I am trying to model my data. As you see the partition key is the user email. In the global secondary index I have a PK of "US", which stands for "User". If I want to get all of the enabled users I just have to query the GSI where GSI1PK = "US" and GSI1SK Starts with "Enabled".
My concern is that all of the users in the app would have the same GSI1PK. Will this be a problem? Can GSIs PK have problems with "hot partitions"? I am Googling this and I do not see a clear answer. There is only one here on StackOverflow that says it will be a problem, but there are other places that say it will not. I am kind of confused.
What would be the best way to structure the data in my table so I can access all of the users without causing hot artition issues?
Placing a potentially large item collection in a single partition will likely lead to a hot partition. Ideally, your chosen partition keys evenly distribute data across partitions. However, it may not always be clear about how to achieve this.
You might consider splitting your large partition into smaller partitions on write (aka write sharding), and re-combining them when reading. For example, when creating GSIPK, you could introduce a randomly generated integer between 1 and 4 in the partition key:
And your GSI would look like this
Now your User data is more evenly distributed across partitions. When reading users from your table, you would pull from all the partitions at once. This could be done in parallel for faster performance.
In this example, I chose a random number to "write shard" the data into separate partitions. However, your data may lend itself to a more natural division (e.g. by country, enabled status, time zone, etc). What I want to highlight is that your strategy to distribute data across partitions can be separate from the data model you use to support your application access patterns.
Related
I am modelling the data of my application to use DynamoDB.
My data model is rather simple:
I have users and projects
Each user can have multiple projects
Users can be millions, project per users can be thousands.
My access pattern is also rather simple:
Get a user by id
Get a list of paginated users sorted by name or creation date
Get a project by id
get projects by user sorted by date
My single table for this data model is the following:
I can easily implement all my access patterns using table PK/SK and GSIs, but I have issues with number 2.
According to the documentation and best practices, to get a sorted list of paginated users:
I can't use a scan, as sorting is not supported
I should not use a GSI with a PK that would put all my users in the same partition (e.g. GSI PK = "sorted_user", SK = "name"), as that would make my single partition hot and would not scale
I can't create a new entity of type "organisation", put all users in there, and query by PK = "org", as that would have the same hot partition issue as above
I could bucket users and use write sharding, but I don't really know how I could practically query paginated sorted users, as bucket PKs would need to be possibly random, and I would have to query all buckets to be able to sort all users together. I also thought that bucket PKs could be alphabetical letters, but that could crated hot partitions as well, as the letter "A" would probably be hit quite hard.
My application model is rather simple. However, after having read all docs and best practices and watched many online videos, I find myself stuck with the most basic use case that DynamoDB does not seem to be supporting well. I suppose it must be quite common to have to get lists of users in some sort of admin panel for practically any modern application.
What would others would do in this case? I would really want to use DynamoDB for all the benefits that it gives, especially in terms of costs.
Edit
Since I have been asked, in my app the main use case for 2) is something like this: https://stackoverflow.com/users?tab=Reputation&filter=all.
As to the sizing, it needs to scale well, at least to the tens of thousands.
I also thought that bucket PKs could be alphabetical letters, but
that could create hot partitions as well, as the letter "A" would
probably be hit quite hard.
I think this sounds like a reasonable approach.
The US Social Security Administration publishes data about names on its website. You can download the list of name data from as far back as 1879! I stumbled upon a website from data scientist and linguist Joshua Falk that charted the baby name data from the SSA, which can give us a hint of how names are distributed by their first letter.
Your users may not all be from the US, but this can give us an understanding of how names might be distributed if partitioned by the first letter.
While not exactly evenly distributed, perhaps it's close enough for your use case? If not, you could further distribute the data by using the first two (or three, or four...) letters of the name as your partition key.
1 million names likely amount to no more than a few MBs of data, which isn't very much. Partitioning based on name prefixes seems like a reasonable way to proceed.
You might also consider using a tool like ElasticSearch, which could support your second access pattern and more.
I've included some links along with our approaches to other answers, which seem to be the most optimal on the web right now.
Our records need to be categorized (eg. "horror", "thriller", "tv"), and randomly accessible both in specific categories and across all/some categories. We generally need to access about 20 - 100 items at a time. We also have a smallish number of categories (less than 100).
We write to the database for uploading/removing content, although this is done in batches and does not need to be real time.
We have tried two different approaches, with two different data structures.
Approach 1
AWS DynamoDB - Pick a record/item randomly?
Help selecting nth record in query.
In short, using the category as a hash key, and a UUID as the sort key. Generate a random UUID, query Dynamo using greater than or less than, and limit to 1. This is even suggested by an AWS employee in the second link. (We've also tried increasing the limit to the number of items we need, but this increases the probability of the query failing the first time around).
Issues with this approach:
First query can fail if it is greater than/less than any of the UUIDs
Querying on any specific category will cause throttling at scale (Small number of partitions)
We've also considered adding a suffix to each category to artificially increase the number of partitions we have, as pointed out in the following link.
AWS Database Blog
Choosing the Right DynamoDB Partition Key
Approach 2
Amazon Web Services: How do we get random item from the dynamoDb's table?
Doing something similar to this, where we concatenate the category with a sequential number, and use this as the hash key. e.g. horror-000001.
By knowing the number of records in each category, we're able to perform random queries across our entire data set, while also avoiding hot partitions/keys.
Issues with this approach
We need a secondary data structure to manage the sequential counts across each category
Writing (especially deleting) is significantly more complex, although this doesn't need to happen in real time.
Conclusion
Both approaches solve our main use case of random queries on category/categories, but the cons they offer are really deterring us from using them. We're leaning more towards approach #1 using suffixes to solve the hot partitioning issue, although we would need the additional retry logic for failed queries.
Is there a better way of approaching this problem? Specifically looking for solutions capable of scaling well (No scan), without requiring extra resources be implemented. #1 fits the bill, but needing to manage suffixes and failed attempts really deters us from using it, especially when it is being called inside a lambda (billed for time used).
Thanks!
Follow Up
After more research and testing, my team has decided to move towards MySQL hosted on RDS for these tables. We learned that this is one of the few use cases were DynamoDB does not fit, and requires rewriting your use case to fit the DB (Bad).
We felt that the extra complexity required to integrate random sampling on DynamoDB wasn't worth it, and we were unable to come up with any comparable solutions. We are, however, sticking with DynamoDB for our tables that do not need random accessibility due to the price and response times.
For anyone wondering why we chose MySQL, it was largely due to the Nodejs library available, great online resources (which DynamoDB definitely lacks), easy integration via RDS with our Lambdas, and the option to migrate to Amazons Aurora database.
We also looked at PostgreSQL, but we weren't as happy with the client library or admin tools, and we believe that MySQL will suit our needs for these tables.
If anybody has anything else they'd like to add or a specific question please leave a comment or send me a message!
This was too long for a comment, and I guess it's pretty much a full fledged answer now.
Approach 2
I've found that my typical time to get a single item from dynamodb to a host in the same region is <10ms. As long as you're okay with at most 1-2 extra calls, you can quite easily implement approach 2.
If you use a keys only GSI where the category is your hash key and the primary key of the table is your range key, you can quickly find the largest numbered single item within a category.
When you add a new item, find the largest number for that category from the GSI and then write the new item to the table with sequence number n+1.
When you delete, find the item with the largest sequence number for that category from the GSI, overwrite the item you are deleting, and then delete the now duplicated item from its position at the highest sequence number.
To randomly get an item, query the GSI to find the highest numbered item in the category, and then randomly pick a number since you now know the valid range.
Approach 1
I'm not sure exactly what you mean when you say "without requiring extra resources to be implemented". If you're okay with using a managed resource (no dev work to implement), you can also make Approach 1 work by putting a DAX cluster in front of your dynamodb table. Then you can query to your heart's content without really worrying about hot partitions. (Though the caching layer means that new/deleted items won't be reflected right away.)
Related to this question, I'm looking for more a more specific answer. In an effort to keep this non-subjective, here is a full thought process for creating an activities table with a stuck point that can be finished with a quick example answer.
In an effort to better understand DynamoDB, I'm creating a personal website that contains an activity feed from a DynamoDB table. The goal is to evenly distribute partition keys while still being able to sort across all partition keys (I'm struggling with this part).
Different types of activities will include blog posts, projects, twitter post references, LinkedIn post references, etc. Using the activity type as a partition key would not be wise as my activity is highly weighted, mostly on the twitter side, hardly ever creating blog posts.
A unique activity id seems to be the best option for evenly distributing activities across DynamoDB partitions. However, this completely removes the ability to sort activities to start, as queries require a partition id to be known first. This is where a secondary global index (SGI) will be helpful. With this, a sort key will not be required on the primary partition key, but paired in an SGI.
This is part where I'm stuck. What do I base the SGI partition key on? At the moment I'm thinking of a single value "activity" for all activities with a sort key of "date", but that is a single partition for all entries. Will a single SGI partition key value limit performance in this project?
Note that this is a small scale project. However, I'm thinking about large scale projects while building this one, attempting to create the best DynamoDB table possible in regards to optimized partition distribution, while still keeping it flexible for sorting all table records.
Consider GSI (Global Secondary Index) same as Main Table indexes while designing your schema as they also get Read/Write provisioning limits and are subject to hot partition throttling as well which back pressures on main table in other words if your GSI gets throttled then your main table will start throttling requests.
Will a single SGI partition key value limit performance in this project?
Single partition for complete table is definitely misuse of DDB scalable capability.
The goal is to evenly distribute partition keys while still being able to sort across all partition keys (I'm struggling with this part).
You can sort across partitions using GSI but you will again need partition key for your GSI and if that partition key is not distributed enough then you get into problems I mentioned above.
DDB is powerful for put/get operations if modeled right and for fairly simple queries with some filters. In general, you will utilize your throughput more efficiently as the ratio of partition key values accessed to the total number of partition key values in a table grows.
For your specific need its not directly possible to get scalable solution from DDB but we still have few options
Option 1:
We can model the data such that it is fairly distributed for writes and will need extra work while reading it back, this pattern is also known as Randomizing Across Multiple Partition Key Values. Since you don't want to access specific item for given time this will work for us.
Idea is to create fixed set (say 1 to 100) and randomly pick a number from it to append to creation date (not timestamp) and have creation timestamps as sort key.
This will distribute your load across multiple random partitions but increases the read complexity as you will need to query all partitions and merge to get final sort view for that date.
Option 2:
Use multiple tables for hot and cold data as it is time series based data. For info read
http://docs.aws.amazon.com/amazondynamodb/latest/developerguide/GuidelinesForTables.html#GuidelinesForTables.TimeSeriesDataAccessPatterns
Option 3:
Scan? Not a good choice if we talk about scalability and when your data grows but for fairly small set of data it surely helps so mentioning it.
These are just an example not saying a good fit for your usecase.
So here is a thought process question for you: write down all your use-cases and access patterns. Figure out their importance which are fine with eventual consistency which are not and see if DDB is good fit for them at first place, don't be tempted to use DDB and then struggling with access pattern scalability.
Also read https://stackoverflow.com/a/38790120/962545 for more questions you must be asking yourself before restricting yourself for specific access pattern you want from DDB.
Don't forget to read best practices: http://docs.aws.amazon.com/amazondynamodb/latest/developerguide/GuidelinesForTables.html
i am working on a migration from MS Sql to DynamoDB and i'm not sure what's the best hash key for my purpose. In MS SQL i've an item table where i store some product information for different customers, so actually the primary key are two columns customer_id and item_no. In application code i need to query specific items and all items for a customer id, so my first idea was to setup the customer id as hash key and the item no as range key. But is this the best concept in terms of partitioning? I need to import product data daily with 50.000-100.000 products for some larger customers and as far as i know it would be better to have a random hash key. Otherwise the import job will run on one partition only.
Can somebody give me a hint what's the best data model in this case?
Bye,
Peter
It sounds like you need item_no as the partition key, with customer_id as the sort key. Also, in order to query all items for a customer_id efficiently you will want to create a Global Secondary Index on customer_id.
This configuration should give you a good distribution while allowing you to run the queries you have specified.
You are on the right track, you should really be careful on how you are handling write operations as you are executing an import job in a daily basis. Also avoid adding indexes unnecessarily as they will only multiply your writing operations.
Using customer_id as hash key and item_no as range key will provide the best option not only to query but also to upload your data.
As you mentioned, randomization of your customer ids would be very helpful to optimize the use of resources and prevent a possibility of a hot partition. In your case, I would follow the exact example contained in the DynamoDB documentation:
[...] One way to increase the write throughput of this application
would be to randomize the writes across multiple partition key values.
Choose a random number from a fixed set (for example, 1 to 200) and
concatenate it as a suffix [...]
So when you are writing your customer information just randomly assign the suffix to your customer ids, make sure you distribute them evenly (e.g. CustomerXYZ.1, CustomerXYZ.2, ..., CustomerXYZ.200).
To read all of the items you would need to obtain all of the items for each suffix. For example, you would first issue a Query request for the partition key value CustomerXYZ.1, then another Query for CustomerXYZ.2, and so on through CustomerXYZ.200. Because you know the suffix range (on this case 1...200), you only need to query the records appending each suffix to the customer id.
Each query by the hash key CustomerXYZ.n should return a set of items (specified by the range key) from that specific customer, your application would need to merge the results from all of the Query requests.
This will for sure make your life harder to read the records (in terms of the additional requests needed), however, the benefits of optimized throughput and performance will pay off. Remember a hot partition will not only increase your overall financial cost, but will also impact drastically your performance.
If you have a well designed partition key your queries will always return very quickly with minimum cost.
Additionally, make sure your import job does not execute write operations grouped by customer, for example, instead of writing all items from a specific customer in series, sort the write operations so they are distributed across all customers. Even though your customers will be distributed by several partitions (due to the id randomization process), you are better off taking this additional safety measure to prevent a burst of write activity in a single partition. More details below:
From the 'Distribute Write Activity During Data Upload' section of the official DynamoDB documentation:
To fully utilize all of the throughput capacity that has been
provisioned for your tables, you need to distribute your workload
across your partition key values. In this case, by directing an uneven
amount of upload work toward items all with the same partition key
value, you may not be able to fully utilize all of the resources
DynamoDB has provisioned for your table. You can distribute your
upload work by uploading one item from each partition key value first.
Then you repeat the pattern for the next set of sort key values for
all the items until you upload all the data [...]
Source:
http://docs.aws.amazon.com/amazondynamodb/latest/developerguide/GuidelinesForTables.html
I hope that helps. Regards.
I am going to implement a notification system, and I am trying to figure out a good way to store notifications within a database. I have a web application that uses a PostgreSQL database, but a relational database does not seem ideal for this use case; I want to support various types of notifications, each including different data, though a subset of the data is common for all types of notifications. Therefore I was thinking that a NoSQL database is probably better than trying to normalize a schema in a relational database, as this would be quite tricky.
My application is hosted in Amazon Web Services (AWS), and I have been looking a bit at DynamoDB for storing the notifications. This is because it is managed, so I do not have to deal with the operations of it. Ideally, I'd like to have used MongoDB, but I'd really prefer not having to deal with the operations of the database myself. I have been trying to come up with a way to do what I want in DynamoDB, but I have been struggling, and therefore I have a few questions.
Suppose that I want to store the following data for each notification:
An ID
User ID of the receiver of the notification
Notification type
Timestamp
Whether or not it has been read/seen
Meta data about the notification/event (no querying necessary for this)
Now, I would like to be able to query for the most recent X notifications for a given user. Also, in another query, I'd like to fetch the number of unread notifications for a particular user. I am trying to figure out a way that I can index my table to be able to do this efficiently.
I can rule out simply having a hash primary key, as I would not be doing lookups by simply a hash key. I don't know if a "hash and range primary key" would help me here, as I don't know which attribute to put as the range key. Could I have a unique notification ID as the hash key and the user ID as the range key? Would that allow me to do lookups only by the range key, i.e. without providing the hash key? Then perhaps a secondary index could help me to sort by the timestamp, if this is even possible.
I also looked at global secondary indexes, but the problem with these are that when querying the index, DynamoDB can only return attributes that are projected into the index - and since I would want all attributes to be returned, then I would effectively have to duplicate all of my data, which seems rather ridiculous.
How can I index my notifications table to support my use case? Is it even possible, or do you have any other recommendations?
Motivation Note: When using a Cloud Storage like DynamoDB we have to be aware of the Storage Model because that will directly impact
your performance, scalability, and financial costs. It is different
than working with a local database because you pay not only for the
data that you store but also the operations that you perform against
the data. Deleting a record is a WRITE operation for example, so if
you don't have an efficient plan for clean up (and your case being
Time Series Data specially needs one), you will pay the price. Your
Data Model will not show problems when dealing with small data volume
but can definitely ruin your plans when you need to scale. That being
said, decisions like creating (or not) an index, defining proper
attributes for your keys, creating table segmentation, and etc will
make the entire difference down the road. Choosing DynamoDB (or more
generically speaking, a Key-Value store) as any other architectural
decision comes with a trade-off, you need to clearly understand
certain concepts about the Storage Model to be able to use the tool
efficiently, choosing the right keys is indeed important but only the
tip of the iceberg. For example, if you overlook the fact that you are
dealing with Time Series Data, no matter what primary keys or index
you define, your provisioned throughput will not be optimized because
it is spread throughout your entire table (and its partitions) and NOT
ONLY THE DATA THAT IS FREQUENTLY ACCESSED, meaning that unused data is
directly impacting your throughput just because it is part of the same
table. This leads to cases where the
ProvisionedThroughputExceededException is thrown "unexpectedly" when
you know for sure that your provisioned throughput should be enough for your
demand, however, the TABLE PARTITION that is being unevenly accessed
has reached its limits (more details here).
The post below has more details, but I wanted to give you some motivation to read through it and understand that although you can certainly find an easier solution for now, it might mean starting from the scratch in the near future when you hit a wall (the "wall" might come as high financial costs, limitations on performance and scalability, or a combination of all).
Q: Could I have a unique notification ID as the hash key and the user ID as the range key? Would that allow me to do lookups only by the range key, i.e. without providing the hash key?
A: DynamoDB is a Key-Value storage meaning that the most efficient queries use the entire Key (Hash or Hash-Range). Using the Scan operation to actually perform a query just because you don't have your Key is definitely a sign of deficiency in your Data Model in regards to your requirements. There are a few things to consider and many options to avoid this problem (more details below).
Now before moving on, I would suggest you reading this quick post to clearly understand the difference between Hash Key and Hash+Range Key:
DynamoDB: When to use what PK type?
Your case is a typical Time Series Data scenario where your records become obsolete as the time goes by. There are two main factors you need to be careful about:
Make sure your tables have even access patterns
If you put all your notifications in a single table and the most recent ones are accessed more frequently, your provisioned throughput will not be used efficiently.
You should group the most accessed items in a single table so the provisioned throughput can be properly adjusted for the required access. Additionally, make sure you properly define a Hash Key that will allow even distribution of your data across multiple partitions.
The obsolete data is deleted with the most efficient way (effort, performance and cost wise)
The documentation suggests segmenting the data in different tables so you can delete or backup the entire table once the records become obsolete (see more details below).
Here is the section from the documentation that explains best practices related to Time Series Data:
Understand Access Patterns for Time Series Data
For each table that you create, you specify the throughput
requirements. DynamoDB allocates and reserves resources to handle your
throughput requirements with sustained low latency. When you design
your application and tables, you should consider your application's
access pattern to make the most efficient use of your table's
resources.
Suppose you design a table to track customer behavior on your site,
such as URLs that they click. You might design the table with hash and
range type primary key with Customer ID as the hash attribute and
date/time as the range attribute. In this application, customer data
grows indefinitely over time; however, the applications might show
uneven access pattern across all the items in the table where the
latest customer data is more relevant and your application might
access the latest items more frequently and as time passes these items
are less accessed, eventually the older items are rarely accessed. If
this is a known access pattern, you could take it into consideration
when designing your table schema. Instead of storing all items in a
single table, you could use multiple tables to store these items. For
example, you could create tables to store monthly or weekly data. For
the table storing data from the latest month or week, where data
access rate is high, request higher throughput and for tables storing
older data, you could dial down the throughput and save on resources.
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.
Source:
http://docs.aws.amazon.com/amazondynamodb/latest/developerguide/GuidelinesForTables.html#GuidelinesForTables.TimeSeriesDataAccessPatterns
For example, You could have your tables segmented by month:
Notifications_April, Notifications_May, etc
Q: I would like to be able to query for the most recent X notifications for a given user.
A: I would suggest using the Query operation and querying using only the Hash Key (UserId) having the Range Key to sort the notifications by the Timestamp (Date and Time).
Hash Key: UserId
Range Key: Timestamp
Note: A better solution would be the Hash Key to not only have the UserId but also another concatenated information that you could calculate before querying to make sure your Hash Key grants you even access patterns to your data. For example, you can start to have hot partitions if notifications from specific users are more accessed than others... having an additional information in the Hash Key would mitigate this risk.
Q: I'd like to fetch the number of unread notifications for a particular user.
A: Create a Global Secondary Index as a Sparse Index having the UserId as the Hash Key and Unread as the Range Key.
Example:
Index Name: Notifications_April_Unread
Hash Key: UserId
Range Key : Unuread
When you query this index by Hash Key (UserId) you would automatically have all unread notifications with no unnecessary scans through notifications which are not relevant to this case. Keep in mind that the original Primary Key from the table is automatically projected into the index, so in case you need to get more information about the notification you can always resort to those attributes to perform a GetItem or BatchGetItem on the original table.
Note: You can explore the idea of using different attributes other than the 'Unread' flag, the important thing is to keep in mind that a Sparse Index can help you on this Use Case (more details below).
Detailed Explanation:
I would have a sparse index to make sure that you can query a reduced dataset to do the count. In your case you can have an attribute "unread" to flag if the notification was read or not, and use that attribute to create the Sparse Index. When the user reads the notification you simply remove that attribute from the notification so it doesn't show up in the index anymore. Here are some guidelines from the documentation that clearly apply to your scenario:
Take Advantage of Sparse Indexes
For any item in a table, DynamoDB will only write a corresponding
index entry if the index range key
attribute value is present in the item. If the range key attribute
does not appear in every table item, the index is said to be sparse.
[...]
To track open orders, you can create an index on CustomerId (hash) and
IsOpen (range). Only those orders in the table with IsOpen defined
will appear in the index. Your application can then quickly and
efficiently find the orders that are still open by querying the index.
If you had thousands of orders, for example, but only a small number
that are open, the application can query the index and return the
OrderId of each open order. Your application will perform
significantly fewer reads than it would take to scan the entire
CustomerOrders table. [...]
Instead of writing an arbitrary value into the IsOpen attribute, you
can use a different attribute that will result in a useful sort order
in the index. To do this, you can create an OrderOpenDate attribute
and set it to the date on which the order was placed (and still delete
the attribute once the order is fulfilled), and create the OpenOrders
index with the schema CustomerId (hash) and OrderOpenDate (range).
This way when you query your index, the items will be returned in a
more useful sort order.[...]
Such a query can be very efficient, because the number of items in the
index will be significantly fewer than the number of items in the
table. In addition, the fewer table attributes you project into the
index, the fewer read capacity units you will consume from the index.
Source:
http://docs.aws.amazon.com/amazondynamodb/latest/developerguide/GuidelinesForGSI.html#GuidelinesForGSI.SparseIndexes
Find below some references to the operations that you will need to programmatically create and delete tables:
Create Table
http://docs.aws.amazon.com/amazondynamodb/latest/APIReference/API_CreateTable.html
Delete Table
http://docs.aws.amazon.com/amazondynamodb/latest/APIReference/API_DeleteTable.html
I'm an active user of DynamoDB and here is what I would do... Firstly, I'm assuming that you need to access notifications individually (e.g. to mark them as read/seen), in addition to getting the latest notifications by user_id.
Table design:
NotificationsTable
id - Hash key
user_id
timestamp
...
UserNotificationsIndex (Global Secondary Index)
user_id - Hash key
timestamp - Range key
id
When you query the UserNotificationsIndex, you set the user_id of the user whose notifications you want and ScanIndexForward to false, and DynamoDB will return the notification ids for that user in reverse chronological order. You can optionally set a limit on how many results you want returned, or get a max of 1 MB.
With regards to projecting attributes, you'll either have to project the attributes you need into the index, or you can simply project the id and then write "hydrate" functionality in your code that does a look up on each ID and returns the specific fields that you need.
If you really don't like that, here is an alternate solution for you... Set your id as your timestamp. For example, I would use the # of milliseconds since a custom epoch (e.g. Jan 1, 2015). Here is an alternate table design:
NotificationsTable
user_id - Hash key
id/timestamp - Range key
Now you can query the NotificationsTable directly, setting the user_id appropriately and setting ScanIndexForward to false on the sort of the Range key. Of course, this assumes that you won't have a collision where a user gets 2 notifications in the same millisecond. This should be unlikely, but I don't know the scale of your system.