Microsoft Azure Cosmos DB Reviews

Azure Cosmos DB is our database of choice for new applications and cloud-native applications. I use it anywhere.

Because it is NoSQL, it has the capability to adapt to changes. As compared to Azure SQL or other SQL databases, Azure Cosmos DB is schema-less. We can add new columns anytime, and the application will not break. It is very efficient for application-facing scenarios.

The most valuable feature of Azure Cosmos DB is its scalability. That is the biggest reason I use Azure Cosmos DB.

I also like its developer-friendliness. It is very easy to begin with. Microsoft and Azure are good with that. With all the getting started information and all the introductions, it is very easy to begin with. Optimization is where it gets a bit trickier. That is where you need to be more active and understand why things are not performing as they used to. Most of the time, performance is not a problem. It is always fast. The problem is more around the cost consequence of that performance.

Its vector capabilities are new. They were implemented just months ago. There are probably three things that we were looking to address by using the vector database. The number one is the cost of Azure AI search and indexing. Before this feature came out in Azure Cosmos DB, the alternative for me was using AI search, which is way more expensive if using it as a vector database. Now with Azure Cosmos DB, that price point becomes much more accessible. That is number one. 

Number two is the developer familiarity aspect of things. AI search is more around enterprise use cases or enterprise search and requires more specialized skills to begin with, whereas Azure Cosmos DB is more or less a commoditized developer platform that is much more accessible to a wider developer audience. That is another aspect that it has addressed. For example, if I have a new starter in my team, it is easier to train that person on Azure Cosmos DB than the AI search. With the explosion of LLMs, AI agents, and things like that, the vector database on Azure Cosmos DB is a good place for developer onboarding. It is just way easier. 

The third part is still related to the developer experience, but in terms of the SDK aspect, the libraries available for Azure Cosmos DB are already well-established in the ecosystem. With the vector database capability, it is just a matter of adding an extension of those existing libraries. That means if the applications that are already using Azure Cosmos DB want to jump to the vector database, the jump would not be that big. It is just a matter of implementing it directly with their existing Azure Cosmos DB that they are already using.

We have used the vector database with Azure AI services. The other aspect is using the vector database with document intelligence. We use document intelligence to process a raw PDF document and things like that. From there, we convert that into embeddings, and then those embeddings will be stored in the vector database. It is something we use as a landing spot for new LLM applications.

The quality has improved because, traditionally, we did things in batches. We processed documents once a day or every twelve hours or so. With this new capability, we are very confident to run those processes in real-time. As new documents come in, the process and the workflow can get triggered. It is not a batch anymore. It is in real-time.

Azure Cosmos DB’s ability to search through large amounts of data is yet to be determined. Large is subjective at the moment. I have only tested it up to 2 gigabytes, and for that, it is working pretty well.

Azure Cosmos DB is our default. We do not question it anymore. After migrating out applications from an SQL database to Azure Cosmos DB, the change in the organization is massive. Especially on the application side of things, app developers are much more productive and lean. Previously, we had to go through a very rigorous process. To add new columns, tables, and other things, we had to work with DBAs. With Azure Cosmos DB, we can have a PoC and POV in weeks, sometimes days, instead of six months. That is how the whole NoSQL ecosystem changed our life cycle and productivity.

Azure Cosmos DB has changed our total cost of ownership for old applications, but not for new applications. For those who are still using SQL Servers or other databases, there was an added TCO because different projects are using different databases, whereas about 10 or 15 years ago, we had just Oracle, SQL Server, or IBM. For new applications, it is the default for us, so there is no change in TCO.

There are several areas for improvement. Firstly, having a local development emulator or simulator for Azure Cosmos DB would be beneficial. It would be very handy to have a Docker container that developers can use locally. Although, I know there is a free tier and so on and so forth, having a local environment would be nice. For example, SQL Server is very portable. You can even install it on your machine. That is the number one thing that is missing in Azure Cosmos DB.

The second improvement area is the IDE of choice. That means how you interact with Azure Cosmos DB. For example, with SQL Server, you have SQL Server Management Studio. I know there is a little bit of support for Azure Cosmos DB in Azure Data Studio, but it is not heavily advertised or it does not feel like first-class citizen support. Developer experience or developer tooling is missing in terms of interacting with the database. Better developer tools or an IDE for interacting with Azure Cosmos DB would enhance the developer experience.

Lastly, there is some mixed messaging about what Azure Cosmos DB is, given its multiple APIs. There are so many Azure Cosmos DB APIs available. There is NoSQL. There are MongoDB, Gremlin, and others. There is still some mixed messaging for others who are new to Azure Cosmos DB about what Azure Cosmos DB is. Is this like MongoDB, but then there is also MongoDB in Azure Cosmos DB? I know it well, and I know that the default one is just NoSQL, but others I have interacted with over the last ten years or so get confused.

I have been using Azure Cosmos DB for over a decade. I have been using it since it was announced.

The solution is very stable, and I cannot recall a time when Azure Cosmos DB let us down. I would rate it a ten out of ten for stability. I never had issues with it.

Its scalability deserves a ten out of ten. I have never hit a limit with Azure Cosmos DB.

We have multiple locations and multiple departments. We are in different countries and regions. For our one project, we have multiple Azure Cosmos DBs. We have about seven developers, and we have tens of thousands of users or consumers. Our clients are enterprises and SMCs.

Early on, about a decade ago, when I started with Azure Cosmos DB, I just played with it and created many things. I ended up having a $10,000 bill. Because it was an accident, I had to send a support ticket. The support team was able to waive that cost for me. That left a very good impression on me up until today. I did not have to pay that money, especially when I was just starting. Now, there are very good partners out there, us included, who are well familiar with Azure Cosmos DB. That ecosystem is well supported now. It is not like you are going to a niche database and hoping for the best. That ecosystem is quite mature.

I would rate customer service and support a nine out of ten. The only reason why it is not a ten is because a lot more triaging is required when raising a support ticket. That is the problem I have.

Positive

Before using Azure Cosmos DB, we primarily used MongoDB and Postgres. I have a mixed experience with both of them. There are also Azure flavors of those. You have MongoDB Atlas on Azure and you have Postgres on Azure. That is why sometimes I am very conflicted about which one to use. Both MongoDB and Postgres have captured the audience around the open-source community and the non-Microsoft enterprise or developer ecosystem.

The deployment is a one-off. It is straightforward. For provisioning Azure Cosmos DB, everything is there. It has been straightforward so far.

Its deployment is done in minutes. In terms of maintenance, Azure Cosmos DB itself does not have any maintenance. However, the application that we are supporting and developing needs maintenance. That is it. Azure Cosmos DB does not require migrations like SQL Server where you have to manage a migration from version 17 to 19 and so on.

We achieved a strong return on investment. Using Azure Cosmos DB enabled us to bring a project to the MVP stage in six weeks. With one recent application that we had, if we had gone through another approach, the project could have taken six months in an enterprise setting where everything is slow and challenging. Getting an MVP of that project would have taken six to eight months, but because we had an active choice of using Azure Cosmos DB and other related cloud-native services of Azure, we were able to get to an MVP stage in a matter of weeks, which is six weeks. That was a very measurable impact that we had. If we went another route, just defining the tables, entities, and other things would have taken us a big amount of time. We had already identified base entities. We knew we could add more columns or remove some columns as we went along. That gave the agility to our project.

We do not have to look at it periodically. I do not get support calls that our application is down.

From a startup point of view, it appears to be expensive. If I were to create my startup, it would not have the pricing appeal compared to the competition, such as Supabase. All those other databases are well-advertised by communities. I know there is a free tier with Azure Cosmos DB. It is just not well advertised.

For mid-tier customers, its pricing is justifiable. The enterprise tier is where it is subjective. For organizations that have built a lot of capabilities around SQL Server, Oracle, or so forth, because of the lack of skills, understanding, and capabilities around Azure Cosmos DB, it would appear to be expensive. The professional services aspect of Azure Cosmos DB is what is driving the cost, not the platform itself. The skills required to manage the service can drive up costs more than the platform itself.

I would recommend Azure Cosmos DB for its scalability and performance. Do not be frightened to give it a try. Because there is no local way of doing things, Azure Cosmos DB will always be considered expensive. It is not very developer-friendly when you have to pay upfront, but there is a free tier. Microsoft needs to do better in terms of communicating that it is free to get started.

I would rate Azure Cosmos DB an eight out of ten because of the lack of local development and so on. It also gets confusing with so many APIs. There is a mixed messaging problem around that. The vector database and so on are also confusing. There is a vector database, but depending on which API you choose, there is a different implementation. It is just a bit confusing. I use this every day, so I know it by heart. I know where it is going, but it is just not very easy to get started for others. Messaging and product categorization are not clear. The way they are bundled or packaged is confusing.

We use it as our main database for our network access control software, and we use it to store all of the information we need to authenticate different devices and users to the networks of our customers. It maintains all of the necessary data for our SaaS product.

It is pretty easy to maintain and to optimize. The main thing that we have to deal with is the RUs. Probably the number one topic of Azure Cosmos DB in the world is how to make sure you have the right RUs set in place for each of your different collections, but the tools that are available in the Azure portal make it very easy for us to check how the performance is going. We can check if we need to adjust anything within the system to ensure that we have the right scale and the right split of our different indexes so that we are getting the most throughput for the data we need. In general, it is very easy for us to maintain. We just need to use the Azure tools to let us know when we need to pay attention to the different throughput variables that are important in the general maintenance of the product.

We use Azure AI Search for all of our search needs. The integration that Azure AI Search has with Azure Cosmos DB is very good. We heavily utilize the integration with Azure AI Search for a lot of the features on our website. The search works very well. I do not know if we have what one would consider large amounts of data because each one of our customers is searching through just their own set of data. On average, each customer has data in the range of only gigabytes, so the search does very well, but I do not know if you would consider what we have as very large amounts of datasets. We are dealing in the few gigabytes range rather than anything huge like terabytes.

The benefits of its global availability and the response time were noticed right away because we immediately had customers who were globally distributed, so the latency was noticed right away. We got a good performance that way. We did not notice any of the other benefits until we got a lot more customers and had a lot bigger scale. As our scale has increased dramatically over the past couple of years, we have noticed that we have not had to do much with Azure Cosmos DB. It just takes all of our additional data and additional queries and all of the additional throughput that we are throwing at it. It does not need very much in terms of maintenance and performance tuning because it handles everything we need pretty much out of the box, so it is a low-maintenance solution for us where we just check on things every once in a while as a best practice. In terms of scalability, we have doubled, tripled, and quadrupled our customer size or number of customers, and we have not had to do very much with it architecturally just because it has been able to handle that scale.

For onboarding, the documentation is very good. As soon as I joined the company, I read a lot of the Azure Cosmos DB documentation so that I understood it. It is well documented, and there are support forums and Microsoft experts. We have a Microsoft Solution Architect dedicated to us, and we have been able to ask him questions. The community surrounding it and Microsoft's ability to answer all of our questions, my questions specifically, have been really good. The documentation is great. I have been able to find all of my answers during my tenure. Everything is generally answered in the documentation, and for what is not, Microsoft has been very quickly able to get to us through our Solution Architect.

Within the first week, I was already executing queries against the database and monitoring its performance. Within a week, I had a basic understanding of how to interact with the database and understand different performance metrics and structures within the database. It is pretty easy to learn if you are familiar with other databases. Because I was familiar with other document stores and SQL-based databases, there was not a lot to understand. There are some differences in the SQL language that you are allowed to use with Azure Cosmos DB, so there was a little bit of a learning curve there, but the documentation was really specific. It is not a sharp learning curve if you are familiar with any other database systems. If you are familiar with SQL Server, MongoDB, MySQL, or Postgres, a lot of the concepts are exactly the same.

We chose Azure Cosmos DB initially because of the type of data that we needed to store. We have a schema that is very nondeterministic and flexible. It is always changing based on whatever data we need to acquire from different devices, so we needed a document store with a flexible schema. 

In addition to that, our customers are globally located, so we needed a database store that could be globally accessed and had minimal latency, good throughput performance, good query performance, as well as scalability. All of the things that you look for in a good piece of software about performance, scalability, high availability, and disaster recovery are available in Azure Cosmos DB. Because of that and because it is a flexible document storage, we went with Azure Cosmos DB.

The one thing that I have been working on with Microsoft about this is the ability to easily split partitions and have it do high-performance cross-partition queries. That is the only place where either our data size or our throughput has grown beyond one partition, so being able to do cross-partition queries efficiently would be my number one request.

The request unit architecture that they have in place is understandable but could be better. What you get out of some solutions like SQL Server or MySQL is a lot more understandable. The request unit architecture of Azure Cosmos DB is not as easy as pure SQL solutions. They could do better in making the RUs more understandable and more flexible because changing your partition keys and your indexes is a larger batch of work than it necessarily needs to be.

The company I have been working for has been using it for more than five years, but I have been with the company for just over two years. I have been working with Azure Cosmos DB during my entire two-plus years at the company.

We have had only one incident where Azure Cosmos DB went down. It was about two years ago in the East US. They had an incident where the update they made caused some downtime for us. I forgot what the duration was, but luckily, we had global replication for our main Azure Cosmos DB setup, so while the East US was down, the West EU region picked up. The majority of our operations continued without issue because of our use of the global replication option available within Azure Cosmos DB.

Latency and availability are great. You can, of course, write code that does bad things for it, and we have had to fix our own code sometimes. Whenever we have written our code properly, latency and availability have been great.

The scale has been wonderful. Our ability to add request units as we have needed them has been easy. We do not have to do much other than tell the system we want more, and then it automatically scales for us, so we do well there. 

The only limitation is around the partitions. Each physical partition maxes out at 10,000 request units as they have documented. We have had to deal with that while designing our data structures to make sure that we take into account that physical partition limitation.

The quality is top-notch. We have been able to talk directly to some of the Azure Cosmos DB experts at Microsoft. We have been able to get extremely detailed answers with very specific recommendations for all of our different questions.

Their speed has been definitely acceptable. Within a day or two, I get at least an acknowledgment of my question. We have not had any high-severity questions to be answered right away. Most of our questions are during the design phase where we just need to know specific recommendations based on our needs, so there has been no real-time pressure. Answering or acknowledging our question within a day or two has been an acceptable time frame. We generally get our answers within a week, which has also been acceptable for us. We have never needed a super fast answer, but I am also clear about that in my communication with them. I tell them that it is not an emergency, and we are just in the design phase and need these questions answered.

I would rate their support a ten out of ten. There is nothing that I would ask for more from the support. They are responsive. They give accurate answers, and they are easy to deal with. That is all that you want from a support experience.

Positive

Historically, in my career, I have used SQL Server, ClickHouse, Postgres, and MySQL. ClickHouse is probably the closest equivalent, but we had to maintain everything in-house, so it was a lot more intensive to maintain. SQL Server was nice, but it does not have the document flexibility, or it did not have that at the point in time we were thinking of using Azure Cosmos DB. So, I am very familiar with all the SQL-based servers historically. They just did not have the necessary document flexibility that we were looking for when we selected Azure Cosmos DB.

It was already in place when I joined.

It does require some maintenance as we grow. Each one of the collections scales up by request units. We use the autoscale feature, but it has a bounded range, so as we grow, we watch the RU maximum, and we adjust the RU maximum of our different containers as we scale up. The maintenance is minimal. We change the RU scale maybe once per quarter. Otherwise, everything goes fairly without maintenance.

I do not know if it has helped us decrease the total cost of ownership, but certainly, it has helped in our DevOps maintenance. Our DevOps people spend a lot less time worrying about or dealing with Azure Cosmos DB. Because of that, we do not spend a lot of person-hours on Azure Cosmos DB. However, Azure Cosmos DB does come at a premium price in terms of being able to do all of its features. It has an appropriately associated cost with it. Because of that, we have not done any formal calculations to see if it saves us more than some of the other solutions such as SQL Server or Postgres. We have not done a cost comparison. What we do know is that it satisfies all of our needs. We do not spend a lot of time thinking about it. As we add features and datasets to it, we do not have to do a lot of performance testing, so we are just able to add things to it. It just works, and we do not have to spend a lot of development time, QA time, or DevOps time worrying about whether it is going to be able to satisfy what we need it to do.

As opposed to running our own VMs or our own databases, it would have reduced our overhead costs. All we have to do is go into the Azure portal, click a couple of buttons, and type a couple of numbers, and then it just happens without any other effort. It takes us seconds to minutes to change things, whereas other solutions might take days or hours to process. From that perspective, there is certainly a reduction because it only takes us a few seconds to scale our Azure Cosmos DB without any other effort. However, you pay for that with the actual price of Azure Cosmos DB itself. That is somewhat built into the price where Microsoft takes on that maintenance cost, but you pay for it.

The pricing and licensing model was initially difficult to understand, but as soon as I learned what was going on and how it was priced, it was pretty easy. What is more difficult is to understand how your system is going to behave specifically with the specific partitioning and querying that you are doing. Some of it is reactive because you cannot always predict what your customers are going to use in your product and in what specific way. So, while we have understood the pricing model, what we have not understood is which parts of our system would end up being the most expensive, costing us the most, or needing to scale the most. It is not necessarily an issue with Azure Cosmos DB itself. It is about understanding your individual software or our individual software when it is running on top of Azure Cosmos DB. It is about understanding what the behavior is going to be.

I would advise taking advantage of all of the features that are available. Especially if you are a globally distributed business, make sure that you have all of the high availability and backup options enabled so that you are not surprised in case of a problem.

Like almost all of the recommendations that you see in different Microsoft videos, make sure that your partition keys are set up properly from a RU perspective so that you know that you will be able to scale your individual containers effectively without running into the limitation of 20-gigabyte physical partition size or 10,000 RU physical partition throughput. Be aware that those exist and design your partition keys for the future so that you will not be limited when your system starts to get heavily utilized in the future.

I would rate Azure Cosmos DB an eight out of ten. There are some improvements that I would like to see around the physical partitions.

In our setup, we rely on Azure Cosmos DB primarily for cloud-native applications that demand global scalability. We use it for connecting web apps and implementing search functionalities.

Cosmos DB's low-latency data access has greatly improved our application performance. It is a game-changer, allowing us to move workloads from on-premises to the cloud, thanks to the reduced latency, and freeing us from the constraints of on-premises environments.

The most valuable features for our organization with Azure Cosmos DB are multi-master capability for applications, automatic failover ensuring high availability, scalability, support for multiple data models, and low-latency access. Additionally, the seamless integration with microservices running in containers adds another layer of efficiency to our operations.

In terms of improvement, slight enhancements in integration interfaces, expanded dashboard functionalities, and broader use-case support would be beneficial.

I would rate the scalability of Cosmos DB as a nine out of ten.

The technical support is quite good.

It is a relatively affordable solution.

For those considering Cosmos DB, my advice is to embrace its versatility. Cosmos DB can handle various data models like documents, wide columns, and graphs seamlessly. You can consolidate your needs into one database, Cosmos, eliminating the need for multiple databases. It simplifies management and offers a comprehensive solution for a wide range of use cases.

Overall, I would rate Microsoft Azure Cosmos DB as an eight out of ten.

This is an event-driven solution. Most oil and gas companies have folder source systems, where they cannot scale, but they still want to provide real-time data to their end consumers for various analytical use cases and AI/ML processing; this is where we input raw data into the Azure environment of this solution. Then, eventually, we built the API on top of Microsoft Azure Cosmos DB because it's highly scalable. The solution is a little bit expensive, but the businesses are ready to accept it. 

In terms of performance versus scalability of this solution, you don't need to worry as long as you have your initial numbers in place. This product works by using performance currency, which is the number of request units per second. Once the data is ingested, based on that, we can know how many users are going to access across the world in every day, hour, or minute. Once you have the ingestion versus consumer pattern identified, you can use this product to input all those numbers, like the volume of data for migration. 

The built-in integration of the solution is tight. It can be used in conjunction with Synapse, Microsoft has also created a Synapse link. In this solution, the OLTP workload will never affect the OLAP workload. Therefore, the solution does data replication asynchronously without affecting the OLTP source system. No specific pipeline is thus required, which is not easily found in other services. 

In the server, there are two ways in which you can provision a call, one is serverless, which has a pay-as-you-go model, and another is a dedicated provision throughout. So, irrespective of what you allocate and whether you use it or not, in this solution, the charges are accounted for the request unit per second. This is a big drawback of the solution.

This is an expensive solution and if you get the initial calculation wrong involving how much you are going to ingest, how many people are going to query and more, then you are going to receive a very large bill at the end of every month.

Additionally, on the serverless option, there is a limitation regarding the amount of data you can ingest; this doesn't allow you to upgrade beyond a point, and the limit cannot be utilized for many use cases. On the execution side, whatever you create as a container, that container cannot be used as a destination when using serverless mode. This is another key limitation of the solution. 

I have been using the solution for one and a half years. 

I had minor issues while using the solution, but they were actively solved, and eventually, a justification was also given. Ninety-five percent of the time I used this solution, there were no issues. Microsoft's service in the cloud market is still growing and so there are some feature limitations. 

The solution is highly scalable. We use the solution in our enterprise both internally and externally, including integration for clients. We created our solution end-to-end by considering different audiences, people who can directly onboard Azure but might not need Cosmos DB.

There are vendors and individuals who cannot directly consume data on the Azure environment and will have a dependency on data. However, we cannot expose the source data for its performance issues and limited scalability, so we deliver this data by using Microsoft Azure Cosmos DB. The parent company Reliance has multiple subsidiaries like Ajio, manufacturing supply chain, Oil and Gas, and more; we used to use the same API for all subsidiaries, which was built on Cosmos. 

Technical support was good. I would rate the customer support an eight out of ten. They were fast and responsive, but the support team runs from different locations within or outside India, so whoever is working during the particular shift will take care of the case initially and then some other individual will take over.

So, my team had to re-explain the same thing over a call or meeting. But it was only a few times, they were able to get all the information based on the previous conversations most of the time. 

Positive

Deployment of the solution was very easy. Once the initial numbers are in place based on request units, only the instance creation was a time-consuming process. The time was consumed due to the dependency on other teams like DevOps, who are responsible for provisioning. So, it was a one-time process, but migrating and running the same workload between different environments was not much of a hassle. 

It took less than a week to configure and install this solution. To complete the setup, it took five to six professionals from our team. One key solution architect, two people from the DevOps team, and two solution architects from Microsoft were needed for the deployment of this product.

Maintenance of the solution is very easy because the solution follows a Platform-as-a-Service type of model. There is actually no need for any downtime or a patch upgrade because it is taken care of by Microsoft. I never have to worry about downtime for this solution. They perfectly deliver on the key characteristics of the product. 

Our business need was to deliver or provide the source data without any latency issues, in less than five or ten minutes latency, to be precise. We had to provide the data to the end consumer without overwhelming our source. We got the business confidence in the initial three months of using Microsoft Azure Cosmos DB. 

The solution is a bit on the expensive side. 

We tried to compare this solution with MongoDB, which is open-source. But we choose this solution because Microsoft is the first implementation partner for us. 

I would rate the solution an eight out of ten. My advice to other people will be first to identify the purpose of availing the solution. There is also a product called Azure Data Explorer, which is a more extensive service used for similar use cases.

Also, in terms of Cosmos, the user should be clear about whether they will be able to use the serverless deployment model or whether they need the dedicated provisioned throughput Model. They should also first use the price calculator by inputting the numbers to decide if they need it. I would also advise you to get in touch with a Microsoft Specialist and walk through all the doubts. 

The solution has a very good service, but the user should be clear about how to start using the product. For the initial three months, we did a lot of trials to get the components and RUs right and check how the calculation is happening. However, after the trials, we were very clear about how we wanted to move forward with the solution to get the maximum ROI. 

We use Microsoft Azure Cosmos DB in our loyalty platform, which is based on our proprietary technology, Synapse LX. In loyalty, we need to enroll, score, and deliver rewards communications in near real-time. There are significant volume spikes in those activities, so our use case is to support the writing of information into our database. Cosmos DB is a no-SQL database that allows us to scale quickly and handle large volume spikes. It allows us to auto or manually scale in many different ways. It gives us much flexibility to handle that requirement and ensure we deliver the right customer experiences.

Cosmos DB is not difficult to use, but like anything, it requires careful planning and consideration of use cases. This is especially important when planning to implement it. From an optimization perspective, Microsoft has made significant efforts in the past 12 to 18 months to facilitate changes after initial implementation and optimize cost.

Cosmos DB provides excellent search result quality. Since implementing it, we have not encountered any issues with our searches.

After deploying Cosmos DB, we initially experienced some performance gains, followed by additional benefits that required a learning curve regarding tuning and configuration. As our understanding deepened, we were able to optimize it further.

In the last three months, Cosmos DB has helped reduce our total cost of ownership. Microsoft recently implemented a feature that allows us to achieve savings of up to 50 percent.

The most valuable aspect of Cosmos DB is its performance. It serves as the foundation for OpenAI's infrastructure, providing us with similar functionality. This not only prepares us for AI use cases but also efficiently supports our loyalty use cases. We can share information with our customers and deliver experiences without concern about performance.

For our Disaster Recovery plan, we currently use geo-replication. We'd like to avoid full DR replication if possible, as this would result in significant cost savings.

I have been using Microsoft Azure Cosmos DB for five years.

We have not had any stability issues with Cosmos DB.

Cosmos DB's scalability is excellent, which is the whole reason to use it for scalability and performance.

The dynamic scaling helps decrease our overhead costs.

The initial deployment was straightforward and consisted of two to three people.

Cosmos DB's pricing structure has significantly improved in recent months, both in terms of its pricing model and how charges are calculated. This has led to substantial cost savings for both us and our customers.

I rate Microsoft Azure Cosmos DB seven out of ten because of the disaster recovery requirements.

Cosmos DB presents a steep learning curve. I would rate it a five out of ten. The challenge lies not so much in understanding its concepts as in utilizing them effectively and efficiently.

It took us 12 to 18 months of focused attention to fully onboard our team. At that point, we began to understand. However, it wasn't until we went live and observed actual user activity that we truly grasped the whole picture. Testing is one thing, but experiencing real-world interactions provides invaluable insights and a deeper understanding.

Cosmos DB requires minimal maintenance, but monitoring its performance and optimizing it as needed is crucial.

Potential users should plan accordingly, as Cosmos DB is a NoSQL database that uses similar design principles. Consider the design and apply those principles beforehand to optimize performance from the start. Understanding your read-and-write ratio is crucial due to cost implications, so ensure you understand the balance between reading and writing to the database. All these factors matter as they can impact your costs, so consider them carefully. 

I like to describe it as a programmer's database. .NET developers, in particular, can design and work with the data easily because it's schema-free. Unlike traditional databases, which are considered rigid with their rules, developers really love Cosmos DB because of its schema-free nature and the freedom it offers.

Cosmos is widely used for web applications. You can also use it for inventory management and IoT solutions... there are a ton of different applications.

It's very easy to integrate Azure Cosmos DB with other Azure services. For example, generating a Power BI report from data in Cosmos is just a few clicks. It's also simple to stream IoT or sensor data into Cosmos.

When it comes to supporting IoT or real-time analytics, the main advantage is speed. Cosmos DB doesn't rely on traditional disks and database structures like a relational database. It uses JSON, which is similar to XML, and that makes it incredibly fast.

The way it was designed is most valuable for global distribution. Unlike old-school SQL Server that was intended for a single data center, Cosmos was built from the ground up for global availability. 

Features like geo-clustering and mirroring were not afterthoughts. If you have a database in Chicago, you can right-click and easily create a failover group in Japan. That works well for global companies with offices across continents; it minimizes latency.

Cosmos's multi-model support made databases more highly available. 

The downside is that Cosmos is new and fairly complex. There's a limited pool of talent who are really good at working with it.

Because of that, I've been approached by recruiters quite a bit; they see my Cosmos DB certification on LinkedIn. It's hard to find people to work on Cosmos projects. 

Sometimes, a really smart developer will design and build a Cosmos implementation and then move on, leaving the company struggling to find someone to work with it and maintain it.

Interestingly, if you need to restore a Cosmos DB database, you have to put in a ticket with Microsoft – they're the only ones who can do that.

I've worked with Cosmos on and off for about three years.

For Cosmos DB, their technical support is very good. They are the experts in that product.

Overall, the customer service and support are excellent. 

Positive

I did a couple days of training on DynamoDB, which is Amazon's comparable product to Cosmos DB.

They're actually quite similar, both being multi-model databases. Relational databases are good for structured data, but once you get into semi-structured and unstructured data, they just don't perform well. 

That's where DynamoDB and Cosmos DB excel – storing, indexing, and quickly working with that less-structured data.

The initial setup is simple and straightforward. You can set up a Cosmos DB in a day, even configuring things like availability zones around the world. 

The harder part is on the developer side – designing the collections (similar to tables) and how the data will flow in.

I've set up a few Cosmos DB instances, and it's about a half-hour tops.

One person can handle the deployment. I'd typically set it up alongside other Azure components like a VM. You choose your settings, networking details, etc., basically walk through a wizard, hit deploy, and it's up within half an hour.

There are some configuration options for database administration on the customer's side. You'll need to go in and enable things like automatic indexing with checkboxes.

With heavy use, like a large-scale IoT implementation, you could easily hit a quarter of a million dollars a month in Azure charges if Cosmos DB is a big part of it.

I would recommend using it, but with a caveat – it's a good fit for companies with deep pockets.

It's powerful and amazing, but the costs can add up.

I'd give it an eight out of ten. It's super powerful and solves real problems with global distribution. I hesitate to give it a perfect ten because it's still new, and good training resources are harder to find. Even the most recent books on Cosmos DB are several years old, which is ancient in IT terms.

I had to work hard to get a certification in it. 

We use it for different companies and different clients. We have Fortune 500, startups, and mid-sized companies as our clients. They are in healthcare, finance, fintech, tech, manufacturing, construction, real estate, telecom, and a lot of other industries. They all love it.

It is the best in terms of performance and scale, and it can do both SQL and NoSQL workloads, so it is pretty impressive. One of the least understood use cases happens to be cost and caching. It has a pretty amazing caching engine, and its cost is amazingly low. Especially with the strategies we have designed, we can show a cost reduction of 99% in certain cases. The request charge reduction is anywhere between 75% to 99%. It has been pretty amazing to get cost, quality, and time. We can get all three with it. It is one of the very few databases that can even get there.

We use the built-in vector database capability. It is pretty fascinating. You get recalls that are pretty high. In the competitive landscape of databases, it is surprisingly better in terms of p95 latency and also requests per second, which is something that every customer wants but does not easily get by default. You can also use HNSW, for example, a lot cheaper than you would otherwise because of the DiskANN technology. It is similar to HNSW, but it is on the disk, so it is cheaper. You are not going to the memory. That saves you a lot of money, which is important because when you are running workloads that are getting to terabytes and terabytes, the cost is a huge concern, especially to support the underlying business. That is pretty amazing in terms of DiskANN, which is a Microsoft technology that is very well implemented in Azure Cosmos DB.

Usually, the vector database is integrated with a bunch of other applications. It could be a CRM system behind the scenes, or it could be any LLM-based application. The interoperability with other solutions is fairly simple because, at the end of the day, it is just an API. You can make it work with anything.

We use a lot of different models including Azure OpenAI and some open-source ones through Azure AI Studio. They are very easy to use with it because it is just an API.

It is as fast as what you would find elsewhere. It scales, and they do that part for you. Performance and scale are the things that Azure Cosmos DB got right. That is definitely a positive. If you do not use the vector database, you may get into hallucination issues. Things might slow down. Such issues do not happen if you are using the vector database correctly. Your LLM is now supported with the rack pattern, which is done very well by DiskANN and Azure Cosmos DB.

DiskANN does a great job with recalls. We can decide how high we want them to be. That is the best it gets. If you are using vector searches, it does a great job. I do not usually use Azure Cosmos to compete with a regular or classic search engine.

In terms of Azure Cosmos DB’s ability to search through large amounts of data, currently, the maximum we have on it is in terabytes, but a lot of that depends on how you do a lot of things. That includes data modeling and partitioning, and then your entire vector strategy, which is what we specialize in. We have seen great results. You get the best of all worlds. For example, there is a higher recall at pretty amazing requests per second, which in some cases is 10x to 15x of what you would get for the same recall with another engine. Your latency is also a lot lower. In some cases, it is incredibly low. For example, it is 10x to 15x lower than others. This combination is very hard to get with other databases that we have tried, so from that angle, Azure Cosmos DB has done a terrific job.

A few years ago, we put out a report that took Azure Cosmos DB as it is and compared it with other databases out there, and it was 92% cheaper on reads and 20% cheaper on writes. After that, we used our optimization, and we were able to further reduce that by another 75% to 99% in different cases. We have an online talk about it where we partially show how to get there. Those are not full solutions. It was a conference where I had 15 minutes, and I ended up doing a demo. 

It is pretty fascinating because it is very hard for other databases to come anything close to it in terms of the cost given the fact that you have pretty amazing performance and scale. A lot of people can beat you on Azure Cosmos DB, but they do not give you the right performance and scale you need for business, so those cost savings are meaningless. For our customers, it has got to be the best of all worlds, and fortunately, Azure Cosmos DB has that.

Pretty much all of the features are valuable. Change feed is a pretty amazing feature. Once you make the changes, they are quickly read for you, and then you also have geo-replication. You can do a lot of things in your region, and the same regions can be replicated all over the world. There are different geographies. I can have my servers pretty much anywhere in the world. The data could be within the country or continent when there is a data restriction policy and things like that. Security is big. There are a lot of very good features.

It is very easy and very simple now given all the improvements, but it is also designed very well. Especially because we specialize in it, it is the easiest thing on our side. Data modeling happens to be a lot easier than SQL and others. The learning curve is a lot smaller than a typical RDBMS. It is very like code, and that is another benefit. Developers love it because you do not have to learn something new. You can use the same object that you are using in your code, and you can write stored procedures in JavaScript if you want to. If you want to do anything else, you could use the SQL API or NoSQL API, or you could use MongoDB API. It supports a lot of different APIs. You do not have to learn anything new, so the learning curve is way smaller than pretty much anything out there.

The best part of Azure Cosmos DB is that you barely have any maintenance. This is what I liked about it in the first place. 

In the long run, there should be an addition of more features, especially because this space is evolving quickly. It all boils down to how many more features you are adding, how many integrations you are supporting, and how many more APIs you have that are standard APIs. The team is already doing a great job. They are already doing all that is needed, but the more features we have, the easier it is for us and our clients.

For example, when you have these vectors, it requires us to know a little bit about the configuration behind things such as HNSW. When it comes to the MongoDB vCore piece of Azure Cosmos DB, people like us know how to get to higher recalls easily but a regular user may not. If they have a feature that provides an easy way to get to a certain recall you need, and that is a configuration by default, that would be great. Currently, the flexibility is amazing, and we love that. The competition is usually not providing that. The competition sometimes gives you a recall of 80%, but they are taking away the latency and requests per second. Azure Cosmos DB does not do that. It is a better solution. If Azure Cosmos DB has configurations and a feature allowing us to pick any of the use cases we want, it would be great. For example, if I have an application that I am okay with, and my application does not require a huge recall that is 80% but needs one that is 60%, for us, it is very easy to take HNSW and do it that way and reduce the requests per second because that might not be a concern. If there is a feature that allows people to pick out of five different permutations and combinations, it would be very easy for anyone else to do it. However, keep in mind that competition does not even have that flexibility, so competition is lagging behind on that, at least in the case of the ones we have tried. If Azure Cosmos DB has such a feature, it will be easy for more people to take advantage of the things we are taking advantage of.

I started using it when the first version of DocumentDB came out, and then a couple of years later, it was renamed to Azure Cosmos DB. I have been working with it even before it was called Azure Cosmos DB. I am still using it, and it is my favorite database.

In 2020, we put out a report. It is on our website. This was the only one out of all the major databases out there that had a linear throughput increase for hundreds of servers. What is crazy is that everyone else said that they do it, but you could literally see that after a certain number, they would slow down. One of them was a pretty majorly known, multibillion-dollar NoSQL database, but after 50 servers, it would just slow down completely. You could literally say, "Wow, it is taking time to go to the next level." Azure Cosmos DB was the only one that had linear throughput. Our team thinks that the underlying infrastructure of Azure Cosmos DB is procured in advance, and that is why they can have a linear scale for as much as you want to go. It totally depends on how big, literally, Azure is. The competition is running on someone else's cloud mostly. They probably procure machines as they go, and maybe that is why they are slow. This is an interesting thing that made us love Azure Cosmos DB.

Latency is pretty straightforward. They guarantee 10 milliseconds read and write times now, which used to be 10 and 15 earlier. That is one thing that is pretty incredible. A few things that we have shared with customers is that there is always a wrong way of doing things. You may know the right thing, and it is easy for you to get the latencies, but you may completely mess up the data modeling as well as your code and hundreds of different things. You may be making five calls when you need to make one. You may be doing a lot of other things that are not necessarily best practices. In some cases, we have seen people having a latency of more than 200 milliseconds, which we brought down to 10 milliseconds.

Azure Cosmos DB’s dynamic scaling decreases an organization’s overhead costs big time. This is where it stands out. A lot of our clients who were previously using RDBMS kind of solutions found them to be slow. It would take forever. They would not scale beyond a certain point, and they would get extremely costly after some time because you have huge machines. Some of them were paying millions of dollars for one machine, whereas now, they just pay and go and they can scale. They can go all the way up, and if they want to go down, they go down. For example, on Thanksgiving, they may be scaling to hundreds of nodes behind the scenes, and the next day, they could be scaling back to three nodes. Imagine the cost savings when you never had to procure those servers or anything else. This is the genius of this whole thing. They are being able to take advantage of a scale that they could have never had as an organization. This is not just for startups. Even big corporations can take advantage of the same exact thing. It works for every single one out there.

I never had to contact support. If you know what you are doing, then it is really good. I do not even know of anyone calling Azure Cosmos DB support for anything.

I have worked with almost all of the competitor solutions. I cannot think of any disadvantages of Azure Cosmos DB unless the competition is for a specific use case. For example, in the beginning, Redis was great for caching. It had a great Pub/Sub Over a period of time, Cosmos DB got there, but Azure Cosmos DB is a lot more than just that use case. Today, it is easy for me to pick Azure Cosmos DB as a caching engine. I would not have done that five or six years ago.

We do not do much with the on-premises version. We only work with the cloud version.

Its deployment is pretty easy. I have been doing it for a long while, so it is easy for us and our clients. I do not know about others.

Our implementation strategy depends on what kind of project it is. We do greenfield, brownfield, and all of the projects. We do integration projects. We also do projects where they are only doing an addition of LLM. We start with understanding the client's needs and then figuring out what is currently there. If there is nothing, we data model the whole thing from scratch and go with the best practices. A lot of times, if it is brownfield, a bunch of work is already done, so we are not going and figuring out what is the most optimum way to do it within client constraints. We create a strategy based on that and implement it based on their tech stack because everyone has a different one. Once we get the approval on that, we move forward with the implementation.

The number of people required and the time required depend on the client and their workload. If you have a small app, you could be onboarded on day one. If you have a big app with petabytes of data, it is usually a month of work. It totally depends on what we are looking at or the use case.

In terms of the learning curve, Azure Cosmos DB is one of the simplest ones out there. It is on the easier side. There are a couple of them that are pretty easy. 

To new users, I would advise understanding different propositions. Start with understanding what kind of data set you have and every single thing. Also, know the tech stack you have and pick your strategy accordingly. What you do not want to do is go with the flow without understanding what a NoSQL database is supposed to be like and make changes down the line.

A lot of people with an SQL background, unfortunately, start using any NoSQL databases, not just Azure Cosmos DB, in a way that is not very good for them because the patterns that usually are the best patterns for SQL may not be the best patterns for NoSQL. For example, there is a reason we do normalization in SQL. That takes away duplicate data, which is perfectly okay, but in the case of NoSQL or Azure Cosmos DB, we can scale and have duplicate data in places if we have a different kind of use case. If I want to make different kinds of searches available, I can have three different kinds of searches available for similar kinds of parameters. I will not be worried about doing that in a NoSQL environment because I can scale out pretty easily, so data does not hold me back. In an RDBMS environment, I might be doing two or three joins to make sure that I am making it fully normalized because if my data increases drastically, that will create a scale up situation. Scale-up is the only thing you can literally do with RDBMSs. Mostly, scale-out is not that easy unless you are on the cloud and you are using the scale of the cloud, and then you have performance issues. In those kinds of different scenarios, the DBAs or people with an RDBMS background need to come up with an open mind and understand what this is.

It is not that you have to learn a lot about Azure Cosmos DB. You will have to learn about this new paradigm. It is not very new. It has been going on for more than a decade. We have been doing it for more than a decade, but we see a lot of people coming from an RDBMS background and getting it wrong, and then you're paying people like us a lot more money to fix it. It is easier to work with someone like us in the beginning or do a little bit more due diligence and learn that paradigm before you get started. That will save you a lot of money and time, and hopefully, you will not need our services at that point of time. That is definitely my advice.

I would rate Azure Cosmos DB a ten out of ten at this time.

The main use case for Microsoft Azure Cosmos DB is as a key-value store where we store all the user data that we have and perform lookups. We use it at a significant scale, with storage of unique data reaching 12 terabytes and handling up to 3 million requests per second.

The scalability of Microsoft Azure Cosmos DB has significantly aided us in managing workloads efficiently.

We were able to realize the benefits of Microsoft Azure Cosmos DB immediately after deployment, making it quite easy to get started.

The most valuable features of Microsoft Azure Cosmos DB include the TTL, the ability to scale up and down as needed, and geo-replication, which comes out of the box. We do not have to do anything for geo-replication. We just have to enable it.

The indexing policy is also very good, and the overall metrics and monitoring system are also quite good.

Microsoft Azure Cosmos DB is fairly easy to use.

Microsoft Azure Cosmos DB can be improved by providing more fine-grained control over certain aspects, such as connections and threads. There could be more control over how many connections are made. I am not sure if it is a knowledge gap issue. A regular connection with the Azure Cosmos DB team might help in addressing knowledge gaps. Being able to fine-tune these features would be helpful for us.

I have been using Microsoft Azure Cosmos DB for about six years.

Over the last two years, Microsoft Azure Cosmos DB has been very stable. It has very good latency and availability. Latency is good on the server side and the client side. We have had only one significant issue that affected our production system. Overall, stability has been excellent.

The scalability of Microsoft Azure Cosmos DB is one of its best attributes. We can scale very efficiently and adjust workloads as needed, which is more challenging with other systems.

We have contacted their support many times. The quality of customer and technical support has improved over the years. Initially, it used to take quite a while for issues to be resolved, but now the support is seamless and very efficient. We have not needed much support in the last couple of years due to the system's stability. It is pretty stable now. I would rate their support a nine out of ten.

Positive

I have used Redis briefly and Aerospike extensively before switching to Microsoft Azure Cosmos DB.

Both Microsoft Azure Cosmos DB and Aerospike have their own advantages. The biggest advantage of Microsoft Azure Cosmos DB is that it is very easy to get started with and it does not require too much effort. It takes just one click to deploy Microsoft Azure Cosmos DB and put it into multiple regions. It does not require too much maintenance, whereas Aerospike requires a lot of maintenance effort. It requires a dedicated team. In this aspect, Microsoft Azure Cosmos DB is very good. However, Aerospike provides control over a few things, which we do not have in Microsoft Azure Cosmos DB. If we want to run or use the maximum amount of resources, Aerospike helps a lot. Both have their advantages and disadvantages.

The initial setup was easy. It was not difficult.

It took us a quarter to be able to use it efficiently. It is fairly easy and straightforward.

We had set up our own autoscaler. There was a pipeline that ran on top of Azure Cosmos DB to see how many RUs were provisioned. It did require a little bit of maintenance because we built custom software on top of that, but that was it. Our autoscaler performed better than Azure Autoscaler. However, because of some billing benefits, we have started using Azure Autoscaler. The Microsoft team said that if we used Azure Autoscaler, they would give us a discount, so we started using that, but our autoscaler performed better.

Initially, the deployment required an entire team, but now, it can be managed by a smaller team of two to three engineers.

It has decreased our total cost of ownership by approximately 20% compared to other alternatives such as Redis.

Its pricing is higher compared to solutions like Aerospike. However, it is justified because of the out-of-the-box features that are provided. The availability and resiliency that we have make it worth the price.

To new users, I would advise first knowing their data. They should know whether it fits their solution, which Azure Cosmos API to use, and what scale they intend to run it.

I would rate Microsoft Azure Cosmos DB a nine out of ten.