Today it's the first time I'm using GWT and JDO. I am running it with Eclipse in the local debug mode.
I do the following thing:
public Collection<MyObject> add(MyObject o) {
PersistenceManager pm = PMF.get().getPersistenceManager();
try {
pm.makePersistent(o);
Query query = pm.newQuery(MyObject.class);// fetch all objects incl. o. But o only sometimes comes...
List<MyObject> rs = (List<MyObject>) query.execute();
ArrayList<MyObject> list= new ArrayList<MyObject>();
for (MyObject r : rs) {
list.add(r);
}
return list;
} finally {
pm.close();
}
}
I already set <property name="datanucleus.appengine.datastoreReadConsistency" value="STRONG" /> in my jdoconfig.xml. Do I have to set some other transaction stuff in the config? Was somebody got a working jdoconfig.xml? Or is the problem somewhere else? Some caching inbetween?
EDIT: Things I have tried:
Setting NontransactionalRead/Write to false
Using the same/a different PersistenceManager though calling PMF.get().getPersistenceManager() multiple times
Using transactions
ignoreCache = true on PersistenceManager
calling flush and checkConsistency
The jdoconfig:
<persistence-manager-factory name="transactions-optional">
<property name="datanucleus.appengine.datastoreReadConsistency" value="STRONG" />
<property name="javax.jdo.PersistenceManagerFactoryClass"
value="org.datanucleus.store.appengine.jdo.DatastoreJDOPersistenceManagerFactory"/>
<property name="javax.jdo.option.ConnectionURL" value="appengine"/>
<property name="javax.jdo.option.NontransactionalRead" value="true"/>
<property name="javax.jdo.option.NontransactionalWrite" value="true"/>
<property name="javax.jdo.option.RetainValues" value="true"/>
<property name="datanucleus.appengine.autoCreateDatastoreTxns" value="true"/>
</persistence-manager-factory>
I must be missing something central here because all approaches fail...
EDIT2: When I split the job into two transaction the log says that the write transaction fished and then the read transaction starts. But it doesn't find the just persited object. It always says Level 1 Cache of type "weak" initialised aswell. Is week bad or good?
It about 30% of requests that go wrong... Might I be some lazy query loading issue?
Franz, the Default read consistency in the JDO Config is STRONG. so if you are trying to approach it in that direction, it wont lead you anywhere
Check this out as i think it mentions something similar to the scenario which you are encountering, with the committed data not returned back in the query. It isnt concurrent as mentioned, but it explains the commit process.
http://code.google.com/appengine/articles/transaction_isolation.html
Also, another approach would be to query using Extents and find out if that solves the particular use case you are looking at, since i believe you are pulling out all the records in the table.
EDIT :
Since in the code snippet that you have mentioned, it queries the entire table. And if that is what you need, you can use an Extent...
The way to use it is by calling
Extent ext = getExtent(<Entity Class name>)
on the persistenceManager singleton object. You can then iterate through the Extent
Check out the documentation and search for Extents on the page here.
http://code.google.com/appengine/docs/java/datastore/jdo/queries.html
Calling the makePersistent() method doesn't write to the datastore; closing the PersistenceManager or committing your changes does. Since you haven't done this when you run your query, you're getting all objects from the datastore which does not, yet, include the object you just called makePersistent on.
Read about object states here:
http://db.apache.org/jdo/state_transition.html
There are two ways around this, you can put this inside a transaction since the commit writes to the datastore (keep in mind GAE 5 transaction/entity type limit on transactions) and commit before running your query;
Example using transaction...
public Collection<MyObject> add(MyObject o) {
PersistenceManager pm = PMF.get().getPersistenceManager();
ArrayList<MyObject> list = null;
try {
Transaction tx=pm.currentTransaction();
try {
tx.begin();
pm.makePersistent(o);
tx.commit();
} finally {
if (tx.isActive()) {
tx.rollback();
}
}
Query query = pm.newQuery(MyObject.class);
List<MyObject> rs = (List<MyObject>) query.execute();
ArrayList<MyObject> list = new ArrayList<MyObject>();
for (MyObject r : rs) {
list.add(r);
}
} finally {
pm.close();
}
return list;
}
or you could close the persistence manager after calling makePersistent on o and then open another one to run your query on.
// Note that this only works assuming the makePersistent call is successful
public Collection<MyObject> add(MyObject o) {
PersistenceManager pm = PMF.get().getPersistenceManager();
try {
pm.makePersistent(o);
} finally {
pm.close();
}
pm = PMF.get().getPersistenceManager();
ArrayList<MyObject> list = null;
try {
Query query = pm.newQuery(MyObject.class);
List<MyObject> rs = (List<MyObject>) query.execute();
list= new ArrayList<MyObject>();
for (MyObject r : rs) {
list.add(r);
}
} finally {
pm.close();
}
return list;
}
NOTE: I originally said you could just add o to the result list before returning; but that isn't a smart thing to do since in the event that there is a problem writing o to the datastore; then the returned list wouldn't reflect the actual data in the datastore. Doing what I now have (committing a transaction or closing the pm and then getting another one) should work since you have your datastoreReadPolicy set to STRONG.
I encountered the same problem and this didn't help. Since it seems to be the top result on Google for "jdo app engine consistency in eclipse" I figured I would share the fix for me!
Turns out I was using multiple instances the PersistenceManagerFactory which led to some bizarre behaviour. The fix is to have a singleton that every piece of code accesses. This is in fact documented correctly on the GAE tutorials but I think it's importance is understated.
Getting a PersistenceManager Instance
An app interacts with JDO using an instance of the PersistenceManager
class. You get this instance by instantiating and calling a method on
an instance of the PersistenceManagerFactory class. The factory uses
the JDO configuration to create PersistenceManager instances.
Because a PersistenceManagerFactory instance takes time to initialize,
an app should reuse a single instance. An easy way to manage the
PersistenceManagerFactory instance is to create a singleton wrapper
class with a static instance, as follows:
PMF.java
import javax.jdo.JDOHelper;
import javax.jdo.PersistenceManagerFactory;
public final class PMF {
private static final PersistenceManagerFactory pmfInstance =
JDOHelper.getPersistenceManagerFactory("transactions-optional");
private PMF() {}
public static PersistenceManagerFactory get() {
return pmfInstance;
}
}
Related
So, the question at hand is about initializing the lazy collections of an "unknown" entity, as long as these are known at least by name. This is part of a more wide effort of mine to build a generic DataTable -> RecordDetails miniframework in JSF + Primefaces.
So, the associations are usually lazy, and the only moment i need them loaded is when someone accesses one record of the many in the datatable in order to view/edit it. The issues here is that the controllers are generic, and for this I also use just one service class backing the whole LazyLoading for the datatable and loading/saving the record from the details section.
What I have with come so far is the following piece of code:
public <T> T loadWithDetails(T record, String... associationsToInitialize) {
final PersistenceUnitUtil pu = em.getEntityManagerFactory().getPersistenceUnitUtil();
record = (T) em.find(record.getClass(), pu.getIdentifier(record));
for (String association : associationsToInitialize) {
try {
if (!pu.isLoaded(record, association)) {
loadAssociation(record, association);
}
} catch (..... non significant) {
e.printStackTrace(); // Nothing else to do
}
}
return record;
}
private <T> void loadAssociation(T record, String associationName) throws IntrospectionException, InvocationTargetException, IllegalAccessException, NoSuchFieldException {
BeanInfo info = Introspector.getBeanInfo(record.getClass(), Object.class);
PropertyDescriptor[] props = info.getPropertyDescriptors();
for (PropertyDescriptor pd : props) {
if (pd.getName().equals(associationName)) {
Method getter = pd.getReadMethod();
((Collection) getter.invoke(record)).size();
}
}
throw new NoSuchFieldException(associationName);
}
And the question is, did anyone start any similar endeavor, or does anyone know of a more pleasant way to initialize collections in a JPA way (not Hibernate / Eclipselink specific) without involving reflection?
Another alternative I could think of is forcing all entities to implement some interface with
Object getId();
void loadAssociations();
but I don't like the idea of forcing my pojos to implement some interface just for this.
With the reflection solution you would suffer the N+1 effect detailed here: Solve Hibernate Lazy-Init issue with hibernate.enable_lazy_load_no_trans
You could use the OpenSessionInView instead, you will be affected by the N+1 but you will not need to use reflection. If you use this pattern your transaction will remain opened until the end of the transaction and all the LAZY relationships will be loaded without a problem.
For this pattern you will need to do a WebFilter that will open and close the transaction.
Let's say I specify an outputText component like this:
<h:outputText value="#{ManagedBean.someProperty}"/>
If I print a log message when the getter for someProperty is called and load the page, it is trivial to notice that the getter is being called more than once per request (twice or three times is what happened in my case):
DEBUG 2010-01-18 23:31:40,104 (ManagedBean.java:13) - Getting some property
DEBUG 2010-01-18 23:31:40,104 (ManagedBean.java:13) - Getting some property
If the value of someProperty is expensive to calculate, this can potentially be a problem.
I googled a bit and figured this is a known issue. One workaround was to include a check and see if it had already been calculated:
private String someProperty;
public String getSomeProperty() {
if (this.someProperty == null) {
this.someProperty = this.calculatePropertyValue();
}
return this.someProperty;
}
The main problem with this is that you get loads of boilerplate code, not to mention private variables that you might not need.
What are the alternatives to this approach? Is there a way to achieve this without so much unnecessary code? Is there a way to stop JSF from behaving in this way?
Thanks for your input!
This is caused by the nature of deferred expressions #{} (note that "legacy" standard expressions ${} behave exactly the same when Facelets is used instead of JSP). The deferred expression is not immediately evaluated, but created as a ValueExpression object and the getter method behind the expression is executed everytime when the code calls ValueExpression#getValue().
This will normally be invoked one or two times per JSF request-response cycle, depending on whether the component is an input or output component (learn it here). However, this count can get up (much) higher when used in iterating JSF components (such as <h:dataTable> and <ui:repeat>), or here and there in a boolean expression like the rendered attribute. JSF (specifically, EL) won't cache the evaluated result of the EL expression at all as it may return different values on each call (for example, when it's dependent on the currently iterated datatable row).
Evaluating an EL expression and invoking a getter method is a very cheap operation, so you should generally not worry about this at all. However, the story changes when you're performing expensive DB/business logic in the getter method for some reason. This would be re-executed everytime!
Getter methods in JSF backing beans should be designed that way that they solely return the already-prepared property and nothing more, exactly as per the Javabeans specification. They should not do any expensive DB/business logic at all. For that the bean's #PostConstruct and/or (action)listener methods should be used. They are executed only once at some point of request-based JSF lifecycle and that's exactly what you want.
Here is a summary of all different right ways to preset/load a property.
public class Bean {
private SomeObject someProperty;
#PostConstruct
public void init() {
// In #PostConstruct (will be invoked immediately after construction and dependency/property injection).
someProperty = loadSomeProperty();
}
public void onload() {
// Or in GET action method (e.g. <f:viewAction action>).
someProperty = loadSomeProperty();
}
public void preRender(ComponentSystemEvent event) {
// Or in some SystemEvent method (e.g. <f:event type="preRenderView">).
someProperty = loadSomeProperty();
}
public void change(ValueChangeEvent event) {
// Or in some FacesEvent method (e.g. <h:inputXxx valueChangeListener>).
someProperty = loadSomeProperty();
}
public void ajaxListener(AjaxBehaviorEvent event) {
// Or in some BehaviorEvent method (e.g. <f:ajax listener>).
someProperty = loadSomeProperty();
}
public void actionListener(ActionEvent event) {
// Or in some ActionEvent method (e.g. <h:commandXxx actionListener>).
someProperty = loadSomeProperty();
}
public String submit() {
// Or in POST action method (e.g. <h:commandXxx action>).
someProperty = loadSomeProperty();
return "outcome";
}
public SomeObject getSomeProperty() {
// Just keep getter untouched. It isn't intented to do business logic!
return someProperty;
}
}
Note that you should not use bean's constructor or initialization block for the job because it may be invoked multiple times if you're using a bean management framework which uses proxies, such as CDI.
If there are for you really no other ways, due to some restrictive design requirements, then you should introduce lazy loading inside the getter method. I.e. if the property is null, then load and assign it to the property, else return it.
public SomeObject getSomeProperty() {
// If there are really no other ways, introduce lazy loading.
if (someProperty == null) {
someProperty = loadSomeProperty();
}
return someProperty;
}
This way the expensive DB/business logic won't unnecessarily be executed on every single getter call.
See also:
Why is the getter called so many times by the rendered attribute?
Invoke JSF managed bean action on page load
How and when should I load the model from database for h:dataTable
How to populate options of h:selectOneMenu from database?
Display dynamic image from database with p:graphicImage and StreamedContent
Defining and reusing an EL variable in JSF page
Measure the render time of a JSF view after a server request
With JSF 2.0 you can attach a listener to a system event
<h:outputText value="#{ManagedBean.someProperty}">
<f:event type="preRenderView" listener="#{ManagedBean.loadSomeProperty}" />
</h:outputText>
Alternatively you can enclose the JSF page in an f:view tag
<f:view>
<f:event type="preRenderView" listener="#{ManagedBean.loadSomeProperty}" />
.. jsf page here...
<f:view>
I have written an article about how to cache JSF beans getter with Spring AOP.
I create a simple MethodInterceptor which intercepts all methods annotated with a special annotation:
public class CacheAdvice implements MethodInterceptor {
private static Logger logger = LoggerFactory.getLogger(CacheAdvice.class);
#Autowired
private CacheService cacheService;
#Override
public Object invoke(MethodInvocation methodInvocation) throws Throwable {
String key = methodInvocation.getThis() + methodInvocation.getMethod().getName();
String thread = Thread.currentThread().getName();
Object cachedValue = cacheService.getData(thread , key);
if (cachedValue == null){
cachedValue = methodInvocation.proceed();
cacheService.cacheData(thread , key , cachedValue);
logger.debug("Cache miss " + thread + " " + key);
}
else{
logger.debug("Cached hit " + thread + " " + key);
}
return cachedValue;
}
public CacheService getCacheService() {
return cacheService;
}
public void setCacheService(CacheService cacheService) {
this.cacheService = cacheService;
}
}
This interceptor is used in a spring configuration file:
<bean id="advisor" class="org.springframework.aop.support.DefaultPointcutAdvisor">
<property name="pointcut">
<bean class="org.springframework.aop.support.annotation.AnnotationMatchingPointcut">
<constructor-arg index="0" name="classAnnotationType" type="java.lang.Class">
<null/>
</constructor-arg>
<constructor-arg index="1" value="com._4dconcept.docAdvance.jsfCache.annotation.Cacheable" name="methodAnnotationType" type="java.lang.Class"/>
</bean>
</property>
<property name="advice">
<bean class="com._4dconcept.docAdvance.jsfCache.CacheAdvice"/>
</property>
</bean>
Hope it will help!
Originally posted in PrimeFaces forum # http://forum.primefaces.org/viewtopic.php?f=3&t=29546
Recently, I have been obsessed evaluating the performance of my app, tuning JPA queries, replacing dynamic SQL queries with named queries, and just this morning, I recognized that a getter method was more of a HOT SPOT in Java Visual VM than the rest of my code (or majority of my code).
Getter method:
PageNavigationController.getGmapsAutoComplete()
Referenced by ui:include in in index.xhtml
Below, you will see that PageNavigationController.getGmapsAutoComplete() is a HOT SPOT (performance issue) in Java Visual VM. If you look further down, on the screen capture, you will see that getLazyModel(), PrimeFaces lazy datatable getter method, is a hot spot too, only when enduser is doing a lot of 'lazy datatable' type of stuff/operations/tasks in the app. :)
See (original) code below.
public Boolean getGmapsAutoComplete() {
switch (page) {
case "/orders/pf_Add.xhtml":
case "/orders/pf_Edit.xhtml":
case "/orders/pf_EditDriverVehicles.xhtml":
gmapsAutoComplete = true;
break;
default:
gmapsAutoComplete = false;
break;
}
return gmapsAutoComplete;
}
Referenced by the following in index.xhtml:
<h:head>
<ui:include src="#{pageNavigationController.gmapsAutoComplete ? '/head_gmapsAutoComplete.xhtml' : (pageNavigationController.gmaps ? '/head_gmaps.xhtml' : '/head_default.xhtml')}"/>
</h:head>
Solution: since this is a 'getter' method, move code and assign value to gmapsAutoComplete prior to method being called; see code below.
/*
* 2013-04-06 moved switch {...} to updateGmapsAutoComplete()
* because performance = 115ms (hot spot) while
* navigating through web app
*/
public Boolean getGmapsAutoComplete() {
return gmapsAutoComplete;
}
/*
* ALWAYS call this method after "page = ..."
*/
private void updateGmapsAutoComplete() {
switch (page) {
case "/orders/pf_Add.xhtml":
case "/orders/pf_Edit.xhtml":
case "/orders/pf_EditDriverVehicles.xhtml":
gmapsAutoComplete = true;
break;
default:
gmapsAutoComplete = false;
break;
}
}
Test results: PageNavigationController.getGmapsAutoComplete() is no longer a HOT SPOT in Java Visual VM (doesn't even show up anymore)
Sharing this topic, since many of the expert users have advised junior JSF developers to NOT add code in 'getter' methods. :)
If you are using CDI, you can use Producers methods.
It will be called many times, but the result of first call is cached in scope of the bean and is efficient for getters that are computing or initializing heavy objects!
See here, for more info.
You could probably use AOP to create some sort of Aspect that cached the results of our getters for a configurable amount of time. This would prevent you from needing to copy-and-paste boilerplate code in dozens of accessors.
If the value of someProperty is
expensive to calculate, this can
potentially be a problem.
This is what we call a premature optimization. In the rare case that a profiler tells you that the calculation of a property is so extraordinarily expensive that calling it three times rather than once has a significant performance impact, you add caching as you describe. But unless you do something really stupid like factoring primes or accessing a databse in a getter, your code most likely has a dozen worse inefficiencies in places you've never thought about.
I would also advice using such Framework as Primefaces instead of stock JSF, they address such issues before JSF team e. g in primefaces you can set partial submit. Otherwise BalusC has explained it well.
It still big problem in JSF. Fo example if you have a method isPermittedToBlaBla for security checks and in your view you have rendered="#{bean.isPermittedToBlaBla} then the method will be called multiple times.
The security check could be complicated e.g . LDAP query etc. So you must avoid that with
Boolean isAllowed = null ... if(isAllowed==null){...} return isAllowed?
and you must ensure within a session bean this per request.
Ich think JSF must implement here some extensions to avoid multiple calls (e.g annotation #Phase(RENDER_RESPONSE) calle this method only once after RENDER_RESPONSE phase...)
In the following case where two DbContexts are nested due to method calls:
public void Method_A() {
using (var db = new SomeDbContext()) {
//...do some work here
Method_B();
//...do some more work here
}
}
public void Method_B() {
using (var db = new SomeDbContext()) {
//...do some work
}
}
Question:
Will this nesting cause any issues? (and will the correct DbContext be disposed at the correct time?)
Is this nesting considered bad practice, should Method_A be refactored into:
public void Method_A() {
using (var db = new SomeDbContext()) {
//...do some work here
}
Method_B();
using (var db = new SomeDbContext()) {
//...do some more work here
}
}
Thanks.
Your DbContext derived class is actually managing at least three things for you here:
the metadata that describes your database and your entity model,
the underlying database connection, and
a client side "cache" of entities loaded using the context, for change tracking, relationship fixup, etc. (Note that although I term this a "cache" for want of a better word, this is generally short lived and is just to support EFs functionality. It's not a substitute for proper caching in your application if applicable.)
Entity Framework generally caches the metadata (item 1) so that it is shared by all context instances (or, at least, all instances that use the same connection string). So here that gives you no cause for concern.
As mentioned in other comments, your code results in using two database connections. This may or may not be a problem for you.
You also end up with two client caches (item 3). If you happen to load an entity from the outer context, then again from the inner context, you will have two copies of it in memory. This would definitely be confusing, and could lead to subtle bugs. This means that, if you don't want to use shared context objects, then your option 2 would probably be better than option 1.
If you are using transactions, there are further considerations. Having multiple database connections is likely to result in transactions being promoted to distributed transactions, which is probably not what you want. Since you didn't make any mention of db transactions, I won't go into this further here.
So, where does this leave you?
If you are using this pattern simply to avoid passing DbContext objects around in your code, then you would probably be better off refactoring MethodB to receive the context as a parameter. The question of how long-lived context objects should be comes up repeatedly. As a rule of thumb, create a new context for a single database operation or for a series of related database operations. (See, for example this blog post and this question.)
(As an alternative, you could add a constructor to your DbContext derived class that receives an existing connection. Then you could share the same connection between multiple contexts.)
One useful pattern is to write your own class that creates a context object and stores it as a private field or property. Then you make your class implement IDisposable and its Dispose() method disposes the context object. Your calling code news up an instance of your class, and doesn't have to worry about contexts or connections at all.
When might you need to have multiple contexts active at the same time?
This can be useful when you need to write code that is multi-threaded. A database connection is not thread-safe, so you must only ever access a connection (and therefore an EF context) from one thread at a time. If that is too restrictive, you need multiple connections (and contexts), one per thread. You might find this interesting.
You can alter your code by passing to Method_B the context. If you do so, the creation of the second db call SomeDbContext will not be necessary.
there a question an answer in stackoverflow in this link
Proper use of "Using" statement for datacontext
It is a bit late answer, but still people may be looking so here is another way.
Create class, that cares about disposing for you. In some scenarios, there would be a function usable from different places in solution. This way you avoid creating multiple instances of DbContext and you can use nested calls as many as you like.
Pasting simple example.
public class SomeContext : SomeDbContext
{
protected int UsingCount = 0;
public static SomeContext GetContext(SomeContext context)
{
if (context != null)
{
context.UsingCount++;
}
else
{
context = new SomeContext();
}
return context;
}
private SomeContext()
{
}
protected bool MyDisposing = true;
protected override void Dispose(bool disposing)
{
if (UsingCount == 0)
{
base.Dispose(MyDisposing);
MyDisposing = false;
}
else
{
UsingCount--;
}
}
public override int SaveChanges()
{
if (UsingCount == 0)
{
return base.SaveChanges();
}
else
{
return 0;
}
}
}
Example of usage
public class ExmapleNesting
{
public void MethodA()
{
using (var context = SomeContext.GetContext(null))
{
// manipulate, save it, just do not call Dispose on context in using
MethodB(context);
}
MethodB();
}
public void MethodB(SomeContext someContext = null)
{
using (var context = SomeContext.GetContext(someContext))
{
// manipulate, save it, just do not call Dispose on context in using
// Even more nested functions if you'd like
}
}
}
Simple and easy to use.
If you think number of connections to Database,and impact of times that new connections must be opened, not an important problem and you have no limitation for support your application to run at best performance, everything is OK.
Your code works well. Because create just a db context has a low impact in your performance,meta data will be cached after first loading, and connection to your database just occurs when the code need to execute a query. With liitle performance consideration and code design, I offer you to make context factory to have just an instance of each Db Context for each instance of your application.
You can take a look at this link for more performance considerations
http://msdn.microsoft.com/en-us/data/hh949853
This is more a solution / work around than an actual question. I'm posting it here since I couldn't find this solution on stack overflow or indeed after a lot of Googling.
The Problem:
I have an MVC 3 webapp using EF 4 code first that I want to write unit tests for. I'm also using NCrunch to run the unit tests on the fly as I code, so I'd like to avoid backing onto an actual database here.
Other Solutions:
IDataContext
I've found this the most accepted way to create an in memory datacontext. It effectively involves writing an interface IMyDataContext for your MyDataContext and then using the interface in all your controllers. An example of doing this is here.
This is the route I went with initially and I even went as far as writing a T4 template to extract IMyDataContext from MyDataContext since I don't like having to maintain duplicate dependent code.
However I quickly discovered that some Linq statements fail in production when using IMyDataContext instead of MyDataContext. Specifically queries like this throw a NotSupportedException
var siteList = from iSite in MyDataContext.Sites
let iMaxPageImpression = (from iPage in MyDataContext.Pages where iSite.SiteId == iPage.SiteId select iPage.AvgMonthlyImpressions).Max()
select new { Site = iSite, MaxImpressions = iMaxPageImpression };
My Solution
This was actually quite simple. I simply created a MyInMemoryDataContext subclass to MyDataContext and overrode all the IDbSet<..> properties as below:
public class InMemoryDataContext : MyDataContext, IObjectContextAdapter
{
/// <summary>Whether SaveChanges() was called on the DataContext</summary>
public bool SaveChangesWasCalled { get; private set; }
public InMemoryDataContext()
{
InitializeDataContextProperties();
SaveChangesWasCalled = false;
}
/// <summary>
/// Initialize all MyDataContext properties with appropriate container types
/// </summary>
private void InitializeDataContextProperties()
{
Type myType = GetType().BaseType; // We have to do this since private Property.Set methods are not accessible through GetType()
// ** Initialize all IDbSet<T> properties with CollectionDbSet<T> instances
var DbSets = myType.GetProperties().Where(x => x.PropertyType.IsGenericType && x.PropertyType.GetGenericTypeDefinition() == typeof(IDbSet<>)).ToList();
foreach (var iDbSetProperty in DbSets)
{
var concreteCollectionType = typeof(CollectionDbSet<>).MakeGenericType(iDbSetProperty.PropertyType.GetGenericArguments());
var collectionInstance = Activator.CreateInstance(concreteCollectionType);
iDbSetProperty.SetValue(this, collectionInstance,null);
}
}
ObjectContext IObjectContextAdapter.ObjectContext
{
get { return null; }
}
public override int SaveChanges()
{
SaveChangesWasCalled = true;
return -1;
}
}
In this case my CollectionDbSet<> is a slightly modified version of FakeDbSet<> here (which simply implements IDbSet with an underlying ObservableCollection and ObservableCollection.AsQueryable()).
This solution works nicely with all my unit tests and specifically with NCrunch running these tests on the fly.
Full Integration Tests
These Unit tests test all the business logic but one major downside is that none of your LINQ statements are guaranteed to work with your actual MyDataContext. This is because testing against an in memory data context means you're replacing the Linq-To-Entity provider but a Linq-To-Objects provider (as pointed out very well in the answer to this SO question).
To fix this I use Ninject within my unit tests and setup InMemoryDataContext to bind instead of MyDataContext within my unit tests. You can then use Ninject to bind to an actual MyDataContext when running the integration tests (via a setting in the app.config).
if(Global.RunIntegrationTest)
DependencyInjector.Bind<MyDataContext>().To<MyDataContext>().InSingletonScope();
else
DependencyInjector.Bind<MyDataContext>().To<InMemoryDataContext>().InSingletonScope();
Let me know if you have any feedback on this however, there are always improvements to be made.
As per my comment in the question, this was more to help others searching for this problem on SO. But as pointed out in the comments underneath the question there are quite a few other design approaches that would fix this problem.
I'm new to Moq, and just started on a project that's already in development. I'm responsible for setting up unit testing. There's a custom class for the DatabaseFactory that uses EnterpriseLibrary and looks like this:
public Database CreateCommonDatabase()
{
return CreateDatabaseInstance(string.Empty);
}
private static Database CreateDatabaseInstance(string foo)
{
var database = clientCode == string.Empty
? DatabaseFactory.CreateDatabase("COMMON")
: new OracleDatabase(new ClientConnections().GetConnectionString(foo)));
return database;
}
Now, here's where that gets used (ResultData is another class of the type DataSet):
public ResultData GetNotifications(string foo, string foo2, Database database)
{
var errMsg = string.Empty;
var retval = 0;
var ds = new DataSet();
var sqlClause =
#"[Some SELECT statement here that uses foo]";
DbCommand cm = database.GetSqlStringCommand(sqlClause);
cm.CommandType = CommandType.Text;
// Add Parameters
if (userSeq != string.Empty)
{
database.AddInParameter(cm, ":foo2", DbType.String, foo2);
}
try
{
ds = database.ExecuteDataSet(cm);
}
catch (Exception ex)
{
retval = -99;
errMsg = ex.Message;
}
return new ResultData(ds, retval, errMsg);
}
Now, originally, the Database wasn't passed in as a parameter, but the method was creating a new instance of the DatabaseFactory using the CreateCommonDatabase method, and using it from there. However, that leaves the class untestable because I can't keep it from actually hitting the database. So, I went with Dependency Injection, and pass the Database in.
Now, I'm stuck, because there's no way to mock Database in order to test GetNotifications. I'm wondering if I'm overly complicating things, or if I'm missing something. Am I doing this the right way, or should I be rethinking how I've got this set up?
Edit to add more info*****
I really don't want to test the database. I want the Data.Notifications class (above) to return an instance of ResultData, but that's all I really want to test. If I go a level up, to the Business layer, I have this:
public DataSet GetNotifications(string foo, string foo1, out int returnValue, out string errorMessage, Database database)
{
ResultData rd = new data.Notifications().GetNotifications(foo, foo1, database);
returnValue = rd.ResultValue;
errorMessage = rd.ErrorMessage;
return rd.DataReturned;
}
So, originally, the database wasn't passed in, it was the Data.Notifications class that created it - but then again, if I left it that way, I couldn't help but hit the database to test this Business layer object. I modified all of the code to pass the Database in (which gets created a the web's Base page), but now I'm just not certain what to do next. I thought I was one unit test away from having this resolved, but apparently, either I'm wrong or I've got a mental roadblock to the right path.
You should be able to create a mock Database object if the methods in it are virtual. If they are not, then you have a little bit of a problem.
I don't know what type "Database" is, but you have a few options.
If you own the source code to Database, I would recommend extracting an interface IDatabase, rather than dealing with a Database class type. This will eliminate some complexity and give you something extremely testable.
If you don't have access to the Database class, you can always solve this with another layer of abstraction. Many people in this case use a Repository pattern that wraps the data access layer. Generally speaking in this case, most people leave testing Respository classes to integration tests (tests without any isolation), rather than unit tests.
Here's how you'd setup your test using option #1:
[TestMethod]
public void GetNotifications_PassedNullFoo_ReturnsData()
{
//Arrange
Mock<IDatabase> mockDB = new Mock<IDatabase>();
mockDB.Setup(db => db.ExecuteDataSet()).Returns(new DataSet() ... );
//Act
FooClass target = new fooClass();
var result = target.GetNotifications(null, "Foo2", mockDB.Object);
//Assert
Assert.IsTrue(result.DataSet.Rows.Count > 0);
}
My dataset code is a little rusty, but hopefully this gives you the general idea.
Based on the code you've given, I would think you would want to talk to the database, and not a mocked version.
The reason is that your GetNotifications code contains DB-specific instructions, and you'll want those to pass validation at the DB Engine level. So just pass in a Database that is connected to your test DB instance.
If you took the testing abstraction to a higher level, where you built unit tests for the database call and a version of this test that used a mocked database, you'd still have to run integration tests, which ends up being triple the work for the same amount of code coverage. In my opinion, it's far more efficient to do an integration test at tier borders you control then to write unit tests for both sides of the contract and integration tests.