Read many fix

[jeet1995]

Description

This PR fixes the issue when a point-read based 404 / 0 response drops existing point-read based documents from the resultant FeedResponse instance in spite of the ID of these documents being specified in the readMany request. The expected behavior is to ignore non-existent documents but fetch other documents.

Testing

Test: CosmosItemTest#readManyWithMultiplePartitionsAndSome404s

• Create a multi-partition container - preferably with more than 2 physical partitions.
• Ensure documents are created on each physical partition.
• Query 1 document per physical partition using the FeedRange instance for the physical partition. This is done to ensure when readMany is used to query the documents back, it only uses point-reads to do so.
• Choose n documents in which a non-existent ID is injected.
• Read these documents back using the readMany API.
• Verify if the count of fetched documents through readMany equals (total no. of physical partitions) - n
• Repeat the above steps by varying n from 0 to the total no. of physical partitions.

Benchmarking results

• The fix shows a marginal tail latency increase (in the order of 1 ms to 5 ms)

[azure-sdk]

API change check

API changes are not detected in this pull request.

[jeet1995]

/azp run java - cosmos - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[FabianMeiswinkel]

LGTM - Thanks

[jeet1995]

/azp run java - cosmos - tests

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jeet1995]

There are some metric-related tests which don't pass - would require some looking at. The CosmosConflictTests on the other passes locally.

[jeet1995]

/check-enforcer override