Reduces the performance overhead when a Spark DataFrame as many partitions - especially when using Cosmos DB as a sink in Spark Streaming scenarios #439
The change is possible now because we had introduced the CosmosDBConnectionCache - so we only need to initialize a single CosmosClient (with the metadata requests impacting master RU budget) per executor and follow a singleton pattern otherwise.
The change is possible now because we had introduced the CosmosDBConnectionCache - so we only need to initialize a single CosmosClient (with the metadata requests impacting master RU budget) per executor and follow a singleton pattern otherwise.