Shadow JSON version number doesn't always increase

[tomghuang]

I tried to build and run the sample under samples/linux/shadow_sample_console_echo to learn the behavior of AWS IoT Shadow Service. I tried to switch the desired windowOpen state between true and false, and expect the device can always receive the delta message. However, about 30% of my testing I received the following message:

WARN:  shadow_delta_callback L#506 Old Delta Message received - Ignoring rx: 222 local: 222

It means that, although I've successfully changed the desired windowOpen from true to false, the device didn't received the delta message, because the SDK consider the received delta message is an old one (with version number 222). This is wrong, because when I change the desired state, the device should receive the delta message.

After tracing the code in aws_iot_shadow_records.c, I found that the bug is caused by receiving two separate messages with the same version number.

In my case, I received an update/accepted message first:

{
  "state": {
    "desired": {
      "windowOpen": false
    }
  },
  "metadata": {
    "desired": {
      "windowOpen": {
        "timestamp": 1468130692
      }
    }
  },
  "version": 222,
  "timestamp": 1468130692
}

Then I received an update/delta message with the same version number (222) and the same timestamp:

{
  "version": 222,
  "timestamp": 1468130692,
  "state": {
    "windowOpen": false
  },
  "metadata": {
    "windowOpen": {
      "timestamp": 1468130692
    }
  }
}

This will cause the following code snippet in the shadow_delta_callback function in aws_iot_shadow_records.c think that the delta message is not new:

if(shadowDiscardOldDeltaFlag) {
    if(extractVersionNumber(shadowRxBuf, pJsonHandler, tokenCount, &tempVersionNumber)) {
        if(tempVersionNumber > shadowJsonVersionNum) {
            shadowJsonVersionNum = tempVersionNumber;
        } else {
            IOT_WARN("Old Delta Message received - Ignoring rx: %d local: %d", tempVersionNumber,
                 shadowJsonVersionNum);
            return;
        }
    }
}

That's because the shadowJsonVersionNum variable has been set to 222 in the AckStatusCallback function when the device received the update/accepted message.

I think the solution is either guarantee that each separate message should have an incremental, unique version number, or the SDK should use a separate variable to track the latest delta message version number.

[chaurah]

Hi @tomghuang,
Thank you for bringing this to our attention. I will discuss this internally with the team and include a fix for consistent version number updates in the next release.

Rahul

[stconnell]

FWIW, I'm working on moving our version of the IoT C SDK (Texas Instruments, CC3200).

I first moved our code base from AWS v1.1.1 --> AWS v2.0.0 - I did not see this issue in the shadow console application.

I have since moved forward, from AWS v2.0.0 --> AWS v2.1.0 - and now I see the "Old Delta Message received" problem.

So, it seems that it's due to some change between AWS v2.0.0 --> AWS v2.1.0

Cheers,

Steve

[chaurah]

Hi @stconnell,
This issue is occurring because we added monitoring for update/accepted in v2.1.0 of the SDK. The change was made here. For now, if this is breaking code, please just remove the check for update/accepted and it should work as expected.

We are still discussing internally how we should maintain consistency for shadow version updates while still being able to receive update/accepted messages. I will update this issue further when we have a fix out. Please let us know if you continue to have the issue after making the above change, or if you have any other suggestions.

Rahul

[stconnell]

Hi Rahul,

Thanks for your prompt reply. I made the change to aws_iot_shadow_records.c to remove the check for update/accepted and the example again works as it did previously.

However, we are a bit hesitant to deviate from the AWS code for the TI release of the SDK (that will sync us up to v2.1.0).

I understand that you're still working on how to best handle this issue, but it would help us to know if you think you expect to leave in the check for update/accepted in the if statement (at line 173) or to keep that check there, and then add some additional logic to handle this properly elsewhere.

For example, if you foresee that you will end up reverting that change to how it was in v2.0.0, then we wouldn't have any issue modifying the code to remove the same (as this would be equivalent to making the change ahead of you guys).

It may make sense to open a private email discussion on this with you (and or the team). If you agree, I'll reach out via email.

Cheers,

Steve

[stconnell]

Hi Rahul,

Per our email discussion, I'm going to go with calling the following function to disable the "discard old delta messages" feature:

aws_iot_shadow_disable_discard_old_delta_msgs()

I like this approach better because I don't have to modify Amazon's framework code. The call to this function can be added to the application, which is better in my opinion.

Thanks for your help.

Cheers,

Steve