ChangeFeedIterator gets stuck on multihashed containers for prefix partitions keys that span multiple physical partitions.

[NaluTripician]

Describe the bug
When reading the ChangeFeed for a prefix partition key that spans multiple physical partitions the ChangeFeed will not be able to handle the partition split and will get stuck in an infinite loop.

Expected behavior
Results should be returned.

Actual behavior
ReadNextAsync becomes stuck in an infinite loop.

// Create Container Hierarchical PartitionKeys
this.containerProperties = new ContainerProperties("mycoll", new List<string> { "/State", "/City", "/ZipCode" });
this.container = await database.CreateContainerAsync(this.containerProperties);

// ...
//Container is then filled with enough items span multiple physical partitions 
// ...

//Create a Prefix PartitionKey
Cosmos.PartitionKey prefixPK = new PartitionKeyBuilder().Add("Redmound").Build();

//Create ChangeFeedIterator
FeedRange feedRange = FeedRange.FromPartitionKey(shardPartitionKey);
FeedIterator<dynamic> feedIterator = container.GetChangeFeedIterator<dynamic>(ChangeFeedStartFrom.Beginning(shardFeedRange), ChangeFeedMode.Incremental, new ChangeFeedRequestOptions() { PageSizeHint = 10 });

while (feedIterator .HasMoreResults)
{
    FeedResponse<dynamic> response = await feedIterator .ReadNextAsync(); // BUG: Code gets stuck here!
}

[ealsur]

Related logical investigation steps that yielded the bug:

1. User sends request with PK with less components
2. RequestInvokerHandler converts the FeedRangePK to FeedRangeEPK https://github.com/Azure/azure-cosmos-dotnet-v3/blob/master/Microsoft.Azure.Cosmos/src/Handler/RequestInvokerHandler.cs#L166-L169
3. RequestInvokerHandler finds that the FeedRangeEPK spans multiple partitions: https://github.com/Azure/azure-cosmos-dotnet-v3/blob/master/Microsoft.Azure.Cosmos/src/Handler/RequestInvokerHandler.cs#L239-L251
4. Error bubbles to https://github.com/Azure/azure-cosmos-dotnet-v3/blob/master/Microsoft.Azure.Cosmos/src/Pagination/CrossPartitionRangePageAsyncEnumerator.cs#L111
5. Then it calls into the FeedRangeProvider.GetChildRanges https://github.com/Azure/azure-cosmos-dotnet-v3/blob/master/Microsoft.Azure.Cosmos/src/Pagination/CrossPartitionRangePageAsyncEnumerator.cs#L113-L118
6. Which calls into FeedRangePartitionKey and always returns a point range, even if the PK spans multiple ranges: https://github.com/Azure/azure-cosmos-dotnet-v3/blob/master/Microsoft.Azure.Cosmos/src/FeedRange/FeedRanges/FeedRangePartitionKey.cs#L25-L37 because that method is simply designed for a normal PK that never goes outside a physical partition.

[amitdhar]

Hi, any updates on this? Thanks!

[NaluTripician]

Hi, any updates on this? Thanks!

Working on a hotfix now, should be out next week.

[amitdhar]

Any status updates?

[NaluTripician]

Issue is fixed with latest release of the SDK (3.35.0)