10

We have a nightly integration that runs all tests and reports any errors against production, currently on Winter '19. Everything was fine until January 14th, 2019, when 13 previously passing tests started to fail with MIXED_DML_OPERATION exceptions. The exact same tests are passing in my Spring '19 sandbox. I created a new demonstration class in my Developer Org, which is also Winter '19, and the tests are passing there, too. Is there a documented change on MIXED_DML_OPERATION, or is a bug being fixed?


@isTest class MixedDMLError {
    @isTest static void test() {
        User u = [select Email, TimezoneSidKey, LocaleSidkey, LastName, EmailEncodingKey, ProfileId, LanguageLocaleKey from user where id = :UserInfo.getUserId()];
        u = u.clone(false);
        u.username = 'user'+math.random()+'@org.random';
        u.alias = (math.random()+'').remove('.').left(8);
        insert u;
        Account a = new Account(Name=Math.random()+'');
        insert a;
    }
}

Edit

It has been mentioned that UserRoleId might be the culprit, so I checked the source code. Here's the actual method creating the user:

public static User createUser( String profileName,String usrName, Boolean doInsert){
    Profile pro =[SELECT Id From Profile WHERE Name = :profileName];
    User usr = new user(
        FirstName='USR_FN',
        LastName='USR_LN',
        Username=usrName,
        ProfileId=pro.Id,
        email='example@example.com',
        emailencodingkey='UTF-8',
        languagelocalekey='en_US',
        localesidkey='en_US',
        timezonesidkey='America/Los_Angeles',
        alias='USR_FN'.substring(0,1)+'USR_LN'.substring(0,1)+string.valueof( Math.random()).substring(0,3)
    );

    if(doInsert)
        insert usr;

    return usr;
}

The role is left null, so unfortunately, that does not explain the difference. Both the unit test and the utility class it calls are identical in the Sandboxes I've tested and Production.


Edit 2

This problem only occurs when running tests. During deployments/validations using Run All Tests or Run Specified Tests, the operation succeeds without any errors. This appears to be a bug in how tests are run.

  • 2
    I had the same thing happen a few releases back - Mixed DML that was previously working just stopped. Salesforce wasn't helpful and we ended up having to rewrite the test class, which was something that needed done anyway – brezotom Jan 15 at 2:17
  • @brezotom Yeah, this seems off though--the documentation doesn't match the behavior, so something needs to be fixed either way. – sfdcfox Jan 15 at 3:55
  • Just wanted to say I am facing the same exact issue. – Bryan Anderson Jan 16 at 19:14
4

Likely not a bug. Note from the documentation on sObjects That Cannot Be Used Together in DML Operations:

You can insert a user in a transaction with other sObjects in Apex code saved using Salesforce API version 15.0 and later if UserRoleId is specified as null.

Your unit test does not filter on this field, nor clear it out before inserting the clone.

If you are going to create a User record by cloning the running user, make sure to set UserRoleId = null before your insert call.

If you do not do so, test runs will not behave the same for each user who runs them, or if certain attributes on that user change.

  • The real code does not use a query, actually; you'll want to see my edit. – sfdcfox Jan 14 at 22:27
  • Yeah makes more sense now. Do you have any triggers or workflows which might be setting UserRoleId inadvertently? Hopefully not in a managed package... – Adrian Larson Jan 14 at 22:29
  • But the documentation also says you cannot update profile id and other object in same transaction. UserRoleId no where mentioned in the code role can be blank in user so my thought is profile id is causing mixed DML – Avijit Chakraborty Jan 15 at 3:13
  • Your assessment seems off base. There is only an insert statement. The passage you cite above mentions update only. Seems quite likely irrelevant. – Adrian Larson Jan 15 at 3:17
  • @AdrianLarson Lightning Sync was causing the problem. We disabled it and the tests started passing. R&D should have a bug for it shortly. – sfdcfox Feb 7 at 20:02
2

Working with support, the root cause was uncovered. They're working with R&D, but I don't have a bug number at the time of this answer.

The root cause was that the System Administrator profile had been enabled for Lightning Sync. This causes erroneous test failures if you follow this advice:

You can insert a user in a transaction with other sObjects in Apex code saved using Salesforce API version 15.0 and later if UserRoleId is specified as null.

In our code, we have a number of tests that were using this technique, and they were all failing when ran directly, but not when deploying (change sets or API).

For now, we've disabled Lightning Sync for the System Administrator profile, and we have a backlog item to update these unit tests to use System.runAs in the future.

  • 1
    I remember speaking with the Salesforce support last year about pretty much the same issue and they pointed out that some test behavior characteristics are different when deploying (read: they do less actual testing) compared to running the tests directly in production. A pretty odd behavior but certainly worth noting. – Semmel Feb 7 at 20:08
1

I had a similar issue at a big client last year where some tests were suddenly failing in production after the API version was upgraded for this instance. The tests did not fail in any sandbox and they did not fail on deployment. After escalating the issue a few levels we finally got a savvy support user that tracked down the issue to the internally used Salesforce objects (the dreaded S2X objects). This, in turn, was actually caused by one of the apps only installed in production (I think it was the Salesforce sync for Exchange or something) that set up the objects behind the scenes in such a way that it triggered the MIXED_DML_OPERATION error. I remember that my colleague fixed the tests after I left the company, so I know it was fixable by rewriting our tests (which ran fine for years!), although I do not know what he did exactly as I was already gone at that point. Still - one of the most disturbing Salesforce bugs to date. I hope this information helps at least someone even if I cannot give you an exact solution to this problem.

  • 1
    +1 That's useful additional feedback. In my case, it was Lightning Sync on a SA profile, so there was at least a viable workaround for us. – sfdcfox Feb 7 at 23:01
0

You can not update account and user in same transaction. Are you sure that it was working previously or you added some field while retrieving user?

This is the documentation reference https://developer.salesforce.com/docs/atlas.en-us.apexcode.meta/apexcode/apex_dml_non_mix_sobjects.htm

You can update a user in a transaction with other sObjects in Apex code saved using Salesforce API version 15.0 and later if the following fields are not also updated:

  • UserRoleId
  • IsActive
  • ForecastEnabled
  • IsPortalEnabled
  • Username
  • ProfileId

Please try removing profile id and see the results, moreover you need not to insert user to do a runAs you can pass the user if you need to test based on profile

  • 5
    Yes, I know about System.runAs, and I've advised our team that we need to clean up the code. But we have logs showing successful nightly runs until this morning at about 3am MST (America/Denver). Further, the test in the question is passing in a Winter '19 Developer Org and a Spring '19 Sandbox Org. We've been doing deployments twice a week without any problems until today. Our diffs show no changes at all to those classes in months, either, they just started failing. – sfdcfox Jan 14 at 22:03
  • You need to remove the profile ID, request you to see the documentation. why you need the profile id in the soql if you are not using – Avijit Chakraborty Jan 14 at 22:04
  • @AvijitChakraborty it works in Winter 19, so it's not a code issue and SF issue. The code provided by sfdcfox is pseudo code. – Pranay Jaiswal Jan 14 at 22:10
  • If you put different code how anybody can answer we both redirected you so same link. – Avijit Chakraborty Jan 15 at 3:15

protected by Adrian Larson Jan 22 at 19:02

Thank you for your interest in this question. Because it has attracted low-quality or spam answers that had to be removed, posting an answer now requires 10 reputation on this site (the association bonus does not count).

Would you like to answer one of these unanswered questions instead?

Not the answer you're looking for? Browse other questions tagged or ask your own question.