Clarification of return values

[jhellan]

Hi

Not strictly an issue, but I did not find an answer in the docs. What are the possible return values from 'save_changes'? I was assuming an array, possibly empty, or an exception. But we have the following lines in our code, where spf_vmm_service is an OData::Service instance:

          @spf_vmm_service.AddToVirtualMachines(vmspec)
          vm = @spf_vmm_service.save_changes[0]

but we occasionally see the error

          undefined method `[]' for true:TrueClass (NoMethodError)

from the second line. It looks like save_changes has returned 'true'. Is that possible, and do you have any idea why?

This is isn't easy for us to reproduce right now - it happened just three times during a 60 hr test run which hit this code abt 1100 times.

Do you have any ideas about this?

Jon Kåre Hellan, UNINETT AS, Trondheim, Norway

[visoft]

Hmm, that's a weird one. My only guess at this point would be that the response from the server is mangled somehow. I'd love to see the response, but 3 times out of 1100 it seems unlikely. To make it easier to capture, edit the save_changes method to verify the result is an array, if not, dump the result to a text file or something. Is that something you could try? I can add a check in there and do something similar. E.g. throw a custom exception, and/or log the output, but I'd like to be able to reproduce the "actual conditions" to test the problem/solution. I have no real idea at this point what would cause that to happen.

[jhellan]

Thanks. I've now instrumented our code. We'll see what that turns up.

We noticed that for batch saves and operations besides 'Add', save_changes returns a boolean. We also see that we get this error a fraction of a second after an earlier failure from AddToVirtualMachines. Is it possible that a failed save may leave behind a save operation in the queue, so that the next save gets turned into a batch save?

We've looked for races in our own code, but the block { AddToVirtualMachines; save_changes } is protected by a mutex. I don't see cases where we call Add but don't call save_changes, either.

[jhellan]

We managed to get a log which made sense. We checked for non-array in single_save, and for batch saves. Our application doesn't do batch saves intentionally.

We got one save_changes call with op AddToVirtualMachines which resulted in "Server returned error but no message. (OData::ServiceError)". There was no associated http_code.

Turned out that the next save_changes call was indeed performed as a batch_save, which returned true. The @save_operations.length was 2.

Is the save_operations queue supposed to be cleared after the first exceptions? If not, how could we clean it up from the application? Would we have to instantiate a new service object?

[visoft]

It doesn't clear the queue after an exception, as you found out. The reasoning here was that if you had 5 changes and one bombed out, you could potentially remove that call. At this point though, there isn't a way to modify the @save_operations array. This should be changed. Perhaps simply exposing that instance variable would suffice, and/or a method to clear out the array.

To reset things, at this point, I believe you'd need to instantiate a new service object like you said.

What would you recommend for a fix? Just expose the array for now?

[jhellan]

Hmm. I think it would be cleaner to make single save and batch save two different operations. If a single save fails, it should be cleared, and it will be up to the user to resubmit.

In a batch situation, can the user tell which ops succeeded and which failed? If he can't, there's obviously no point in keeping the queue around. If he can, I guess it would be best to expose the clear queue operation.