I am designing a simple messaging service using ASP.NET MVC / Windows Azure Table Storage. I have two kinds of entities - messages and message threads. Relation between them is simple - each thread can have multiple messages but the message can only be assigned to one thread.
Table storage is not a relational DB, so representing relations is always a bit tricky. I need to decide between 2 approaches:
Having one big table for threads and one for messages. And having threadId as a partition key of message entity so that messages are partitioned by threads.
Dynamically creating a special table for each message thread and having threadId as a name of the table.
I tend to prefer the second because it fits better into architecture of the rest of the service. But there will obviously be large number of tables created in a storage account.
Do you think this may be a problem?
You could also consider having just one table, that stores both Thread and Message entities. This would give you transaction support, and you could use Lucifure's hybrid approach on this table.
Creating a large number of tables may be an issue, depending on how you want to manage them. The underlying REST API for listing tables works like a query for table entities. It only returns the first 1000 tables, after that you have to use a continuation token. All of the storage explorers I've seen don't allow you to query tables based on name, they simply like the first 1000 tables. If you end up with 20000 threads, it could take you a while to get to the table you want.
One way you could mitigate this is to put your message table in its own storage account. This way your storage account with all of your other tables won't get crowded out by all of these dynamic tables that you will be creating and possibly deleting.
Deleting is actually one of the ways in which using a separate table for each thread would be easier. To delete all of the related messages you simply have to delete one table rather than iterating over each message and deleting it.
Everything else however will be more complicated than keeping all of the messages in one table. If this is core functionality to your app and you can dedicate enough time to develop it this way, one table per thread is probably a good idea. Otherwise the easy way to do things is with one big table.
You may consider a hybrid approach to keep the number of tables to a manageable level, depending on your scalability needs.
My experience has been that date based partitioning at the table level is a very effective approach and can be leverage across the board.
For example you could partition tables based on date and with a granularity of day or month. So a table name like “Thread201202” could be used for all threads started in February 2012.
Your thread id would implicitly include the “201202” and be something like “201202-myid01” although you would not need to explicitly store it in the partition key since it would be implied in the table name.
Aged threads could then be easily disposed by deleting tables say more than a year old.
Related
I am working on an asset tracking system that also manages the concept of "projects". The users of this application perform maintenance activities on their customer's assets, so they need an action log where actions on an asset start life as a task in a project. For example, "Fix broken frame" might be a task where an action would have something like "Used parts a, b, and c to fix the frame" with a completed time and the employee who performed the action.
The conceptual data model for the application starts with a Customer that has multiple locations and each location has multiple assets. Each asset should have an associated action log so it is easy to view previous actions applied to that asset.
To me, that should all go in one table based upon the logical ownership of that data. Customer owns Locations which own Assets which own Actions.
I believe I should have a second table for projects as this data is tangential to the Customer/Location/Asset data. However, because I read so much about how it should all be one table, I'm not sure if this delineation only exists because I've modeled the data incorrectly because I can't get over the 3NF modeling that I've used for my entire career.
Single table design doesn't forbid you to create multiple tables. Instead in encourages to use only a single table per micro-services (meaning, store correlated data, which you want to access together, in the same table).
Let's look at some anecdotes from experts:
Rick Houlihan tweeted over a year ago
Using a table per entity in DynamoDB is like deploying a new server for each table in RDBMS. Nobody does that. As soon as you segregate items across tables you can no longer group them on a GSI. Instead you must query each table to get related items. This is slow and expensive.
Alex DeBrie responded to a tweet last August
Think of it as one table per service, not across your whole architecture. Each service should own its own table, just like with other databases. The key around single table is more about not requiring a table per entity like in an RDBMS.
Based on this, you should answer to yourself ...
How related is the data?
If you'd build using a relational database, would you store it in separate databases?
Are those actually 2 separate micro services, or is it part of the same micro service?
...
Based on the answers to those (and similar) questions you can argue to either keep it in one table, or to split it across 2 tables.
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 am planning to create a website using ASP.NET and SQL Server. However, my plan for the database design leaves me wondering if there is a better way.
The website will serve as a repository of information for various users. I figure I would have two databases, a Membership and Profile database.
The profile database would contain user data for all users, where each user may have ~20 tables. I would create the tables when the user account is created and generate a key used to name the tables. The tables are not directly related.
For Example a set of tables for two different users could look like:
User1 Tables - TransactionTable_Key1, AssetTable_Key1, ResearchTable_Key1 ....;
User2 Tables - TransactionTable_Key2, AssetTable_Key2, ResearchTable_Key2 ....;
The Key1, Key2 etc.. values would be retrieved based on the MembershipID data when the account was created. This could result in a very large number of tables over time. I'm not sure if this will limit scalability by setting up the database in this way. Any recommendations?
Edit: I should mention that some of these tables would contain 20k+ rows.
Realistically it sounds like you only really need one database for this.
From the way you worded your question, it sounds like you're trying to dynamically create tables for users as they create accounts. I wouldn't recommend this method.
What you want to do is create a master table that contains a primary key for each individual user. I'm assuming this is the Membership table. Then create the ~20 tables that you need for the profiles of these members. Every record, no matter the number of users that you have, will go into these tables. These 20 tables would need to have a foreign key pointing to the unique identifier of the Membership table.
When you want to query a Member for their user information, just select from the tables where the membership table's primary Id matches the foreign key in the profile tables.
This would result in only a few tables in the end and is easily maintainable and follows better database design.
Your ORM layer (EF, LINQ, DAL code) will hate having to deal with one set of tables per tenant. It is much better to have either one set of tables for all tenant in a single database, or a separate database per tenant. The later is only better if schema upgrade has to be vetted by tenant (like Salesforce.com has). If you can afford to upgrade all tenant to a new schema at once then there is no reason for database per tenant.
When you design a schema that hold multiple tenant the important things to remember are
don't use heaps, all tables must be clustered index
add the tenant ID as the leftmost key to every clustered
add the tenant ID as the leftmost key to every non-clustered index too
add the Left.tenantID = right.tenantID predicate to every join
add the table.TenantID = #currentTenantID to every query
These are fairly simple rules and if you obey them (with no exceptions) you will get a perfect partitioning per tenant of every query (no query will ever ever scan rows in a range of a different tenant) so you eliminate contention between tenants. To be more through, you can disable lock escalation to make sure no tenant escalates to block every other tenant.
This design also lends itself to table partitioning and to sharing the database for scale-out.
You definitely don't want to create a set of tables for each user, and you would want these only in one database. Even with SQL Server 2008's large capacity for tables (note really total objects in database), it would quickly become unmanageable. Your best bet is to use 20 tables, and separate them via a column into user areas. You might consider partitioning the tables by this user value, but that should be tested for performance reasons too.
Yes, since the tables only contain id, key, and value, why not make one single table?
Have the columns:
id, user ID, key, value
Put an Index on the user ID field.
A key idea behind a relational database is that the table structure does not change. You create a solid set of tables, and these are the "bones" of your application.
Cheers,
Daniel
Neal,
The solution really depends on your requirement. If security and data access are concern and you have only a handful of users, you can set up a different db for each user with access for him set to only his/her database.
Other wise, what Daniel Williams suggested is a good alternative where you have one DB and tables laid out with a indexed column partitioning the users data rows.
It's hard to tell from the summary, but it looks like you are designing for dynamic attribution by user. This design approach is called EAV (Entity-Attribute-Value) and consists of a simple base collection key (UserID, SiteID, ProductID...) and then rows consisting of name/value pairs. In a more complex version, categories are sometimes added as "super columns" to the tuple/row and provide sub-groupings for a set of name/value pairs.
Designing in this way moves responsibility for data type integrity, relational integrity and tuple integrity to the application layer.
The risk with doing this in a relational system involves the breaking of the tuple or row into a set of rows. Updates, deletes, missing values and the definition of a tuple are no longer easily accessible through human interaction. As your application evolves and the definition of a tuple changes, it becomes almost impossible to tell if a name/value pair is missing because it's part of an earlier-version tuple or because it was unintentionally deleted. Ad-hoc research as well becomes harder to manage as business analysts must keep an understanding of the virtual structure either in their heads or in documentation provided.
If you are looking to implement an EAV model, I would suggest you look at a non-relational solution (nosql) like MongoDB or CouchDB. These stores allow a developer to save and retrieve "documents" or json-formatted messages that are essentially made up of a collection of name/value pairs and can look very much like a serialized object. The advantage here is that you can store dynamic attribution without breaking your tuple. You always know that you have a complete tuple because you can store and retrieve it as a single "blob" of information that can be serialized and deserialized at-will. You can also update single attributes within the tuple, if that's a concern.
MongoDB also provides some database-like features such as multiple-attribute indexes, a query engine that is robust in comparison to other similar non-relational offerings and a sharding solution that is much less trouble than trying to do it with MySQL.
I hope this helps.