoice service

Web formatting service

6

First two approaches solutions

Already covered in previous research:
Web testing:
http://www.site.uottawa.ca/~bernard/Testing_a_servlet.pdf

SOAP application testing:

http://www.site.uottawa.ca/~bernard/TestingWebServices.pdf

Problems with the black box separate approaches

With the first approach (web application testing), if there is a failure we wont be able to determine its cause accurately:
In the web application In the underlying services

With the second approach, we may conclude that the services are OK, but we can not guarantee that:
the web application places the correct service requests. the web application processes correctly the service responses.
8

Black box testing from a user and services perspective

The SUT is the web application only The web application is disconnected from the real services. A TTCN-3 parallel test component emulates the services messaging.
TTCN-3 test suite

TTCN-3 Web application Handler component

SUT WEB application

TTCN-3 services Handler components 9

Advantages of approach 3
This approach fully verifies that the web application sends the correct requests to the services, but only according to the testers opinion. Once this verified, if the web response to the user is still wrong, we can then conclude with confidence that the problem is located in the web application processing.
10

Grey box testing from an integration testing perspective

All messages between all components are tested. The test suite intercepts the communication between the web application and the services and verifies them.
TTCN-3 test suite

Web application Handler component

Catalog service Handler component

Invoice service Handler component

Html formatting service Handler component

SUT SUT WEB server

Catalog service

invoice service

Html formatting 11 service

Reasons for integration testing

Unit testing does not guarantee that components will work together correctly. Multi-user traffic could reveal faults:
In the web application (mixing client side data between users) In the services (mixing server side data between responses and time outs due to load) Inability of the SOA infrastructure to deliver responses with an appropriate quality of service

12

Re-using unit testing Test suite elements

The test suite elements of the two first approaches can be fully re-used for the two last approaches. The only differences are:
the test configuration merging some behavior.

13

Implementation of Grey box SOA testing

testcase SOABasedWebTesting() runs on MTCType system SystemComponentType { var SOAComponentType theSOAComponent; var UserComponentType theUserComponent[2]; theUserComponent[0] := UserComponentType.create; theUserComponent[1] := UserComponentType.create; theSOAComponent := SOAComponentType.create; // map all ports here theSOAComponent.start(serviceEventsTest()); theUserComponent[0].start(User_1_events()); theUserComponent[1].start(User_2_events()); theUserComponent[0].done; theUserComponent[1].done; servCoordPort.send("end test"); all component.done; log("testcase SOABasedWebTesting completed"); } 14

User behavior specification

function User_1_behavior() runs on UserComponentType { var ResponseType theResponse; timer theTimer; var integer i; userWebPort.send(web_req_A); theTimer.start(5.0); alt { [] userWebPort.receive(web_resp_B) { log("user 1 has received web_resp_B"); theTimer.stop; setverdict(pass) } [] userWebPort.receive(ResponseType:?) -> value theResponse { log("user 1 has receive wrong response: " & theResponse); setverdict(fail) } [] theTimer.timeout { log("User 1 timed out"); setverdict(inconc) }

} }
15

Testing challenges
In Web pages, data is mixed with formatting information. Caching: Requests to the web application do not produce always a request to an underlying service. Service messages can not always be correlated directly to a specific users web applications messages.
16

Testing caching
Caching in the web application results in a non event at the service level. How can we test that an event has not occurred? Answer: check if the cached event occurs, and if yes, set the verdict to fail. This requires a TTCN-3 implementation for:
Representing a caching mechanism Representing the non-event detection
17

Caching testing implementation

type component SOAComponentType { integer nbRequests = 0; var RequestsType cachedRequests := {}; } function serviceEventsTest() runs on SOAComponentType { alt { [] soaWebPort.receive(service_req_A) -> value incomingMsg { if(isNotCached(incomingMsg.theRequest)) { updateCache(theRequest); servicePort.send(incomingMsg.theRequest); chek_for_response_B(incomingMsg.theSessionId); serviceEventsTest() } else { log("has received a cached message_A"); setverdict(fail); stop }

[]
} 18

TTCN-3 features for caching

Caching is very easy to implement in TTCN-3 because:
Easy building of lists or sets containing cached complex messages. Easy lookup of the cache due to the powerful TTCN-3 matching mechanism for complex types. Possibility to select messages subject to caching.
function isNotCached(RequestType theRequest) runs on SOAComponentType return boolean { var integer i; for(i:=0; i < nbRequests; i:=i+1) { if(match(theRequest, cachedRequests[i])) { return false; } } return true; } 19

Correlation gap handling

Temporal ordering problem. The web application may place its requests to the service in a different order as received from the users. Potential lack of indicators to assign a service request to a specific user. End-to-end tracking may work for a single user, but does not work in the case of multiple users.
20

10

Single user end to end tracking message flow

User 1 Web application
Web request A service request A

service

service response B Web response B

21

Single user end to end tracking testing architecture

TTCN-3 User 1 test component
Web request A service request A service request A

Web application

TTCN-3 Service test component

service

service response B service response B Web response B

22

11

Single user TTCN-3 behavior implementation

Simplified behavior using only the MTC

userWebPort.send(web_req_A); soaWebPort.receive(service_req_A) -> value incomingMsg { servicePort.send(incomingMsg.theRequest); //service req A servicePort.receive(service_resp_B) soaWebPort.send(service_resp_B) to incomingMsg.theSessionId userWebPort.receive(web_resp_B) { setverdict(pass) 23

Multiple user message flow

ideal case
TTCN-3 User 1 test component TTCN-3 User 2 test component
Web request A service request A service request A service response B service response B Web response B

Web application

TTCN-3 Service test component

service

Web request C service request C service request C service response D service response D Web response D

24

12

Multiple user message flow

one of many realistic cases
TTCN-3 User 1 test component TTCN-3 User 2 test component
Web request A service request A service request A Web request C service response B service request C service request C service response B service response D service response D Web response D Web response B

Web application

TTCN-3 Service test component

service

25

Multiple user handling

Each user is portrayed by a TTCN-3 parallel test component. The service handler is portrayed by a single TTCN-3 parallel test component. For each service request received the service handler performs two kinds of checking: It checks if such a message was expected, if yes, then forwards it to the service It enforces the expected response from the service and if successful forwards the service response to the web application. The MTC tells to the service handler what requests to expect but not in which order. This is handled in a template: template RequestsType expectedRequests := { "service_req_A", "service_req_C"}; At the end of the test, the service handler checks if the set of messages it was told to expect by the MTC matches the set of actually received messages. if(match(expectedRequests, receivedRequests)) { setverdict(pass); } else { setverdict(fail); };

26

13

Service test component re-usability

There are two ways to implement this architecture:
Hard code the service component with each expected alternative message in a function. Make a generic service message handler that checks if a received message is present in the expected messages list.

27

Hard coded Service handler

function serviceEventsTest(RequestsType expectedRequests) runs on SOAComponentType { var ServiceRequestWrapperType incomingMsg; alt { [] soaWebPort.receive(service_req_A) -> value incomingMsg { servicePort.send(incomingMsg.theRequest); check_response_B(incomingMsg.theSessionId); serviceEventsTest(expectedRequests) } [] soaWebPort.receive(service_req_C) -> value incomingMsg { servicePort.send(incomingMsg.theRequest); check_response_D(incomingMsg.theSessionId); serviceEventsTest(expectedRequests) } [] servCoordPort.receive("end test") { if(match(expectedRequests, receivedRequests)) { log("the expected service requests set does match the actual received requests"); setverdict(pass); } else { log("the expected service requests set does NOT match the actual received requests"); setverdict(fail); }; stop } 28

14

Generic Service handler

function serviceEventsTest(RequestsType expectedRequestResponses) runs on SOAComponentType { var ServiceRequestWrapperType incomingMsg; var ServiceResponse correspondingResponse; alt { [] soaWebPort.receive(?:ServiceRequestType) -> value incomingMsg { if(wasExpected(incomingMsg.theRequest, expectedRequestResponses) { servicePort.send(incomingMsg.theRequest); correspondingResponse := getCorrespondingResponse(incomingMsg.theRequest, expectedRequestResponses); servicePort.receive(correspondingResponse); serviceEventsTest(expectedRequestResponses) } [] servCoordPort.receive("end test") { if(match(extractRequests(expectedRequestResponses), receivedRequests)) { log("the expected service requests set does match the actual received requests"); setverdict(pass); } else { log("the expected service requests set does NOT match the actual received requests"); setverdict(fail); }; stop } } } 29

Testcase re-usability
testcase SOABasedWebTesting() runs on MTCType system SystemComponentType { var SOAComponentType theSOAComponent; var UserComponentType theUserComponent[2]; theUserComponent[0] := UserComponentType.create; theUserComponent[1] := UserComponentType.create; theSOAComponent := SOAComponentType.create; // map all ports here theSOAComponent.start(serviceEventsTest(theExpectedServiceRequests)); theUserComponent[0].start(User_1_behavior()); theUserComponent[1].start(User_2_behavior()); theUserComponent[0].done; theUserComponent[1].done; servCoordPort.send("end test"); all component.done; log("testcase SOABasedWebTesting completed"); } 30

15

Advantages of the generic service handler

The service handler does not need to be rewritten for each test campaign. Expected tuples of service requests/responses can be implemented with:
Templates for the sets of request/response tuples. Parametric functions that take the expected Request/Response tuples sets.

All of this thanks to the powerful TTCN-3 matching mechanism.

31

Using TTCN-3 to verify log files

Perform the black box testing of the web application from a users point of view. Use the services log files as part of the SUT:
Decode the log files Create a TTCN-3 function containing a behavior to verify log files automatically

This approach moves the status of log files from post-mortem analysis to a fully active status Advantages: it does not disturb the normal operation of the SUT.
32

16

Use of log files in testing

TTCN-3 test suite

Web Application Test component

Service Log file Test component

WEB server

service

Log file
33

Conclusions
Service Oriented Architecture presents challenges and opportunities to accurately pinpoint precise location of faults and quality issues TTCN-3 matching mechanism and control enables:
Scalable multi-user matching of request/responses Precise detection and location of faults and quality of service issues Reusable service-based test sets that can be leveraged across disparate web applications In-process or log-based fault analysis
34

17

Contact information
E-mail:
[email protected] [email protected] [email protected]

Further reading:
http://www.site.uottawa.ca/~bernard/ttcn.html http://www.site.uottawa.ca/~lpeyton/

35

18