Coding Stories - Optimize calls to Azure Table

This week I had the opportunity to make a code review for a startup. They decided to go on Azure from the first moment. The business logic that is behind the application is not very complex, but they are using pretty heavily Azure Storage, especially blobs and tables.
After a load tests, they observed that there are some performance problems on the storage side. The cause of this problem was not Azure. The code that was written didn't used Azure SDK properly.
Below you can find two causes of this problems.

CreateIfNotExists
This method is used when we don't know if a specific resource exists (like blobs or table).

CloudTable fooTable = tableClient.GetTableReference("foo");
foo.CreateIfNotExists();
…

There is not problem with this method as long as we don't have to many calls to it. Each time when we call it, a HTTPS request is send to Azure backend to check if that specific resource exists or not (if the resource doesn't exist, than it will be created automatically).
Normally, we could try to cache the instance of the CloudTable and reuse it as many times we can.
In their case this was not possible because each different client would have a dedicated table where he would add data. When you have 10 clients that are trying to access your system in the same time it is not a problem, but when you have 5.000 or 10.000 clients, than an additional REST call can impact your performance and increase the latency of your system.
In reality they don't need to check if the table exists each time. The table instance can be created in the moment when the user register into the system. Once the table is created, we don't need anymore to check if the table exist or not.
Don't forget that from the cost perspective it is the same thing if we have 1 table or 50.000 tables under an Azure Storage account. Costs and performance are not impacted.

Optimize Queries over Azure Table
Each user of the application is allowed to retrieve data from his own Azure Table. A simple filtering support would allow each user to filter data based on some simple rules.

CloudTable fooTable = tableClient.GetTableReference("fooTable");
TableQueuryDynamicTableEntity customQuery = new TableQueuryDynamicTableEntity()
  .Where(
    TableQuery.CombineFilters(
      TableQuery.GenerateFilterCondition("PartitionKey", QueryComparisons.Equal, "fooPartition"),
      TableOperators.And,
      TableQuery.CombineFilters(
        TableQuery.GenerateFilterCondition("RowKey", QueryComparisons.Equal, "rowKey1"),
        TableOperators.Or,
        TableQuery.GenerateFilterCondition("RowKey", QueryComparisons.Equal, "rowKey2"),
List<DynamicTableEntity> fooResults = fooTable.ExecuteQuery(customQuery).ToList();

In the above code two different queries are grouped together under the same query request. The problem with this query is that it will be executed over all the entities that are under the given table partition. Because of this, the performance of the request is very poor, even if we are making only one API call.
If we already know the partition and row key of the entity, there is no need to define a complex query that will scan all partitions.
Remarks: In Azure Table, each table has to have a partition key and a row key. The combination between this will form the unique key of the entity. For scalability and access speed optimization, Azure can slit a table based on partitions and store each partition in a different logical unit.
To make this code running faster, especially when the number of entities under a partition is high, is to make two different requests to Azure Table. Each request would add a different entity. Even if we are doing multiple requests, the overall performance will be much better.

TableOperation fooOperation1 = TableOperation.Retrive("fooPartition","rowKey1");
TableOperation fooOperation2 = TableOperation.Retrive("fooPartition","rowKey2");
DynamicTableEntity fooEntity1 = (Dynamic<TableEntity>) fooTable.Execute(fooOperation1).Result;
DynamicTableEntity fooEntity2 = (Dynamic<TableEntity>) fooTable.Execute(fooOperation2).Result;

Because the query that is executed over Azure Table is more simple - it doesn't need to search in all partition entities, the performance  will be very high.

Small things like this, can improve drastically the performance of our system. Before using a service or an SDK it is important to understand how it is working and what are the base principles.