Working with Azure Search in C#.NET


What is Azure Search

Azure Search is a service in Microsoft Azure Cloud Platform providing indexing and querying capabilities. Azure Search is intended to provide developers with complex search capabilities for mobile and web development while hiding infrastructure requirements and search algorithm complexities.

I like the idea of having an index service provider being part of my solution as it allows developers to perform searches quickly and effortlessly without going through the pain of writing essay-length SQL query.

The key capabilities of Azure Search include scalable full-text search over multiple languages, geo-spatial search, filtering and faceted navigation, type-ahead queries, hit highlighting, and custom analyzers.


Why Azure Search (or index provider in general)

  1. It’s Brilliant – Having an index provider sitting in between application and database is a great way to shield the database from unnecessary or inefficient Read requests. This allows database I/O and computing power to be reserved for operations that truly matters.
  2. It’s Modern – Traditional search approach requires developers to write essay-length SQL query to retrieve simple aggregated data. Index provider such as Azure Search allows developers to initiate a search request without writing complicated search algorithm (for example geo-spatial search) and no complex SQL query is required (for example faceted navigation).
  3. It Scales – Solutions that gain the most benefited from index provider are relatively larger enterprise system. Such system often require scaling to certain extain. Scaling Azure Search in Microsoft Azure Cloud Platform is several clicks away compared to having an on-premise service provider such as Solr.
  4. It’s FREE – Microsoft Azure provides a free tier for Azure Search. Developers can create and delete Azure Search service for development and self-learning purpose.

Use Case

A car classified site is a good example to make use of index service provider such as Azure Search. I will use a local car classified site, to illustrate several features the system can potentially tap into.

Type-ahead Queries allow developer to implement auto suggestions as user types. In the following example, as I was typing “merce”, the site returns me a list of potential Mercedes-Benz car model that I might be interested in.


Facet allow developers to retrieve aggregated data (such as car make, car model, car type, car location, car price range, etc) without writing writing complex SQL query, which also means saving the Read load at database. In the following example, the site returns me a list of Mercedes-Benz model and the count in the bracket to indicate how many classified ads are available for the specific model.


Filter allows developer to retrieve documents that fit the searching criteria without writing complex SQL queries with endless INNER JOIN and WHERE clauses. In the following example, I specified that I want all the used Mercedes-Benz, model of E-Class E200, variant of Avantgarde from Kuala Lumpur with the price range of RM100,000 to RM250,000. You can imagine the kind of INNER JOIN and WHERE clauses the poor developer has to design dynamically if this were to be retrieved from a database directly.


Another feature the car classified site can potentially tap into is Geo-spatial Search although it is not seen implemented. For example if I were to search for a specific car in a dealership, the portal can suggest similar cars from other dealerships nearby to the dealership I’m looking at. That way when I make a trip to visit a dealership, I can also visit other nearby dealerships that have the similar cars.

Using C#.NET to work with Azure Search

Let’s roll our sleeves and hack some codes. I will be using a C#.NET console application to illustrate how we can design and create an index, upload documents into the index and perform several types of searches on the index. This solution will be used to simulate some of the potential codes required by a car classified portal.

First, we create a solution name AzureSearchSandbox.

We will need “Microsoft.Azure.Search” NuGet package from Open your Package Manager Console and run the following command:

Install-Package Microsoft.Azure.Search -Version 1.1.3

Upon successful installation, you will see several NuGet packages are added into your packages.config file in your solution.

  <package id="Microsoft.Azure.Search" version="1.1.3" targetFramework="net452" />
  <package id="Microsoft.Rest.ClientRuntime" version="1.8.1" targetFramework="net452" />
  <package id="Microsoft.Rest.ClientRuntime.Azure" version="2.5.2" targetFramework="net452" />
  <package id="Microsoft.Spatial" version="6.13.0" targetFramework="net452" />
  <package id="Newtonsoft.Json" version="7.0.1" targetFramework="net452" />

Note that you will only need to install “Microsoft.Azure.Search”, the other packages are dependencies. The dependencies are resolved automatically. 

In your solution, add a new class Car.cs

using System;

namespace AzureSearchSandbox
    public class Car
        public string Id { get; set; }
        public string Name { get; set; }
        public double Price { get; set; }
        public string Category { get; set; }
        public DateTimeOffset LaunchDate { get; set; }

        public int? SafetyRating { get; set; }
        public override string ToString()
            return $"Id: {Id}\tName: {Name}\tPrice: {Price}\tCategory: {Category}";

This Car object will be used to represent your Azure Search document.

Next we create a static Helper.cs class to take care of the initialization of the index in Azure Search.

using Microsoft.Azure.Search;
using Microsoft.Azure.Search.Models;

namespace AzureSearchSandbox
    public static class Helper
        public const string IndexName = "carindex";

        public static SearchServiceClient Initialize(string serviceName, string apiKey)
            SearchServiceClient serviceClient = new SearchServiceClient(serviceName, new SearchCredentials(apiKey));

            return serviceClient;

        private static void CreateIndex(SearchServiceClient client)
            var indexDefinition = new Index()
                Name = IndexName,
                Fields = new []
                    new Field("Id", DataType.String)                    { IsKey = true},
                    new Field("Name", DataType.String)                  { IsSearchable = true, IsFilterable = true},
                    new Field("Price", DataType.Double)                 { IsSortable = true, IsFilterable = true },
                    new Field("Category", DataType.String)              { IsFilterable = true, IsFacetable = true },
                    new Field("LaunchDate", DataType.DateTimeOffset)    { IsSortable = true, IsFacetable = true },
                    new Field("SafetyRating", DataType.Int32)           { IsFilterable = true, IsSortable = true },


        private static void DeleteIfIndexExist(SearchServiceClient client)

Initialize() is the main method to kick start our Azure Search index. Unlike other on-premise index service that require certain amount of setup such as Solr, it doesn’t take long to have our index up and running. In Solr, I have to install Solr using NuGet, install the right Java version, set the environment variable and finally create a core in Solr admin portal. With Azure Search, no upfront set up is required.

The index is created in CreateIndex() method, where we tell Azure Search client SDK that we want an index with the fields we define in our Index object.

To ensure this set of code is running on a fresh index, we have DeleteIfIndexExist() method to ensure the previous index is removed. We call this right before CreateIndex() method.

Next, we add a new class Uploader.cs to deal with the documents we are about to upload into our Azure Search index.

using Microsoft.Azure.Search;
using Microsoft.Azure.Search.Models;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Threading;

namespace AzureSearchSandbox
    public class Uploader
        private List<Car> PrepareDocuments()
            List<Car> carDocuments = new List<Car>();

            Car c1 = new Car()
                Id = "1",
                Name = "Proton Iriz",
                Category = "Hatchback",
                LaunchDate = new DateTimeOffset(2015, 1, 20, 0, 0, 0, TimeSpan.Zero),
                Price = 55000,
                SafetyRating = 5

            Car c2 = new Car()
                Id = "2",
                Name = "Perodua Myvi",
                Category = "Hatchback",
                LaunchDate = new DateTimeOffset(2004, 6, 15, 0, 0, 0, TimeSpan.Zero),
                Price = 40000,
                SafetyRating = 3

            Car c3 = new Car()
                Id = "3",
                Name = "Perodua Axia",
                Category = "Hatchback",
                LaunchDate = new DateTimeOffset(2014, 12, 25, 0, 0, 0, TimeSpan.Zero),
                Price = 30000,
                SafetyRating = 2

            Car c4 = new Car()
                Id = "4",
                Name = "BMW 320i Sport",
                Category = "Sedan",
                LaunchDate = new DateTimeOffset(2000, 8, 31, 0, 0, 0, TimeSpan.Zero),
                Price = 300000,
                SafetyRating = 4


            return carDocuments;

        public void Upload(ISearchIndexClient indexClient)
                var documents = PrepareDocuments();
                var batch = IndexBatch.Upload(documents);

            catch (IndexBatchException e)
                    $"Oops! The following index failed...\n { e.IndexingResults.Where(r => !r.Succeeded).Select(r => r.Key) }");


PrepareDocuments() is a simple method to construct a list of dummy Car object for our searches later on.

Upload() method gets the dummy Car objects from PrepareDocuments() and pass these object into Azure Search client SDK to upload the documents into index in a batch. Note that we added a 2000 millisecond sleep time to allow our service to upload and process car documents properly before moving on to next part of the code, which is search. However in practical sense, we would not want to add sleep time in our upload code. Instead, the component that takes care of searching should expect index is not available immediately. We also catch IndexBatchException implicitly to handle the index in case the batch upload of index failed. In this example, we merely output the index key. In practical sense, we should implement a retry or at least logging the failed index.

Once the index upload operation is completed, we will add another class Searcher.cs to take care of the searching capability.

using Microsoft.Azure.Search;
using Microsoft.Azure.Search.Models;
using System;
using System.Collections.Generic;

namespace AzureSearchSandbox
    public class Searcher
        public void SearchDocuments(ISearchIndexClient indexClient, string searchText, string filter = null, List<string> facet = null)
            var sp = new SearchParameters();

            if (!string.IsNullOrEmpty(filter))
                sp.Filter = filter;

            if (facet != null)
                sp.Facets = facet;
            DocumentSearchResult<Car> response = indexClient.Documents.Search<Car>(searchText, sp);

            if(response.Facets != null)
                FacetResults facetResults = response.Facets;
                foreach (var facetResult in facetResults)
                    Console.WriteLine($"Facet Key: {facetResult.Key}");
                    foreach (var f in facetResult.Value)
                        Console.WriteLine($"{f.Value}, ({f.Count})");
                foreach (SearchResult<Car> result in response.Results)

SearchDocuments() method is to handle the searching mechanism on the index we created earlier. No fancy algorithm, only passing specific instruction to Azure Search client SDK on what we are looking for and display them. In this method, we take care simple text search, filter and facets. There are much more capabilities Azure Search client SDK can provide. Feel free to explore the SearchParameters and response object on your own.

Putting them all together in Program.cs

using Microsoft.Azure.Search;
using System;
using System.Collections.Generic;

namespace AzureSearchSandbox
    class Program
        static void Main(string[] args)
            string searchServiceName = "daniel";

            string apiKey = "38E9EC1BBC7F012747BE72FDBECDFB3F";
            SearchServiceClient serviceClientApi = Helper.Initialize(searchServiceName, apiKey);
            ISearchIndexClient indexClientApi = serviceClientApi.Indexes.GetClient(Helper.IndexName);

            Uploader uploader = new Uploader();

            Searcher searcher = new Searcher();

            Console.WriteLine("Begin searching Perodua....");
            searcher.SearchDocuments(indexClientApi, "Perodua");

            Console.WriteLine("Begin searching Hatchback category....");
            searcher.SearchDocuments(indexClientApi, "*", "Category eq 'Hatchback'");

            Console.WriteLine("Begin searching Price more than 100,000....");
            searcher.SearchDocuments(indexClientApi, "*", "Price gt 100000 and Category eq 'Sedan'");

            Console.WriteLine("Begin searching by Category facet");
            List<string> facets = new List<string>();
            searcher.SearchDocuments(indexClientApi, "*", null, facets);


First we define index service name and API key to create a search index client instance. The instance is returned by Helper.Initialize(). We will make use of this instance for both search and upload later.

After initializing the index, we call Upload() method to upload some dummy car documents to the index.

Next, we perform the following searches:

  1. Simple text search. We will search for the text “Perodua” in the documents.
Console.WriteLine("Begin searching Perodua....");
searcher.SearchDocuments(indexClientApi, "Perodua");

The result as following. Azure Search index returns 2 documents which contains the keyword “Perodua”


2. Using Filter. A more targeted and efficient approach to look for documents. In the following example, first we look for Category field which is equal to ‘Hatchback’; second we look for Price field which is greater than 100,000 and is a ‘Sedan’ category. More details on how to write expression syntax in Azure search.

Console.WriteLine("Begin searching Hatchback category....");
searcher.SearchDocuments(indexClientApi, "*", "Category eq 'Hatchback'");

Console.WriteLine("Begin searching Price more than 100,000....");
            searcher.SearchDocuments(indexClientApi, "*", "Price gt 100000 and Category eq 'Sedan'");

The result as following: Car with the category of Hatchback and car cost more than 100,000 and is a Sedan.


3. Searching facets. With facets, developer will no longer need to write long query that combines Count() and endless GROUP BY clauses.

Console.WriteLine("Begin searching by Category facet");
List<string> facets = new List<string>();
searcher.SearchDocuments(indexClientApi, "*", null, facets);

If we have a traditional database table that represent the dummy car data, this is equivalent to “SELECT Category, Count(*) FROM Car GROUP BY Category”.

Result as following:


This example might not look like a big deal if you have a small data set or simple data structure. Facet is super handy and fast when you have large number of data and when your query is getting more complex. The ability to define which facet is required in C# codes make the”query” much cleaner and easier to maintain.

One last thought…

You can clone the above source code from my GitHub repo. I will be happy to accept Pull Request if you want to demonstrate how to make use of other capabilities in Azure Search client SDK. Remember to change the API key and service name before compiling. Have fun hacking Azure Search!

Entity Framework vs Stored Procedure


When developer queries SQL Server for data, Entity Framework (EF) and Stored Procedure (SP) are 2 of the most common options. Often, individual preferences could be debatable.

In an existing system I am working on, there are about 40+ stored procedures being implemented.

Stored procedure adds complexity for maintenance. For example, you need to ensure certain version of the SP is compatible with certain version of the codes. During deployment, you need to deploy code + SP instead of just codes. Imagine rolling back changes if your deployment fails. Another factor is debugging. Debugging codes is obviously more pleasant than debugging SP.

I prefer to only use my database as a medium of storage. I like to keep the database as simple and clean as possible and off load other works to external components or services that are best at doing what they are designed for. For example, if I need fast searching I prefer index service such as Solr / Azure Search by designing relevant facets rather than building up big indexes that add storage overhead to database; if I need to process business logic, I will do it in application because it is much easier to scale an application over a database.

Having said that, SP has it’s advantage such as being pre-compiled, it is included in the execution plan hence has performance advantage. The reusability factor of SP is debatable because if the codes is structured in a reusable manner, the codes that stored the database logic can definitely be reused just like a SP.

Since the system I’m working on has a bunch of SP (and the main argument is performance). Out of curiosity, I did a performance profiling comparing EF vs SP on an actual database.

Some initialization codes…

            int loop = 10;
            Guid id = Guid.Parse("XXXXXX-XXXX-XXXX-XXXX-XXXXXXXXXXXX");
            string someId = "XXXXXXXX";

            DateTime now = DateTime.Now;
            Stopwatch stopWatch = new Stopwatch();
            Stopwatch stopWatchScope = new Stopwatch();
            string milisecond = string.Empty;
            TimeSpan ts = new TimeSpan();

Codes to query database using Entity Framework (with LINQ) that I re-implement base on an existing stored procedure:

            MyDatabaseEntities context = new MyDatabaseEntities();
            Console.WriteLine("Time for Entity Framework");
            for (int i = 0; i < loop; i++)
                var query = from c in context.Customers
                            join cs in context.CustomerStatus on c.PK_CustomerStatus equals cs.PK_CustomerStatus
                            join p in context.Programs on c.ProgramID equals p.ProgramID
                            where c.CustomerID == id
                                && (cs.CustomerStatusCode == "Condition 1" || cs.CustomerStatusCode == "Condition 2")
                                && p.someID == someId
                                && p.StartDate < now && p.EndDate > now
                            select c;
                Customer customer = query.ToList().FirstOrDefault();
                var a = customer.CustomerID;
                ts = stopWatch.Elapsed;
                milisecond = ts.TotalMilliseconds.ToString();
                Console.WriteLine($"Attempt {i}: {milisecond} milisecond");

            ts = stopWatchScope.Elapsed;
            milisecond = ts.TotalMilliseconds.ToString();
            Console.WriteLine($"=>Total time for EF: {milisecond} milisecond");

Codes to query database using Stored Procedure:

            SqlConnection conn = new SqlConnection();
            conn.ConnectionString =
                "Data;" +
                "Initial Catalog=xxxxxxx;" +
                "User id=xxxxxx;" +

            SqlCommand cmd = new SqlCommand();
            SqlDataReader reader;

            cmd.CommandText = "MyExistingSimpleStoredProcedure";
            cmd.CommandType = CommandType.StoredProcedure;
            SqlParameter parmId = new SqlParameter("@CustomerID", SqlDbType.UniqueIdentifier);
            parmId.Value = id;
            parmId.Direction = ParameterDirection.Input;
            SqlParameter parmSomeId = new SqlParameter("@someId", SqlDbType.VarChar);
            parmSomeId.Value = someId;
            parmSomeId.Direction = ParameterDirection.Input;

            cmd.Connection = conn;

            Console.WriteLine("Time for Stored Procedure");
            for (int i = 0; i < loop; i++)
                reader = cmd.ExecuteReader();
                ts = stopWatch.Elapsed;
                milisecond = ts.TotalMilliseconds.ToString();
                Console.WriteLine($"Attempt {i}: {milisecond} milisecond");

            ts = stopWatchScope.Elapsed;
            milisecond = ts.TotalMilliseconds.ToString();
            Console.WriteLine($"=>Total time for SP: {milisecond} milisecond");

Where this is the query in the stored procedure:

		C.EndDate As CustomerEndDate,
		C.StartDate As CustomerStartDate,
	From dbo.Customer  C with (nolock)
	Inner Join [dbo].CustomerStatus CS with (nolock) On C.PK_CustomerStatus = CS.PK_CustomerStatus 	And   cs.CustomerStatusCode in ('Condition 1','Condition 2') 
	Inner Join Program P  with (nolock)  On P.ProgramID  = C.ProgramID
		Where  C.CustomerID = @CustomerID
		AND P.SomeID = @someId
		AND (@date BETWEEN P.StartDate AND P.EndDate)

Here are the result after 5 rounds of execution:

Stored Procedure won

Stored Procedure won

Entity Framework won

Entity Framework won

Stored Procedure won

The overall winner is Stored Procedure, where Stored Procedure won 3 times while Entity Framework won 2 times.

A few interesting insight from the profiling:

  1. Stored Procedure performed marginally better in overall.
  2. Entity Framework is marginally slower but it is not as slow as making Stored Procedure a clear winner.
  3. The first call for Entity Framework is significantly higher compare to the consecutive calls. If we were to exclude initial call in EF, EF will be a clear winner.

Choosing between Entity Framework or Stored Procedure might not be straight forward especially if you have an existing application that has massive implementation in either approach. However here are some guidelines to make your decision making less painful.

  1. If you have an existing system that has massive implementation on either approach, stick with the existing approach.
  2. If you have existing development team members who have strong background in stored procedure or entity framework respectively, continue to tap onto their strength.
  3. If you are starting a new project, go for Entity Framework for ease of development and good maintainability although the cost to pay is slight performance downgrade.

Until next time!

Technical Interview Part 2


What I do instead

Before technical interview session, a technical assignment given to the candidate on Friday, so that they can work on it over the weekend. The technical assignment typically takes few hours to complete. I have a range of questions from demonstrating a design pattern, build a simple application with database interaction, to SEO analysis algorithm.

The objectives of the technical assignment are:

  1. Ensure the candidate can codes.
  2. Evaluate how modern his development approach (for example, whether the candidate will use Elmah or Log4Net over writing a custom logger class)
  3. Evaluate how serious he treats his codes (if a candidate deliver a half-hearted solution, it indicates the same for his work codes)
  4. Evaluate whether he goes the extra miles (such as implementing unit tests and proper exception handling)

During the technical interview, I will do the following:

  1. Have the candidate to explain the core of the solution. I will then ask a few questions base on his implementation. For example if I want to implement certain change, where should I modify the code. This is to ensure I’m talking to the person who wrote the codes.
  2. Find a flaw in the system and press on it again and again – in a professional and respectful manner. This is to evaluate how well the candidate respond to criticism, whether the candidate get defensive and whether the candidate is open to feedback.
  3. Challenge the candidate on how he can upgrade his solution to be production ready on both code and infrastructure level. This is to evaluate how much thought he has given to his solution and how much of exposure the candidate has dealing with production system.

Next, I will move on to a list of generic topic on software development. Example of the topics I cover.

Source Control

Every developer uses source control to certain extend. I will normally ask what kind of source control has the candidate use. The top 3 answers are TFS, Git and SVN. I will ask the candidate to share with me, what are the differences between the top 2 source controls he is familiar with. The idea here is to discuss about what the candidate is familiar with so that he can show his best thoughts. Depending on what the candidate bring up to the table, I will get a sense of what kind of developer the candidate is.

For example, if a candidate tells me checking out a branch in TFS is downloading a whole new copy of the code; while Git is merely applying the delta difference on the same copy of the code, it indicates this candidate used to work with giant code base with some level of branching experience and he appreciates Git is much more efficient on client side storage.

Another example, if the candidate brought up terms such as Rebase, I will follow up by asking what is the difference between Rebase and Merge on theoretical level and when is a good scenario to use Rebase over Merge on practical level. Depending on the scenario given, I might (or might not) have further question to validate the usages. The idea here is I’m following up on the topics suggested by the candidate himself. If a candidate cannot provide solid evidence on how familiar is he with the topic suggested by himself, that indicates the candidate is throwing fancy terms around hoping to impress the interviewer.

Design Pattern

Despite the challenges I highlighted earlier on design pattern, I still think design pattern is a good topic to cover during technical interview because the right application of design pattern indicate the complexity of codes the candidates has dealt with, hence the need of design pattern.

Ever since I take the role of being an interviewer, I do make it a point to read up on additional design patterns that I have never used. The good news is, most candidates consistently brought up only a handful of design pattern. The top 3 are such as Singleton, Abstract Factory and Repository.

Although Dependency Injection is not strictly a design pattern, a lot of candidates did mention Dependency Injection as something they know under the design pattern topic. I do not dismiss this answer just because it did not fit into the definition. My objective here is to assess the candidate’s ability to design his code structure, not a competition of giving definition.


What do I look out for during Design Pattern discussion? Take Singleton for example. After the candidate mentioned he knows Singleton, I will follow up with question “What is a good use case to use Singleton?”. The typical answer I got is something along the line “when you only need to have a single instance of the class”. Good. At this point, I know that the candidate is aware of the definition of Singleton although did not provide the use case I asked for. I will rephrase my question slightly differently to remind the candidate I’m looking for a use case.

One “interesting” answer always pop up is to apply Singleton in data layer (CRUD operation to database). I call this interesting because anyone who give a little more thoughts or have done some research in Singleton will realize it’s a bad idea to apply the pattern in data layer. However, this misconception comes up very frequently.

I will take this opportunity to explain to the candidate the kind of problem will surface for applying Singleton in data layer. Why do I do that? Having the right skill or information is important, but having the right attitude is equally (if not more) important. You can teach someone new skill but it is extremely challenging change someone’s attitude. At this point if the candidate appears to be enlighten with the new information, I know the candidate is coachable. In most situation, I would rather to have a coachable new hire (although not having the top notch skill set) over someone with superstar skill set but a poor attitude.


Working with database will come across a developer’s path very frequently. It is an unspoken rule that a developer must be able to work with database. With the amount of storage options in the market, it is difficult to discuss all of them but we will stick with the most popular option for most .NET developer – SQL Server in this article.

When hiring a junior developer, the candidate will have to prove his ability in writing T-SQL. Insert, Update, Delete and different kind of joins. No big deal. For senior developer, I would normally ask the candidate what other exposure does he has apart from T-SQL. Asking the actual involvement in SQL Server gives me very good indicator what kind of system the candidate has dealt with.

For example, if the candidate claimed he takes care of database backup, I will follow up with what is the backup cycle and types of back up he was using. If all the candidate did was doing a full back up on daily basis, it indicates the database size he was dealing with was not very large and the data lost does not seem like a big deal which means the data is not extremely critical.

If the candidate mentioned he scales SQL Server, I will follow up with what type of replication he applied and what is the rational behind the decision. I will also ask what other strategies he has considered before using replication because replication is an expensive option. If the candidate brought up Redis cache and index provider such as Solr or Azure Search, it shows the candidate has looked beyond SQL Server context which indicates he is someone having very broad skill set across technologies.

Once a candidate told me he implemented table partitioning in his database. I asked what is the logical condition he applied his partition base on. He said primary key which is GUID data type. That was an interesting answer because the generally approach to create partition is to base on date or some other logical conditions. I explained to him how I would implement table partitioning instead and the reason behind it. His eyes were brighten up.

Notice that I did not say “This is wrong. The correct way is this”. Instead, I make it as a discussion on “This is what I will do instead”. The same information was delivered across, but the outcome will be very different.

The candidate impressed me because he knows about table partitioning that most developer don’t. It suggests that the candidate is someone who took the extra effort to learn new skills to solve problems. Most importantly, the way he responded to the information I shared with him suggests he is someone coachable. I took this candidate into my team and he has proven to be a star team member.

Few final thoughts…

There are a lot of other topics that I cover during the interview. Most of them are generic topics such as tweaking software performance and security. The purpose of having a standardized list of topic is to ensure I use the similar benchmark for candidates for the same position. The reason to start with generic topic and drill further down is to allow the candidate to talk about areas that they are familiar with so that they can showcase their sharpest thoughts.

When candidate brought up certain topic for discussion, I’m assuming he knows about the topic very well. I’m handing over the power to drive the discussion to the candidate to certain extend. I prefer to talk about what the candidate is familiar with (instead of mine) so that I can truly assess his level of technical competency. Frankly, there is very little value to talk about a topic the candidate has only read an articles on 6 months ago. However, whichever topics that the candidate brought up, I will drill really deep to ensure he indeed knows about them rather than just throwing some fancy words around. 

During technical interview, I’m looking at more than just technical skills. Technical skills is learnable. What really interest me are:

  • Whether the candidate is coachable?
  • How big of passion the candidate has over technology?
  • What is the candidate’s approach in solving problems?
  • What is the candidate’s attitude dealing with technology and PEOPLE?
  • How much of potential the candidate has so that the company can groom him to be a superstar developer and beyond?

The technical topics I have for the candidate were merely for me to expose those areas I’m interested to learn about the candidate. I’m never interested to know the difference between a clustered index vs non-clustered index or the difference between Azure Web Job vs Azure Worker Role vs Azure Function. Given a laptop with internet, anyone can Google them in 5 seconds. What I am interested to discover is whether this candidate is coachable, his passion, his approach, his attitude and his potential!

Ideally, we should hire the right person with the right skill. However such angels rarely come by. If I have to choose between the right person or the right skill, I will choose the right person any day. Of course, provided the candidate still has reasonable level of skill set on the role he is applying. New skills are learnable and very often it is very quick to learn a new skill. Coaching a person takes a much more time, energy and cha-ching – if you are lucky.

If you are not lucky, a bad apple not only bring down productivity but also break the current harmonious team. It is much more effective to filter the potential troublemaker than to “coach” or “develop” him later. There is no point hiring bad apples just to hit headcount. With people, slow is fast.

Some companies practice having a couple strong technical guys to interview candidates whom they might not eventually work with. The interviewers are hiring for the company wide. Some companies practice having the Team Lead / Architect within the team to interview the candidates whom they will eventually work with. They are hiring for the team. I have been in both the situations and personally I prefer to the latter.

Being able to work with the person whom I interviewed earlier will give me additional consideration and deeper thoughts into whether the candidate will be a good fit into my team. Another good reason is to allow me to validate and refine my interview techniques. Interview is all about perception and assumption made on the candidate. I have made good decisions and I have made bad decisions. However, in the situation where I made a wrong assumption base on a wrong perception, I can adjust my interview technique on a continuous basis if I have first hand experience working with the candidate I interviewed.

Finally, I don’t claim what I’m doing is the only way or the best way. We live and we learn 🙂 I found this approach to be working quite well hence I continue practicing. If you have any thought on this, please leave me a comment. Hope you have found something useful in this article. Until next time. Cheers!

Scaling SQL Server


Scaling database refers the ability to serve significantly more request to both read and write data without compromising performance. In many enterprise applications, performance bottleneck often happens at database, hence scaling database is a critical part on improving system performance. In the last article on Microservices, we discussed scaling database horizontally and vertically on high level. In this article we will talk more in-depth about:

  1. Three types of replication in SQL Server
  2. Distribute database load using log shipping
  3. Tools to shield your database from being hit.

3 Types of Replication

In typical enterprise applications, Read request significantly outnumber Write request. By implementing Replication, you effectively offload the bulk of Read request to the Subscribers while reserving Publisher for Writing.

Transactional Replication

Transactional Replication is the simplest form of replication to understand and to implement. Transactional Replication is implemented by having a Publisher to publish the changes. One or more than one Subscriber will replay the transaction log. Data changes and schema modifications made at the Publisher are delivered to the Subscriber(s) as they occur (almost real time). In this way, transactional consistency is guaranteed.

The incremental changes in Publisher will be propagated to Subscribers as they occur. If a row changes for 3 times in Publisher, the Subscriber will also change for 3 times. It is not just the net data change that get propagated over.

For example, if a row in Product table changes price three times from $1.00, to $1.10, to $1.20 and finally to $1.30, transactional replication allows an application to respond to each change. Perhaps, send a notification to user when the price hit $1.20. It is not simply the net data change to the row by changing the price from $1.00 to $1.30. This is ideal for applications that require access to intermediate data states. For example, a stock market price alert application that tracks near real time stock price changes to send price alert to users.

Is it possible to scale your Publisher horizontally? Yes, Bidirectional Transactional Replication and Peer-to-Peer Transactional Replication will help you to achieve it. However, Microsoft strongly recommend that write operations for each row be performed at only one node – for 2 reasons. First, if a row is modified at more than one node, it can cause a conflict or even a lost update when the row is propagated to other nodes. Second, there is always some latency involved when changes are replicated. For applications that require the latest change to be seen immediately, dynamically load balancing the application across multiple nodes can be problematic.

From experience, the most optimum solution is to scale your Subscriber horizontally by having multiple nodes. Keep your Publisher in one node and scale the Publisher vertically, when you really have to.



Merge Replication

In Merge Replication, Subscriber synchronizes with the Publisher when connected to the network and exchanges all rows that have changed between the Publisher and Subscriber since the last synchronization occurred. You may see Merge Replication as a batch update from Subscriber to Publisher that propagates only the net data changes. For example, if a row changes five times at a Subscriber before it synchronizes with Publisher, the row will change only change once at the Publisher to reflect the net data change (which is the 5th value). Then, the unified changes in Publisher will be propagated back to other Subscribers.

Merge Replication is suitable for situation where Subscribers need to receive data, make changes offline, and later synchronize changes with the Publisher and other Subscribers. For example a nationwide POS (point of sale) system where retail branches are spread across multiple physical locations. The retail branches will first initialize a snapshot from the Publisher database, make local offline changes (for example, through sales) to Subscriber database. The sales are not required to be propagated back to Publisher immediate. The many other retail branches also do not need immediate update on the changes happened at another retail branch although a more recent update will be beneficial, for example knowing whether another branch nearby has the stock that the local branch has run out to recommend customers where to go accordingly. Once a day or multiple times a day, depending on business need, the sales number in retail branches (Subscribers) will be propagated back to HQ (Publisher) and the executives in the HQ office can view the daily sales report.


Conflict can happen in Merge Replication and conflict will happen. Good news is, conflicts are resolved without the need of user intervention because SQL Server has built-in mechanism to resolve conflicts on data changes. However if you have unique use cases where you want to ensure SQL Server is doing exactly what you intended while resolving a conflict, you can view the conflict at Microsoft Replication Conflict Viewer and the outcome of the resolution can be modified.

Snapshot Replication

Snapshot Replication propagate data exactly as it appears at a specific moment in time and does not monitor for updates to the data. When synchronization occurs, the entire snapshot is generated and is sent to Subscribers. In simpler terms, Snapshot Replication takes a snapshot of the Publisher data state and overwrite it at the Subscribers. No conflict will happen as this replication basically overwrite the whole data set. Snapshot Replication is also used in both Transactional Replication and Merge Replicate to initialize the database at the Subscribers.


Snapshot Replication is suitable for system where

  • Subscriber deal with data that does not change frequently.
  • Subscriber do not require the most recent set of data for a long time.
  • The data set is small.

I was working on a small project with an advertising agency where the requirement is to analyze the trend of discussion and sentiment related to Telcos in a community forum. I quickly hack some web scraping codes to scrap what I needed and populate them into my database. As a result, I got 730 discussion topics and the total size of the database is about 5MB. From there I need to write some more algorithm to find out the trend and sentiment of discussion. During the development of my algorithm, I did not need the most updated dataset that reflects what is happening in the live forum. I pleasantly work (read) on a Subscriber node to develop and test out my algorithm. Few weeks later when the forum is updated with a lot more discussion topics, I simply replicate the changes in Publisher to my Subscriber in a fairly short amount of time. On production, knowing the analytical codes read database extensively, I pointed my codes to run on the Subscriber node. My Publisher node will not get any hit apart from getting the necessary insert operation from my web scraping service. Since my users do not need to know what was being discussed on a daily basis or on real time basis (because accurate trend and sentiment require months and months worth of data), I have configured my Snapshot Replication to happens on a monthly basis. The replication can be completed in a fairly short amount of time. On monthly basis the users will get a fresh copy of the trend and sentiment report base on at least 1 year worth of backdated forum discussion data without any performance downgrade. By implementing Snapshot Replication, I allow my web scraping service to write to Publisher without worrying if anyone need the database for reading to generate report at any particular time. Through Subscriber, I have also set up the foundation to make it possible to generate much more sophisticated reports without downgrading the performance by spinning up more Subscriber nodes when and if I need to.

Log Shipping

Log Shipping is used to automatically send transaction log backups from primary database to one or more secondary databases. The primary and secondary database should sit on different nodes. The transaction log backups are applied to each of the secondary database. Log Shipping is often used as part of disaster recovery strategy but creative database administrator often use Log Shipping for various other purposes *wink*.

In one of the projects I was working on, the SQL Server database was storing 130k registered users and their related activities such as payment history, credit spending history, Account & Contact relationship, products, login audit, etc. The company was at a rapid expansion stage where the CFO decided it’s time to get in a Business Intelligence guy to churn out some reports to give a sense how the business is doing on daily basis. The obvious thing to do here is to replicate a database for the BI colleague to run his heavy queries because running the reporting queries on production database is going to kill the poor database. The most suitable type of replication will be Transactional Replication. However the challenge was Replication requires SQL Server Enterprise Edition and we were running on SQL Server Standard Edition. The bad news is, the company did not have a lot of cash lying around for our disposal. We will have to find an alternative. After getting the green light from CTO, I implemented Log Shipping for BI reporting. In essence, I was “scaling” the SQL Server using a disaster recovery technique by offloading the reporting query load to a secondary database by replaying the transaction log to simulate Transactional Replication on an interval basis. It was the most optimum option we have in order to satisfy various stakeholders while keeping the cost low.


You can use this trick as long as your client does not require real time data. Take note on the following practical issues while “scaling” your SQL Server using Log Shipping.

  1. Understand that Log Shipping is a 3 steps process. First, primary database backup the transaction log at the primary server instance. Second, copy the transaction log file to the secondary server instance. Third, restore the log backup on the secondary server instance. From experience, third step Restore is the most fragile step where it broke often. To find out why Restore fails, go to SQL Server Agent to view the job history to see the detail error message.
  2. Understand your transaction backup cycle. If you have another Transaction Log backup automation happening by another agent / service, your Log Shipping will stop working fairly quickly (note: not immediately). Log Shipping works by taking all the transaction log since the last Full backup and clear off the log. It is important for SQL Server to match to log sequence. If there is another Transaction Log backup happened somewhere else, the tail of your newest Transaction Log will not match the head of last Transaction Log hence fail. If you need to use Log Shipping, disable / stop all other Transaction Log backup.
  3. Move your agent job interval up gradually. Note that Backup, Copy, Restore jobs are run by SQL Server agent on an interval. When you are setting up your Log Shipping, after the initial full database backup is restored at secondary server, your Transaction Log is ready for action. When do your Backup, Copy, and Restore jobs kick in depend on what is the interval you set while configuring your Log Shipping. I recommend you set a super short interval in the beginning so that you can monitor the failure in your setup fairly quickly. You do not want to wait for 6 hours for your Backup, Copy, Restore jobs to kick in to find out they failed, you then make some changes and wait for another 6 hours. I always start with 1 minute. Then, 5 minutes then to the actual time frame depending on business requirement. In my case, it was 2 hours.
  4. Enable Standby mode so that your client can read the data. I highly recommend you to check “Disconnect user in the database when restoring backups.”. SQL Server Agent Restore job will handle the Transaction Log intelligently by replaying the log that was missed previously. However as I mentioned earlier, Restore job is the most fragile step. Sometimes when Restore job breaks, you have to resetup the whole log shipping mechanism. Imagine a database size of 300GB (the actual size I was working with), it is pretty painful to wait for the whole process to complete. Hence, to ensure the integrity of the Transaction Log sequence, I would rather terminate all open connection to ensure my Restore step can be executed successfully. standby-mode
  5. Monitoring your Log Shipping continuity. Again, Log Shipping is pretty fragile especially if this is the first time you are doing it. You need certain mechanism to monitor your Log Shipping to ensure they are still running as expected 6 months down the road. You can open up SQL Server Agent to view the job history ensuring they are all green or configure Alert Job to raise an alert when job does not complete successfully. But personally, nothing is more assuring than knowing the data actually changes in the secondary database during specific time frame. What I do is monitoring a table column that is supposed to change after replay of Transaction Log. In my case, I monitor the latest user login time because I know this table is updated fairly frequently. The probability that no one login in the last 2 hours is close to zero. I make sure that every 2 hour the value in column changes. If you do not have a user login audit, you can make use of any table that you are sure the data will most likely change, for example your CreatedOn column on the highest transaction table.

Shielding your database

Shielding your database is a great way to keep your database load low so that you can serve more requests. But if you shield the database, where does the data come from? The data has to come from somewhere and that somewhere is known as index provider. I’m not referring to building more indexes within SQL Server because indexes served from SQL Server still add load to SQL Server and not to mention additional storage on disk. The index provider I was referring to are fast-searching services such as Solr, Elasticsearch, Azure Search, or even Redis.

This approach is to drastically offload the reading at database level to another service which are built for super fast searching. Not only the response time is much faster, you will save the poor programmers from write essay-length query to retrieve data.

Example 1: You have a clean 3rd normalized form database. With SQL Profiler you observed that clients have been issuing long query such as a query with 15 JOIN and 5 GROUP BY on very large tables. Not only the that query result comes back slowly, it also drags down the server performance and other queries are badly affected. You have discussed with your best SQL query guru to review the queries, studied the execution plans, revisited the data structure and the conclusion is that is really what it takes the get the desired result. So, what do you do now?

Index provider will come into rescue when you design your schema for your query. The joining is no longer required because the index has flatten the defined fields during indexing. The grouping is also no longer required because that info come from facets. Now instead of the JOIN and GROUP BY happen in SQL Server level, you get the data from an index lightening fast – as a simple read provided you have designed your index schema properly.


Example 2: You are working with 2 microservices where each of the service has its own database keeping different domain data. Your application requires data from both the services and both of them are responding to you slowly. Again, after reviewing the request with respective microservice owner and you all came to the conclusion that both the codes and SQL statement sare at the most optimum form, what do you do now?

Index provider will come into rescue when you index both the data from different microservices into one index provider and you query from the index provider instead. Not only you bypass the two slow microservices, you are also querying from one index instead of multiple databases. Of course, this is easier said than done because now it involves the two microservices to index the required data and additional effort to maintain the index provider.


Here are some of the practical tips you can consider while implementing index provider to shield your database from getting hit:

  1. Initial indexing will take a lot of time, especially from various sources. This is one of the reason why some deployments take long time. A trick to overcome this is to not relying on the on-the-fly index API call provided by index provider. Export your data as .csv and import them for indexing. Do it manually if your data is large enough. Automate this process if full reindexing happen on recurring basis.
  2. Minimize your index schema changes. Every schema change will require you to reindex your indexes, which usually take time and lead to longer down time. The key is when designing your schema, think holistically what is the use case and potential use cases instead of designing a schema just for one client and one use case. In the meantime, you also do not want to design an overly generic schema else it will be inefficient. The art is finding this balance and the way to do it is to understand your domain and context well before designing your schema.
  3. Ensure your stakeholders are aware that index is a reflection of your database. Your database remains as the single source of truth. Expect delay in your index. Eventual consistent is what you aim to achieve. Often, it is acceptable to have a few seconds or even minutes delay depending on how critical the data is. Ensure you get the acknowledgement from your stakeholders.
  4. Queue your indexing. A simple phone number update into database could result in 100 of reindexing request. Have a queueing mechanism to protect your index provider from sudden surge of reindexing request.
  5. Take note of the strongly type fields. Index provider such as Solr takes everything as string while index provider such as Azure Search has strongly-typed field (Edm.Boolean, Edm.String, Edm.Int32, etc). If you plan to switch index provider in the future, take care of the data type from day one else you will end up having an additional layer of mapper to deal with the data type later on.

Hope these help you in your journey to scale your SQL Server. Have fun!



I was first introduced to Microservices Architecture in 2014. At that point of time, I have no idea what I was doing is known as Microservices. We designed our system that way simply because it made practical sense. We started off with 1 PHP web service, 1 PHP frontend application, 1 .NET web service, 2 .NET frontend applications and a CRM. The number of Microservices grow along with the business needs. Since then I learned that managing Microservices is as interesting and fun as building Microservices.

Microservices are relatively small applications that interacts with each other to achieve specific business requirement. In each Microservice, they are designed to do one thing and to do it really well. Sometimes, they are independent on their own but they often work together to accomplish more complex business requirement. The opposite of Microservices is a monolithic system, the kind of system where you have 5,000,000 line of codes in single code base.

Why do we use Microservices?

Technology Heterogeneity – It means the diversity of technology within a solution. Your Microservices have independent stacks of technology. You can choose the most suitable stack of technology depending on the problem you are solving. For example, a Photo Album Printing business might have a PHP frontend application (because the want to tap into WordPress as CMS), a .NET backend business rules exposed as Web API (because there is a legacy logic and SQL Server database), a Java image processing engine (because there are proprietary image processing libraries written in Java), and an R application to crunch big data on customer sentiment. In Microservices architecture, you can have different stacks of technologies that work together seamlessly. They interact through set of API exposed to each other.


This reason also align with Scrum. Each Scrum team potentially owns one Microservice and there will be multiple Scrum teams based on the technological domain. By the time the Scrum team gets too big, it also serves as an indicator it is time to break the Microservice to be smaller. Ideally you do not want to wait for the Microservice to be too big before you break it. You should be alert on not to stuff your Microservice to be bloated in the first place. Kick start another Microservice whenever you can logically scope the context boundary into a separate Microservice.

Scaling – The fundamental of scaling boils down to 2 approaches: Vertical and Horizontal. Vertical scaling is quick and easy but could get very expensive especially when hitting the top tiers of resources. Horizontal scaling is cheaper but could be difficult to implement if the solution is not designed to scale horizontally. For example, a stateful monolithic system. As a general rule of thumb, always design your solution to scale horizontally. To put this into perspective, one large virtual machine could be substantially more costly than three small virtual machines that provide the equal amount of processing power, depending on which cloud provider you are working with.

Building solution as Microservices provides the foundation to scale horizontally. Using the earlier Photo Album Printing system, say there are many users who submit photos in bulk for processing during 9.00 AM to 12.00 PM. The DevOps guy only need to scale up the Java image processing engine service.

9.00 AM-12.00 PM

Scaling for 9.00 AM-12.00 PM

At 6.00PM to 11.00PM, say there are many visitors come to the website to browse the photos. The DevOps guy only need to scale up the PHP frontend application.


Scaling for 6.00PM-11.00PM

If we have a gigantic monolithic system, we have to scale the entire system regardless of which component is being utilized most. To put this into perspective, imagine you keep your car engine running just because you want the air-cond. Heads will not roll, it is just not the most efficient way to use your technologies.

Ease of Deployment – If you have tried waking up 2.00 AM in the morning for a “major deployment”, or been through a 20-hour deployment, you probably will agree it is important to have clean and quick deployment. I can vividly remember how nervous my CIO got whenever we have a “major deployment”. Sometimes he will come in early morning together with us give us moral support by supplying us with coffee and McDonalds. Despite the heart-warming breakfast, it was really stressful for everyone go through such deployment. Long story short, we have improved our deployment to be able making 4 productions deployment within a day with 0 down time. It is not possible (or significantly more difficult) if we have not built our codes on Microservices architecture.

Deployment for Microservices is definitely easier compared to a gigantic monolithic system. The database that the Microservice using is simpler which make altering database schema changes less painful. The amount of code is lesser which indirectly means there are less configuration to deal with during deployment. The scope of what the Microservices is designed for is smaller which makes post-deployment (both automated or manual) testing faster. In worse case scenario, rolling back a small service is significantly straightforward compared to rolling back a monolithic system with 25 other dependencies where some of they need to be rolled back together.

Scaling Microservices

The secret to scaling Microservices is: start small, think big. You might start your Microservice as a small service coded by a solo developer in 2 weeks. Although a service could be small, you need to think about how to deal with it when your audience size grow 10 times larger. As we discussed earlier, scaling vertically is easy but could be fairly expensive when you get nearer to the top tiers. You want to design your Microservice to scale horizontally from day one.

How to build a horizontal-scale friendly Microservice? The most common reason some services cannot scale horizontally efficiently is due to session. When you have a session stuck in your service memory, your client will always have to go back to the same service, else you will discover all kind of weird behavior. Of course, you can overcome this problem by enabling stickiness in your load balancer, or have an additional SQL Server database to keep all the session (InProc mode). However why would you want to get yourselves into this situation in the first place? If having session within the service is going make your scaling effort more challenging, avoid relying on session from day one so that you can scale horizontally, effortlessly. Building your service base on RESTful principles is a good starting point.

If your Microservice really have to make use of session, make use of additional session service such as Redis instead of keeping your session in service memory.

Front your service with a load balancer. Having your service instances sit behind a load balancer acts as the foundation for horizontal scaling. Configure auto-scaling in whichever cloud provider you are using. By the time the additional load kicks in, your auto-scaling will automatically boot up additional hosts to serve the load.


Another advantage of having your Microservice instances sit behind a load balancer is to avoid single point of failure. You should consider having at least 2 hosts to avoid single point of failure. For example, perhaps your Microservice only need 1 medium size virtual machine computing power. While 2 small size virtual machines provide the equal computing power as 1 medium size virtual machine (you have to work out the Maths yourselves). Having 2 small size virtual machines sit behind a load balancer rather than 1 medium size virtual machine being connected directly is a good remedy to avoid single point of failure.

Scaling Databases

As the number of your Microservice instance grows, it usually means more load in your database. Database IO is the most common bottleneck in software performance. Unless you have put in explicit design to protect your database from getting hit, you will often end up with scaling your database vertically. However you might want to keep this option as your last card.

Out of the box, SQL Server provides you the option to do Transactional Replication, Merge Replication, and Snapshot Replication. You have to determine which mode is most optimum for your system. If you have no idea what all these are about, Transactional Replication is your safest bet. But if you are adventurous, you can mix and match different approaches.

Transactional Replication works by Publisher and Subscriber model. All Write will happen in Publisher while all Read will happen in Subscribers. In typical services, the number of Read far out number Write. In this set up, you distribute the Read load to multiple Subscriber hosts. You can continue to add more Subscriber hosts as you see fit.


The drawback is, you need to code your Microservice in a way to perform all data manipulation and insertion in the Publisher while the reading in the Subscriber(s), which requires conscious effort from developers to code in such manner. Another drawback for adopting replication is the skill set required to set up and maintain the replication. Replaying transactional log is pretty fragile from my experience. You need someone who understand the mechanism behind replication to troubleshoot the failure effectively.

I highly recommend you to give some forward thinking on how to avoid your database get hit unnecessary in the first place. Tap into search engines such as Solr and ElasticSearch when suitable. Identify how your data will be used up front. Keep the on-the-fly data aggregation to minimum. Design your search indexes accordingly. At the very minimum, make use of caching.

The key for scaling your data is to achieve eventual consistency. It is alright to have your data out of sync for a short period of time especially on non-mission critical system. As long as your data will be consistent eventually, you are heading to the right direction.

Scaling database could be tricky. If you need something to be done by tomorrow 9.00AM, the easiest option is to scale vertically. No code change and no SQL Server expert involved. Everyone will be happy… probably except the CFO.

Keep Failure in Mind

Failure is inevitable in software. In Microservices architecture, the chances for software to fail is even greater. The risk of failure is exponentially higher because your service no longer depend on yourselves alone. Every Microservice that your Microservice depends on could go wrong at any given time. Your Microservice need to expect other Mircroservices to fail and handle the failure gracefully.

Bake in your failure handling. Your Microservice depends on other Microservices at one point or another. Can your Microservice core features operate as usual when other Microservices start failing? For example, a CMS depends on a Comment Service. In an article page, if the Comment Service is not responding, how will that affect CMS ability to display the article? Will the CMS article page just crash when the visitor visits? Or will the CMS be able to handle the Comment Service failure gracefully by not showing the comments but the rest of the article is loaded as usual?

Another example, I was using Redis to keep my user token after every successful login. At one point, Redis decided not to keep token for me anymore by actively rejecting the new connection. My users could not login although they have entered the correct username and password. The users could not login simply because part of the non-critical authentication process has failed. We discovered the root cause later. However, in order to avoid such embarrassing moment from happening again, at code level we changed the interaction with Redis to an asynchronous call because creating a token in Redis is not the main criteria in authentication. By changing the Redis call to asynchronous, users can continue to utilize the core functionalities although minor portion of features that relies on Redis token will not work.

It is fine if you do not have a sophisticated failure handling mechanism. At the very minimum, code defensively. Expect failure to happen at every interaction point with other Microservices. Ideally we do not want any of the Microservices to fail. But when they do fail (and they will), defensive coding help your Microservice being minimally affected. It is better to have 70% of your core functionality working working than the whole service crashing down.

The Backends for Frontends

This is another concept I discovered by accident when I was hacking some codes in Android Studio in 2013. The Backends for Frontends design is extra practical for mobile application, although you can still apply the concept on any Microservices.

In essence, the Backends for Frontends design is to back your frontend application with a backend service. The primary objective of this backend service is to serve your frontend application. This is a very good choice for mobile application for several reasons.


First, mobile application is known for having connectivity limitation. Instead of asking your mobile app to connect to 7 different other Microservices to request various information and do the processing at the client (mobile) side, it makes more sense to get the Backend for Frontend service to make the necessary server-to-server calls, process data, then only send the necessary data back to mobile client.

Second, the Backend for Frontend service also serve as a security gateway. Obviously you do not want to expose all your backend core services (for example your CRM) to the public. You need to design your network to have your backend core services sit in a private network. Then, grant permission for your public facing Backend for Frontend service to access to this private network. By doing this, your backend core services are protected from public access yet there are explicit permission granted to specific Backend for Frontend service. You can implement whichever security model you find fit in the Backend for Frontend service where your client application must and can comply to.


Third, mobile application sits at client side, which makes updating the application more challenging. You want to minimize the logic in the client side. The Backend for Frontend service plays the perfect role for handling business logic. You can update the logic much easier in the Backend for Frontend service compared to the client application. In other words, your frontend application will be lightweight and is only responsible for UI presentation.

One Last Thought…

Microservices is a huge topic by itself. This article serves as a triggering point for you to get to know Microservices without going through at 400 pages book. If you would like to learn more, there are many books available. I recommend you to look at Building Microservices by Sam Newman. I hope you have discovered something new in this article. Until next time!

Json Data in SQL Server


The rise of NoSQL database such as mongoDb is largely due to the agility to store data in non-structured format. A fixed schema is not required like traditional relational databases such as SQL Server.

However, NoSQL database such as mongoDb is not a full-fledged database system. It is designed for very specific use cases. If you don’t know why you need to use NoSQL in your system, chances are you don’t need to. For those who find it essential to use a NoSQL database, often they only use NoSQL database for certain portion of their system and then use another RDBMS for the remaining part of their system that have more traditional business use cases.

Wouldn’t it be nice if RDBMS is able to support similar data structure – having the ability to store flexible data format without altering database tables?

Yes, it is possible. For years, software developers have been storing various JSON data in one table column. Then, developers will make use of library such as Newtonsoft.Json within the application (data access layer) to deserialize the data to make sense out of the JSON data.

Reading / Deserializing JSON

string json = @"{
  'Name': 'Bad Boys',
  'ReleaseDate': '2015-4-7T00:00:00',
  'Genres': [

Movie m = JsonConvert.DeserializeObject<Movie>(json);

string name = m.Name;
// Output: Bad Boys

This works. However “JsonConvert.DeserializeObject” method is working extremely hard to deserialize the whole JSON data to only retrieve a simple field such as Name.

Imagine there is a requirement for searching certain Genres on a table that has 1 million row of records, the application codes will have to read 1 million row of records, then perform filtering on the application side. Bad for performance. Now imagine if you have a more complex data structure than the example above…

The searching mechanism will be much efficient if developers can pass a query (SQL statement) for database to handle the filtering. Unfortunately SQL Server does not support querying JSON data out of the box.

It is impossible to directly query JSON data in SQL Server until the introduction of a library known as JSON SelectJSON Select allows you to write SQL statement to query JSON data directly from SQL Server.

How JSON Select Works

First you need to download an installer from their website. When you run the installer, you need to specify the database you wish to install this library at:


What this installer essentially does is to create 10 functions in the database you have targeted. You can see the functions at:

SSMS > Databases > [YourTargetedDatabase] > Programmability > Functions > Scalar-valued Functions


Next, you can start pumping in some JSON data in your table to test it out.

I create a Student table with the following structure for my experiment:


In my StudentData column, I enter multiple rows of records in the following structure:

     "firstName": "Daniel",
     "lastName": "Foo",
     "age": 29,
         "streetAddress": "57 Jalan Senjolong",
         "city": "Cheras",
         "state": "Kuala Lumpur",
         "postalCode": "56100"
           "type": "home",
           "number": "+603-92848831"
           "type": "mobile",
           "number": "+6012-6461560"

For demonstrating the query purpose, I have entered multiple rows as following:


If you want to write a simple statement to read the list of student names in JSON data, you can simply write:

    dbo.JsonNVarChar450([StudentData], 'firstName') [First Name],
    dbo.JsonNVarChar450([StudentData], 'lastName') [Last Name]

You will get result as following in SSMS:


How about more complex query? Does it work with Aggregate Functions?

If you want to find out about how many students come from each city and what is their average age, you can write your SQL Statement as following:

  dbo.JsonNVarChar450([StudentData], '') City,
  AVG(dbo.JsonInt([StudentData], 'age')) [Average Age],
  COUNT(*) [Count]
  dbo.JsonNVarChar450([StudentData], '')

You will get result as following in SSMS:


It appears the library allows you to query any JSON data in your table column using normal T-SQL syntax. The only difference is you need to make use of the predefined scalar-valued functions to wrap around the values you want to retrieve.

Few Last Thoughts…

  1. The good about this library is it allows developers to have hybrid version of storage (NoSQL & relational database) under one roof – minus the deserialize code at application layer. Developer can continue using the classical RDBMS for typical business use cases and leverage on the functions provided in the library to deal with JSON data.
  2. The bad about this library is it lacks proven track record and commercial use cases to demonstrate the robustness and stability.
  3. Although the library is not free, the license cost is relatively affordable at $AU 50. However the library is free for evaluation.
  4. SQL Server 2016 provides native support for JSON data. This library is only useful for SQL Server 2005 to 2014 where upgrading to 2016 is not a feasible option.

Multi-Country Address Database Design


Database design is one of the most critical elements in software development. Realizing that

  1. New programming design patterns and frameworks are introduced from time to time, codes get refactored and system get revamped;
  2. New UI frameworks are introduced from time to time, presentation layer get thrown away and redesigned;
  3. However database design stays the same for as long as possible because business simply do not want to risk valuable production data and development team do not want to deal with legacy codes to accomodate the new design.

A good database design can stand against the agility of business needs. The development team is not expected to frequenty adding or modifying database (for example adding new tables and altering relationship between tables) unless it is “totally necessary”.

While “totally necessary” could be subjective depending on how well a database is designed to handle the changes, a good database design is expected to handle various changes in business requirements without major redesign of the existing database.

Case Study: Handling addresses in multi-country database

I was working for a car classified MNC that have businesses in 3 countries. We were rolling out CRM for Malaysia, then Indonesia and finally Thailand. Each country has a different address format.







Following Agile principle of Just Enough and Just In Time (and a typical product manager would choose to deal with the other countries in a later time rather than taking the overhead now), development team would design a database schema to cater only for Malaysia user address format.

The following database schema design is efficient to get the job done. Address is the child of Customer. Section, District, Postcode, Locality, Province, and Country are the referencing tables for Address table. Among different location levels, the parent-child relationships have been defined to indicate the hierachy of the location levels. The database is fully normalized and has strong data integrity by defining appropriate foreign keys:


The development team proceed with loading data to the respective reference tables (Section, District, Locality, Postcode, Province, Country) and launched the CRM for Malaysia users. Soon production data start flooding into Customer table and Address table. Everyone is happy with the system.

2 months later business informed the development team, “Guys, it’s time to let our Indonesia users enjoy our amazing system!”, “Sure!” comes the reply.

When development team looked at Indonesia address format, it is quiet different from Malaysia address format. Leaving AddressLine1, AddressLine2 and CountryId field aside, the remaining fields are different.


The existing database schema cannot fit Indonesia data in any elegant manner due to the parent-child relationship between location level and obviously it is not a good idea to have a Postcode table filled with Malaysia Postcode data, Indonesia Locality data and Thailand Province data (refers to row #3).

Not only Indonesia, even Thailand has a different format compare to Malaysia and Indonesia. The mapping will only get more complicated when more countries join the party.

In order to support Indonesia, the existing database schema need to go through moderate design change to cater for Indonesia address format. Later when system need to support Thailand address, another database schema change would take place.

Instead of continue maintaning a structured table design to map one field in a table to one business entity property, while making changes to support Indonesia (or even better when designing database schema in the first place), a hierarchical relationship could be defined for maintaining the dynamic relationship of address location levels.

Simply put, a hierarchical relationship is self referencing mechanism to dynamically define n-level of parent-child relationship.

The most common usage for hierarchical relationship is defining Employee and Manager relationship. A database could have an Employee table. Within the Employee table there is a ManagerId field and the ManagerId field has a foreign key to the same Employee table Id field. The record could keep referencing within the table itself until hitting a record where his ManagerId is null, which indicate that record is the highest level in the hierarchy. In this case, most likely the record is the CEO record.

To put it into context of Address, take a look at the following database schema design:


In this design:

  1. Tables are much lesser and relationship is much cleaner
  2. Development team can be assured that this design could handle any address structure in any other countries.

The above database design schema shows:

  1. There is a Customer table.
  2. There is an Address table that could have multiple records for a particular Customer (one to many relationship). The Address table has a foreign key to Location table to tell which is the smallest level of location the Customer belong to.
  3. There is a Location table that has a hierachical relationship with itself. Base on the ParentId field, you can “figure out” the full set of address.

So how do we “figure out” the hierachy?

Let’s take a look at a sample set of data in Location table:


By using the following SQL query with Id = 27:

WITH LocationTree
AS ( SELECT Name, ParentId
FROM Location
WHERE Id = 27
SELECT Location.Name, Location.ParentId
FROM Location
INNER JOIN LocationTree ON Location.Id = LocationTree.ParentId)
FROM LocationTree
You would get result as following:
This result shows the child parent relationship between the location. Cheras is the parent of 56100 (postcode), Selangor is the parent of Cheras and Malaysia is the parent of Selangor. In address context, Postcode 56100 is inside Cheras, Cheras is inside Selangor and Selangor is inside Malaysia.
The Id = 27 is set in Address table LocationId field to represent the lowest level of location the application could identify for the particular user.
Now if we modify the query to Id = 28, you would get a different set of location hierachy:
Not only the hierachical relationship would work for single country, but also other countries with lesser location level. Assuming that we have a customer from Indonesia and the smallest location level we could identify is City, say Id = 22, where the Id = 22 is set in the Address record belongs the Indonesia user. By running the same query, we would get the following:
The mechanism for hierachical relationship is simple in the address context. Howeer at the application level, the code need to set the smallest location level into LocationId in Address table. At the application level the input would need to be in the form of selection rather than free text input (else it would defeat the normalized database implementation). For example:
Regardless of how many location level we have, or what is the sequence of the location level, a hierachical relationship will allow us to have a normalized yet flexible database schema that is resilient against business changes like supporting more countries.

An application developer that handles the programming codes for multiple countries might complain he could not map the list of address result hierachy into his object property. Imagine he has 4 levels for Malaysia, 3 levels for Indonesia and unknown level for Thailand. How would he design his codes to map what does each level means or represent for each country without having 300 lines of if-else statement? A trick for handling such complexity is through dependancy injection per country. After all, it is the application code job to give context to the data while the database job is to provide storage for the data.

One last thought…

Having a hierachical relationship would increase the flexibility of database schema, but it increases complexity in application codes and would not provide a clear one to one mapping on table field to object property. Such implementation is fundamentally different from how a typical development team would design the Address table. Such complexity is not required if a development team is designing a system to support only 1 country address format. However hierachical relationship should be given consideration if the database needs to handle multiple-country address with various combination of location level relationship.