Open archerzz opened 2 years ago
Label prediction was below confidence level 0.6
for Model:ServiceLabels
: 'Search:0.11814453,Storage:0.101195894,Azure.Core:0.100758575'
Hi @archerzz, Is there a reason that the order changed? Was there a code change that went along with this? Assuming that is the case, we'd normally recommend just re-recording for situations like this. While it may be possible to do what you suggest here, the serialization format is expected to be stable and unexpected changes should be flagged.
To make re-recording easier, you can use attribute your tests with the RecordedTest
attribute, which will automatically re-record for you on playback failure.
Hi @JoshLove-msft I think some code change in the autorest.csharp
caused the order change.
Re-recording should fix the issue. The only pain point is that some recording will take hours. So I filed this issue as feature request, which can reduce the chances to do a re-recording.
This is not a blocker.
@JoshLove-msft Given that we have moved to the proxy, is this still a valid concern?
The concern is still valid, but this would need to be addressed in the Test Proxy so I can move it to the tools repo.
Library name
Azure.Core.TestFramework
Please describe the feature.
Description
Looks like in the playback mode, the test framework is comparing JSON content by text diff, which will cause false alerts when only the sequence of some properties has been changed.
Example
We see repetitive alerts below after the SDK codes are refreshed.
But actually the content is not changed semantically.
Solution
We should try to compare the JSON body content by key/value, not the text diff.