I am new to DynamoDB though I am not totally new to NoSQL paradigm. I worked with Firebase years ago.
I spent the last couple days learning and studying most of the materials that I can find on the Internet about single-table design, a design approach advocated by the DynamoDB team. I think I have got the essence of it and am fascinated by the concept.
All the materials that I read on single-table design are under the context of related entities, however. This makes me wonder what about unrelated entities (as in I don't need to perform JOINs on the entities if I were to implement the service with a SQL database). What are the pros and cons of putting unrelated entities in the same table vs putting them in separate tables? (in terms of performance, monetary cost, maintainability and etc.)
There might be some cost benefit to storing unrelated entities in the same table.
But only if using provision capacity, and really only then if the I/O to the unrelated data is insignificant in terms of the I/O to the main table.
So if you could have 1 table with 35 RCU/WCU vs having that table plus another table with 1 RCU/WCU, you could save a few pennies on capacity. Storage cost would be the same regardless.
But don't forget that the DDB "always free" tier includes 25GB of storage, 25 WCU, 25 RCU. Number of tables isn't a factor.
At scale, it'd be better to have them separate so you could better tune the capacity to the workload.
I suppose if you needed a million 1 xCU tables rather than 1 25 xCU table.... it'd make a difference. But pay-per-request is likely a better option in that case.
Related
Coming from an SQL background, it's easy to fall into a SQL pattern and mindset when designing NOSQL databases like DynamoDB. However, there are many best practices that rely on merging different kinds of data with different schemas in the same table. This can be very efficient to query for related data, in lieu of SQL joins.
However, if I have two distinct types of data with different schemas, and which are never queried together, since the introduction of on demand pricing for DynamoDB, is there any reason to merge data of different types into one table simply to keep the number of tables down? Prior to on demand, you had to pay for the capacity units per hour, so limiting the number of tables was reasonable. But with on demand, is there any reason not to create 100 tables if you have 100 unrelated data schemas?
I would say that the answer is "no, but":
On-demand pricing is significantly more expensive than provisioned pricing. So unless you're just starting out with DynamoDB with a low volume of requests, or have extremely fluctuating demand you are unlikely to use just on-demand pricing. Amazon have an interesting blog post titled Amazon DynamoDB auto scaling: Performance and cost optimization at any scale, where they explain how you can reserve some capacity for a year, then automatically reserve capacity for 15 minute intervals (so-called autoscaling), and use on-demand pricing just to demand exceeding those. In such a setup, the cheapest prices are the long-term (yearly, and even 3 year) reservations. And having two separate tables may complicate that reservation.
The benefit of having one table would be especially pronounced if your application's usage of the two different tables fluctuates up and down over the day. The sum of the two demands will usually be flatter than each of the two demands, allowing the cheaper capacity to be used more and on-demand to be used less.
The reason why I answered "no, but" and not "yes" is that it's not clear how much these effects are important in real applications, and how much can you save - in practice - by using one instead of two tables. If the number of tables is not two but rather ten, or the number of tables changes over the evolution of the application, maybe the saving can be even greater.
If each of my database's an overview has only two types (state: pending, appended), is it efficient to designate these two types as partition keys? Or is it effective to index this state value?
It would be more effective to use a sparse index. In your case, you might add an attribute called isPending. You can add this attribute to items that are pending, and remove it once they are appended. If you create a GSI with tid as the hash key and isPending as the sort key, then only items that are pending will be in the GSI.
It will depend on how would you search for these records!
For example, if you will always search by record ID, it never minds. But if you will search every time by the set of records pending, or appended, you should think in use partitions.
You also could research in this Best practice guide from AWS: https://docs.aws.amazon.com/en_us/amazondynamodb/latest/developerguide/best-practices.html
Updating:
In this section of best practice guide, it recommends the following:
Keep related data together. Research on routing-table optimization
20 years ago found that "locality of reference" was the single most
important factor in speeding up response time: keeping related data
together in one place. This is equally true in NoSQL systems today,
where keeping related data in close proximity has a major impact on
cost and performance. Instead of distributing related data items
across multiple tables, you should keep related items in your NoSQL
system as close together as possible.
As a general rule, you should maintain as few tables as possible in a
DynamoDB application. As emphasized earlier, most well designed
applications require only one table, unless there is a specific reason
for using multiple tables.
Exceptions are cases where high-volume time series data are involved,
or datasets that have very different access patterns—but these are
exceptions. A single table with inverted indexes can usually enable
simple queries to create and retrieve the complex hierarchical data
structures required by your application.
Use sort order. Related items can be grouped together and queried
efficiently if their key design causes them to sort together. This is
an important NoSQL design strategy.
Distribute queries. It is also important that a high volume of
queries not be focused on one part of the database, where they can
exceed I/O capacity. Instead, you should design data keys to
distribute traffic evenly across partitions as much as possible,
avoiding "hot spots."
Use global secondary indexes. By creating specific global secondary
indexes, you can enable different queries than your main table can
support, and that are still fast and relatively inexpensive.
I hope I could help you!
Does GSI Overloading provide any performance benefits, e.g. by allowing cached partition keys to be more efficiently routed? Or is it mostly about preventing you from running out of GSIs? Or maybe opening up other query patterns that might not be so immediately obvious.
https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/bp-gsi-overloading.html
e.g. I you have a base table and you want to partition it so you can query a specific attribute (which becomes the PK of the GSI) over two dimensions, does it make any difference if you create 1 overloaded GSI, or 2 non-overloaded GSIs.
For an example of what I'm referring to see the attached image:
https://drive.google.com/file/d/1fsI50oUOFIx-CFp7zcYMij7KQc5hJGIa/view?usp=sharing
The base table has documents which can be in a published or draft state. Each document is owned by a single user. I want to be able to query by user to find:
Published documents by date
Draft documents by date
I'm asking in relation to the more recent DynamoDB best practice that implies that all applications only require one table. Some of the techniques being shown in this documentation show how a reasonably complex relational model can be squashed into 1 DynamoDB table and 2 GSIs and yet still support 10-15 query patterns.
https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/bp-relational-modeling.html
I'm trying to understand why someone would go down this route as it seems incredibly complicated.
The idea – in a nutshell – is to not have the overhead of doing joins on the database layer or having to go back to the database to effectively try to do the join on the application layer. By having the data sliced already in the format that your application requires, all you really need to do is basically do one select * from table where x = y call which returns multiple entities in one call (in your example that could be Users and Documents). This means that it will be extremely efficient and scalable on the db level. But also means that you'll be less flexible as you need to know the access patterns in advance and model your data accordingly.
See Rick Houlihan's excellent talk on this https://www.youtube.com/watch?v=HaEPXoXVf2k for why you'd want to do this.
I don't think it has any performance benefits, at least none that's not called out – which makes sense since it's the same query and storage engine.
That being said, I think there are some practical reasons for why you'd want to go with a single table as it allows you to keep your infrastructure somewhat simple: you don't have to keep track of metrics and/or provisioning settings for separate tables.
My opinion would be cost of storage and provisioned throughput.
Apart from that not sure with new limit of 20
I have a messaging app, where all messages are arranged into seasons by creation time. There could be billions of messages each season. I have a task to delete messages of old seasons. I thought of a solution, which involves DynamoDB table creation/deletion like this:
Each table contains messages of only one season
When season becomes 'old' and messages no longer needed, table is deleted
Is it a good pattern and does it encouraged by Amazon?
ps: I'm asking, because I'm afraid of two things, met in different Amazon services -
In Amazon S3 you have to delete each item before you can fully delete bucket. When you have billions of items, it becomes a real pain.
In Amazon SQS there is a notion of 'unwanted behaviour'. When using SQS api you can act badly regarding SQS infrastructure (for example not polling messages) and thus could be penalized for it.
Yes, this is an acceptable design pattern, it actually follows a best practice put forward by the AWS team, but there are things to consider for your specific use case.
AWS has a limit of 256 tables per region, but this can be raised. If you are expecting to need multiple orders of magnitude more than this you should probably re-evaluate.
You can delete a table a DynamoDB table that still contains records, if you have a large number of records you have to regularly delete this is actually a best practice by using a rolling set of tables
Creating and deleting tables is an asynchronous operation so you do not want to have your application depend on the time it takes for these operations to complete. Make sure you create tables well in advance of you needing them. Under normal circumstances tables create in just a few seconds to a few minutes, but under very, very rare outage circumstances I've seen it take hours.
The DynamoDB best practices documentation on Understand Access Patterns for Time Series Data states...
You can save on resources by storing "hot" items in one table with
higher throughput settings, and "cold" items in another table with
lower throughput settings. You can remove old items by simply deleting
the tables. You can optionally backup these tables to other storage
options such as Amazon Simple Storage Service (Amazon S3). Deleting an
entire table is significantly more efficient than removing items
one-by-one, which essentially doubles the write throughput as you do
as many delete operations as put operations.
It's perfectly acceptable to split your data the way you describe. You can delete a DynamoDB table regardless of its size of how many items it contains.
As far as I know there are no explicit SLAs for the time it takes to delete or create tables (meaning there is no way to know if it's going to take 2 seconds or 2 minutes or 20 minutes) but as long your solution does not depend on this sort of timing you're fine.
In fact the idea of sharding your data based on age has the potential of significantly improving the performance of your application and will definitely help you control your costs.
I have a big analytics module in my system and plan to use vertica for it.
Someone suggested that we also use vertica in the rest of our app (standard crud app with models from our domain) so not to manage multiple databases.
would vertica fit this dual scenario?
High frequency UPDATEs is probably where Vertica lags behind the worst. I would avoid using it for such data models.
Alec - I would like to respectfully challenge your comments on Vertica. In no way do you need to denormalize or sort data before loading. Vertica also holds the record for fastest loading of data over all databases.
You also talk about Vertica not being able to do complex analytics as well as an RDBMS. Vertica IS an RDBMS and can do analytics faster than any other RDBMS and they prove it over and over.
As far as your numbers, in my use case I load roughly 5 million records per second into my Vertica cluster and have 100's of billions of records.
So Yaron - I would highly recommend you look at Vertica before you rule it out based on this information.
As is often the case these days, a meaningful answer depends on what you need to do. In a general sense, 'big data' solutions have grown from large data volume deficiencies in RDBMS systems. No 'big data' solution can compete with the core capabilities of RDBMS systems, ie complex analytics, but RDBMS systems are poor (expensive) solutions for large data volume procesing. Practical solutions for now have to be hybrid solutions. Vertica can be good once data is loaded, but I believe (not an expert) it requires denormalisation of data and pre-sorting before loading to perform at it's best. For large data volumes this may add significantly to the required resources. There is a definite benefit to using one system for all your needs, but there are also benefits to keeping your options open.
The approach I take is to store and index new data and then provide specific feeds to various reporting/analytic engines as required. This separates the collection and storage of raw data from the complex analytic processing. I am happy to provide more details if you are interested. This separation addresses a core problem which has always been present in database systems. In the past you used to hear 'store fast, report slowly or store slowly, report fast, but you cannot do both'. The search for a complete solution has, in the last few years, spawned the many NoSQL offerings which typically address the 'store fast' task. Some systems also provide impressive query performance by storing data in memory or cache but this requires many servers for large data volumes. I believe NoSQL and SQL solutions can, and will be, integrated, but this is till down the track.
To give you some context, I work with scenarios where at least 1 billion records a day are loaded. If you are dealing with say 100 million records a day (big is relative), then your Vertica approach will probably suffice, otherwise I think you need to expand your options.
Test it. Each use case is different. Assuming Vertica is a solution for every use case is almost as bad as using MongoDB for every use case.
Vertica is a high performance analytics database, column oriented, designed to analyze incredibly large datasets and scale horizontally. It's also expensive, hard to administer, and documentation is spotty. The payoff in the right environment can be easily worth the work, obviously
MySQL is a traditional RDBMS, row oriented, designed to model relationships between structured data, and works well on a single node scale (though many companies have retrofitted it to great success, exemplar gratia, Facebook). It's incredibly well documented and seemingly works on any platform, language, or framework and can be used by anyone.
My guess is using Vertica for an employee address book database is like showing up to a blue collar job in a $3000 suit. Sure it works, but is it the right tool for the job? Maybe if you already have a Vertica license and your applications already have the requisite data adaptors/ORM/etc..., go ahead and give it a shot. It's still a SQL database so it should work fine in those situations. If your goal is minimal programming as opposed to optimal performance, then why use Vertica at all? Sounds like something simpler would be more ideal. Vertica may or may not give better performance in a regular CRUD application environment since it's not optimized for that, but you can always test both and see.
Vertiy have many issues with high concurrency (Many small transaction per minute )
In MPP systems , the data is segmented across the cluster and any time there is need to take cluster level lock ( mainly in commit time ) , so many commits many cluster level X locks .
high concurrency is less the use case in DWH and reporting , so vertica is perfect for that .
In most of the cases OLTP solutions ( like CRM and etc ) required to provide high concurrency for that very is bad choice
Thanks