Creating new instances of CloudTableClient and CloudTable per transaction

Question

After doing some research, I am still unsure how best to maintain a 'connection' to Azure Table Storage. Should CloudTableClient or CloudTable instances be reused across requests?

We are using Table Storage behind a public, high-traffic API. We require high read availability and performance. All queries are POINT queries (both partition key and row key are available) and the response payment is small in size (less than 1 kilobyte). Write performance is not a big concern. Each request on the API could read up to 10 point queries across a few partitions.

From my reading, I have understood the following:

• CloudTableClient is not thread-safe and should be created for every transaction. Apparently this shouldn't hamper performance when being recreated continuously.

• A CloudTable instance will thus also have to be created for each transaction.

Are these the correct assumptions to make?

I am thus reinitialising CloudTableClient and CloudTable for every request. It feels wasteful.

See implementation:

public class EntityStorageComponent : IEntityComponent
{
    private CloudStorageAccount storageAccount;

    public CloudTable Table
    {
        get
        {
            var tableClient = storageAccount.CreateCloudTableClient();

            ServicePoint tableServicePoint = ServicePointManager.FindServicePoint(storageAccount.TableEndpoint);
            tableServicePoint.UseNagleAlgorithm = false;
            tableServicePoint.ConnectionLimit = 100;

            var context = new OperationContext();
            context.Retrying += (sender, args) =>
            {
                Debug.WriteLine("Retry policy activated");
            };

            // Attempt delays: ~200ms, ~200ms, ~200ms
            var requestOptions = new TableRequestOptions
            {
                RetryPolicy = = new LinearRetry(TimeSpan.FromMilliseconds(200), 3),
                MaximumExecutionTime = TimeSpan.FromSeconds(60)
            };

            var table = tableClient.GetTableReference("farematrix");
            table.CreateIfNotExists(requestOptions, context);

            return table;                
        }
    }

    public EntityStorageComponent(IOptions<ConfigurationOptions> options)
    {
        storageAccount = CloudStorageAccount.Parse(options.Value.TableStorageConnectionString);
    }

    public SomeEntity Find(Guid partitionKey, Guid rowKey)
    {
        var retrieveOperation = TableOperation.Retrieve<SomeEntity>(partitionKey, rowKey);

        var retrievedResult = Table.Execute(retrieveOperation);

        return retrievedResult.Result as SomeEntity;
    }
}

Answer 1

Apart from the usual overhead of creating an object, I don't see any issue in creating multiple instances of CloudTableClient and CloudTable objects. So if you're simply doing the following, I don't think you're going to get hit performance wise:

        var tableClient = storageAccount.CreateCloudTableClient();
        var table = tableClient.GetTableReference("farematrix");

However I do see an issue with the way you're creating CloudTable in your code (Table member). Essentially in your code, anytime you get the Table property from EntityStorageComponent, you're trying to create a table in your storage account.

        var table = tableClient.GetTableReference("farematrix");
        table.CreateIfNotExists(requestOptions, context);

This is a problem because table.CreateIfNotExists(requestOptions, context); will make a network call and would slow down your system considerably. You may want to move out table.CreateIfNotExists(requestOptions, context); code and put that in your startup code so that you're always sure (mostly) that the table is present.