数据库测试的基础要素

{"type":"doc","content":[{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"众所周知，在测试行业，模拟数据库和其他持久化层会降低测试效率。在测试时，如果一个组件不属于测试的一部分，就很难测试它与其他组件之间的交互行为。遗憾的是，这个行业只专注于功能层面的测试，很少有人接受过其他类型测试的培训。这篇文章通过引入数据库测试的概念来纠正这个问题。这些技术也适用于其他类型的持久化机制，比如调用微服务。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了了解如何测试数据库，我们先“忘记”与单元测试和集成测试相关的一些概念。直接一点说，现如今对这些术语的定义已经偏离了它们最初的含义。所以，在文章的剩余部分，我们将不再使用它们。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"测试最本质的目的是生成信息。一个测试用例在执行完之后应该生成与被测试的东西相关的信息，这些信息是你原先不知道的。生成的信息越多越好。因此，我们倾向于“一个测试用例应该尽可能提供可以证明某个事实所需的断言”，而不是“一个测试用例只提供一个断言”。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另一个有问题的观点是“所有的测试都应该是独立的”。人们通常会误读这个观点，认为每一个测试都应该使用Mock，你所测试的每一个功能应该与它们的依赖项隔离。但这样是毫无意义的，因为在生产环境中，这些功能不可能与它们的依赖项隔离。相反，你应该尽可能像在生产环境中那样测试，这样才会发现尽可能多的问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“所有的测试都应该是独立的”这句话真正的意思是说，每一个测试都可以独立于其他测试运行。或者，换句话说，你可以按照任意的顺序、在任意时刻运行每一个测试或一组测试。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"很多人在测试时把事情弄复杂了。他们在执行每一个测试（甚至是每一个单独的测试用例）之前都会完整地重建数据库。这带来了一些问题。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"首先，测试变慢了。创建新数据库和填充数据需要时间，这通常是造成数据库测试变慢的直接原因，而这又反过来让人们不愿意去执行测试，甚至不准备这类测试。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"另一个问题与数据库里的记录数量有关。当数据库里只有一条代码，有些代码运行得很好，但当有成千上百条记录时就会失败。在某些情况下，比如查询语句里缺少了WHERE子句，只要两条记录就会导致测试失败。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"因此，我们需要编写数据库端的测试。不管在任何时候，你都应该用生产环境的数据副本来执行测试，并看着它们全部执行成功。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"“"},{"type":"link","attrs":{"href":"https:\/\/grauenwolf.github.io\/DotNet-ORM-Cookbook\/index.htm","title":"","type":null},"content":[{"type":"text","text":".NET ORM Cookbook"}]},{"type":"text","text":"”就给出了一个很好的示例。这个项目有1600多个数据库端测试用例，它们可以按照任意顺序执行。为了理解其中的原理，我们将构建一些简单的CRUD测试来解释这些概念。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来的问题是一致性。人们常说，每一个测试都应该具备完美的一致性，也就是说，每次运行同一个测试都应该得到相同的结果。为了获得一致性，不能使用基于时间或随机生成的测试数据，也不能被环境影响到。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在测试数据库时，这是无法实现的。因为总有一些不可预测的问题出现，比如网络连接问题、磁盘问题、旧数据，等等。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"但并不是说不具备这种一致性的测试就是不可靠的。尽管一些属性会不一致，但测试在大部分时间都会返回相同的结果。随机出现的失败让可以你知道应用程序在哪些情况下会有怎样的表现。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"注意：本文所有的"},{"type":"link","attrs":{"href":"https:\/\/github.com\/Grauenwolf\/TestingWithDatabases\/tree\/main\/TestingWithDatabases","title":"","type":null},"content":[{"type":"text","text":"例子"}]},{"type":"text","text":"都可以再GitHub上找到。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"创建记录"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"我们的第一个测试是创建一条记录。为了简单起见，我们选择了EmployeeClassification类，它只有四个字段："}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"int EmployeeClassificationKey \nstring? EmployeeClassificationName \nbool IsEmployee \nbool IsExempt\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在检查数据库模式时，我们发现EmployeeClassificationKey是一个自生成数字字段，所以就不用管它了。EmployeeClassificationName有唯一性约束，这是给很多人造成麻烦的地方。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example1_Create()\n{\n var repo = CreateEmployeeClassificationRepository();\n var row = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test classification\",\n };\n await repo.CreateAsync(row);\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个测试是不可重复运行的，因为在第二次运行它时，相同的名字已经存在了。为了解决这个问题，我们加了一个区分方式，比如时间戳或GUID。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example2_Create()\n{\n var repo = CreateEmployeeClassificationRepository();\n var row = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n await repo.CreateAsync(row);\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"这个测试并没有真正测试什么东西。我们知道，CreateAsyn没有抛出异常，但它可能是一个空方法。为了让测试完整，我们需要加入读操作。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"创建和读取记录"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在创建和读取测试中，我们先确保可以从数据库读取到非0的键。然后，我们用这个键读取记录，并验证从数据库读取的记录字段与原先的一样。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example3_Create_And_Read()\n{\n var repo = CreateEmployeeClassificationRepository();\n var row = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n var key = await repo.CreateAsync(row);\n Assert.IsTrue(key != 0);\n\n var echo = await repo.GetByKeyAsync(key);\n Assert.AreEqual(key, echo.EmployeeClassificationKey);\n Assert.AreEqual(row.EmployeeClassificationName, echo.EmployeeClassificationName);\n Assert.AreEqual(row.IsEmployee, echo.IsEmployee);\n Assert.AreEqual(row.IsExempt, echo.IsExempt);\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"注意：当没有读取到记录时Repository并不会抛出异常，所以，在属性级别的断言之前加入Assert.IsNotNull，可以更好地捕获测试失败情况。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"断言太多会导致一些问题。首先，如果一个断言失败了，你不知道是哪一个。IsEmployee和IsExempt都是Boolean类型，所以你都没有办法通过上下文信息来判断是哪个失败了。你可以通过加入更多信息来解决这个问题，如果测试框架支持的话。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"其次，难以诊断。如果多个断言失败了，只有第一个被捕获到，后续的信息丢失了。为了解决这个问题，我们使用了AssertionScope对象。所有与之相关的断言会被集中在一起，在using代码块最后统一报出来。AssertionScope的实现示例可以在"},{"type":"link","attrs":{"href":"https:\/\/github.com\/Grauenwolf\/TestingWithDatabases\/blob\/main\/TestingWithDatabases\/AssertionScope.cs","title":"","type":null},"content":[{"type":"text","text":"GitHub"}]},{"type":"text","text":"上找到。对于更为复杂的创建，可以考虑使用"},{"type":"link","attrs":{"href":"https:\/\/fluentassertions.com\/introduction#assertion-scopes","title":"","type":null},"content":[{"type":"text","text":"流式AssertionScope"}]},{"type":"text","text":"或者NUnit的"},{"type":"link","attrs":{"href":"https:\/\/docs.nunit.org\/articles\/nunit\/writing-tests\/assertions\/multiple-asserts.html","title":"","type":null},"content":[{"type":"text","text":"Assert.Multiple"}]},{"type":"text","text":"。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example4_Create_And_Read()\n{\n var repo = CreateEmployeeClassificationRepository();\n var row = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n var key = await repo.CreateAsync(row);\n Assert.IsTrue(key != 0, \"New key wasn't created or returned\");\n\n var echo = await repo.GetByKeyAsync(key);\n\n using (var scope = new AssertionScope(stepName))\n {\n scope.AreEqual(expected.EmployeeClassificationKey, actual.EmployeeClassificationKey, \"EmployeeClassificationKey\");\n scope.AreEqual(expected.EmployeeClassificationName, actual.EmployeeClassificationName, \"EmployeeClassificationName\");\n scope.AreEqual(expected.IsEmployee, actual.IsEmployee, \"IsEmployee\");\n scope.AreEqual(expected.IsExempt, actual.IsExempt, \"IsExempt\");\n } \n} \n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"随着测试用例越来越多，这会变成一项枯燥的重复性工作，所以我们需要一个辅助方法。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"row.EmployeeClassificationKey = key;\nPropertiesAreEqual(row, echo); \n\nstatic void PropertiesAreEqual(EmployeeClassification expected, EmployeeClassification actual, string? stepName = null)\n{\n Assert.IsNotNull(actual, $\"Actual value for step {stepName} is null.\");\n Assert.IsNotNull(expected, $\"Expected value for step {stepName} is null.\");\n\n using (var scope = new AssertionScope(stepName))\n {\n scope.AreEqual(expected.EmployeeClassificationKey, actual.EmployeeClassificationKey, \"EmployeeClassificationKey\");\n scope.AreEqual(expected.EmployeeClassificationName, actual.EmployeeClassificationName, \"EmployeeClassificationName\");\n scope.AreEqual(expected.IsEmployee, actual.IsEmployee, \"IsEmployee\");\n scope.AreEqual(expected.IsExempt, actual.IsExempt, \"IsExempt\");\n }\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"你也可以不用手动写这个方法，直接使用"},{"type":"link","attrs":{"href":"https:\/\/github.com\/GregFinzer\/Compare-Net-Objects","title":"","type":null},"content":[{"type":"text","text":"CompareNETObjects"}]},{"type":"text","text":"库。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"创建、更新和读取记录"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"接下来的测试我们要更新记录，涉及一个创建操作和两次读取操作。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example5_Create_And_Update()\n{\n var repo = CreateEmployeeClassificationRepository();\n var version1 = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n var key = await repo.CreateAsync(version1);\n Assert.IsTrue(key != 0, \"New key wasn't created or returned\");\n version1.EmployeeClassificationKey = key;\n\n var version2 = await repo.GetByKeyAsync(key);\n PropertiesAreEqual(version1, version2, \"After created\");\n\n version2.EmployeeClassificationName = \"Modified \" + DateTime.Now.Ticks;\n await repo.UpdateAsync(version2);\n\n var version3 = await repo.GetByKeyAsync(key);\n PropertiesAreEqual(version2, version3, \"After update\");\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"为了能够知道为什么比较操作会失败，我们给PropertiesAreEqual方法加了一个stepName参数。"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"创建和删除记录"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"到目前为止，我们已经涵盖了CRUD的C、R和U，就差D了。在删除测试中，我们仍然会读取数据两次。但是，我们会使用Repository另一个方法，当找不动记录时返回null。如果你的Repository没有这个方法，请参考第7个示例。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\npublic async Task Example6_Create_And_Delete()\n{\n var repo = CreateEmployeeClassificationRepository();\n var version1 = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n var key = await repo.CreateAsync(version1);\n Assert.IsTrue(key != 0, \"New key wasn't created or returned\");\n version1.EmployeeClassificationKey = key;\n\n var version2 = await repo.GetByKeyOrNullAsync(key);\n Assert.IsNotNull(version2, \"Record wasn't created\");\n PropertiesAreEqual(version1, version2, \"After created\");\n\n await repo.DeleteByKeyAsync(key);\n\n var version3 = await repo.GetByKeyOrNullAsync(key);\n Assert.IsNull(version3, \"Record wasn't deleted\");\n}\n[TestMethod]\npublic async Task Example7_Create_And_Delete()\n{\n var repo = CreateEmployeeClassificationRepository();\n var version1 = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n };\n var key = await repo.CreateAsync(version1);\n Assert.IsTrue(key != 0, \"New key wasn't created or returned\");\n version1.EmployeeClassificationKey = key;\n\n var version2 = await repo.GetByKeyAsync(key);\n PropertiesAreEqual(version1, version2, \"After created\");\n\n await repo.DeleteByKeyAsync(key);\n\n try\n {\n await repo.GetByKeyAsync(key);\n Assert.Fail(\"Expected an exception. Record wasn't deleted\");\n }\n catch (MissingDataException)\n {\n \/\/Expected\n }\n}\n"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"如果你的数据库使用了软删除，你还需要检查相应的记录是否更新了删除标记。这可以通过以下几行代码来实现。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"var version4 = await GetEmployeeClassificationIgnoringDeletedFlag(key);\nAssert.IsNotNull(version4, \"Record was hard deleted\");\nAssert.IsTrue(version4.IsDeleted);\n"}]},{"type":"heading","attrs":{"align":null,"level":2},"content":[{"type":"text","text":"创建记录的改进"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null},"content":[{"type":"text","text":"在第一个测试中，可选数据列总是使用默认值。这个可以通过数据驱动测试来解决。下面的例子针对的是 MSTest，不过其他主流的测试框架也有类似的东西。"}]},{"type":"paragraph","attrs":{"indent":0,"number":0,"align":null,"origin":null}},{"type":"codeblock","attrs":{"lang":"java"},"content":[{"type":"text","text":"[TestMethod]\n[DataTestMethod, EmployeeClassificationSource]\npublic async Task Example9_Create_And_Read(bool isExempt, bool isEmployee)\n{\n var repo = CreateEmployeeClassificationRepository();\n var row = new EmployeeClassification()\n {\n EmployeeClassificationName = \"Test \" + DateTime.Now.Ticks,\n IsExempt = isExempt,\n IsEmployee = isEmployee\n };\n var key = await repo.CreateAsync(row);\n Assert.IsTrue(key > 0);\n Debug.WriteLine(\"EmployeeClassificationName: \" + key);\n\n var echo = await repo.GetByKeyAsync(key);\n Assert.AreEqual(key, echo.EmployeeClassificationKey);\n Assert.AreEqual(row.EmployeeClassificationName, echo.EmployeeClassificationName);\n Assert.AreEqual(row.IsEmployee, echo.IsEmployee);\n Assert.AreEqual(row.IsExempt, echo.IsExempt);\n}\n\npublic class EmployeeClassificationSourceAttribute : Attribute, ITestDataSource\n{\n public IEnumerable