1

The save API of dynamodb provides multiple save-behaviors including UPDATE, UPDATE_SKIP_NULL_ATTRIBUTES, CLOBBER, and APPEND_SET.

According to this doc Optimistic Locking via version attribute is enabled if save-behavior is UPDATE and disabled if it is CLOBBER. Two questions regarding this:

  1. Why would someone want to disable Optimistic Locking? Optimistic Locking is giving you safe writes to your records!
  2. What about the other save-behaviors? Does it even make sense to use UPDATE_SKIP_NULL_ATTRIBUTES and APPEND_SET with a version attribute? Reason being, you could simply skip the version attribute while calling save() which is a bit hard to digest.
Improve this question

1 Answer 1

Reset to default
1

I'll try to answer each of your questions:

  1. Why would someone want to disable Optimistic Locking? Optimistic Locking is giving you safe writes to your records!

True - optimistic concurrency gives you safe access to your records, but it's not free(*), and you have to decide what to do if an update fails. On the other hand if the table is structured such that there is no possibility for concurrent updates (for instance, you have a table where items are only ever written once, each with a unique key, then why take the extra cost of the optimistic lock?)

(*) an UPDATE with optimistic concurrency requires you to first read the record, then try to update it supplying the version you've just read which means that it is more expensive (you have to pay for a read and a write, potentially multiple reads and write if there is contention) and takes longer (again you have to read and only after that you can attempt to update). If you have another way to ensure that only one writer may be able to update the record, there is no need to use optimistic concurrency.

  1. What about the other save-behaviors? Does it even make sense to use UPDATE_SKIP_NULL_ATTRIBUTES and APPEND_SET with a version attribute? Reason being, you could simply skip the version attribute while calling save() which is a bit hard to digest.

The UPDATE_SKIP_NULL_ATTRIBUTES is actually pretty powerful. It enables you to only apply partial updates to an item. Let's say that you have a complex item that stores state for multiple components of a system, all in the same record. When one of the components changes state, you can simply apply an update with UPDATE_SKIP_NULL_ATTRIBUTES without having to worry about accidentally modifying the other attributes.

The APPEND_SET mode is similar to UPDATE_SKIP_NULL_ATTRIBUTES but it changes the behavior of updates to attributes stored as sets from the default of overwriting the whole set to appending to the set. Personally I haven't found it super useful but I can image it could be useful.

The final question is, does it even make sense to use something like UPDATE_SKIP_NULL_ATTRIBUTES with optimistic concurrency

I think it kind of depends on how you architect your system. It is conceivable to have a table, with GSIs that projects parts of each item, with a key and the version attributes. Then, you could read from the GSI only what you need, and then if you would like to apply a partial update you can do so optimistically.

Improve this answer
Sign up to request clarification or add additional context in comments.

4 Comments

"On the other hand if the table is structured such that there is no possibility for concurrent updates" In this case you would probably not define a version at all. CLOBBER disables optimistic locking if a version is already defined which kind of baffles me.
True, but not defining a version at all is essentially the same thing as CLOBBER. The one thing I would say is if you are still confused, try reading through the DynamoDB document API which is the underlying API that the mapper was build upon as a sort if ORN
@MikeDinescu Does this mean, if I want to have a case where I read a value, and update in Dynamo and it throws a ConditionUpdatefailure because of mis-match in version number, I'll need to write code to retry till it succeeds? Or is there something provided by Dynamo which already does that?
Correct - if the operation fails due to a ConditionalCheckFailed error then you have to write your logic to retry which will involve reding the item and fetrying

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.