Clarify message idempotency section #4144
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
📖 Docs PR preview links |
dandavison
force-pushed
the
message-idempotency
branch
from
4747f02 to
4d64af2
Compare
dandavison
force-pushed
the
message-idempotency
branch
from
d87ed40 to
6e961f5
Compare
| However, if you are using Updates with [Continue-As-New](/workflow-execution/continue-as-new) you should implement the deduplication in your Workflow code, since Update ID deduplication by the server is per Workflow run. | ||
|
|
||
| To deduplicate in your message handler, you can use an idempotency key. | ||
| (In addition to these application-level identifiers, both Signals and Updates automatically use request IDs to deduplicate retried client calls. You do not need to do anything to enable this.) |
There was a problem hiding this comment.
Don't mind dropping the parentheses.
There was a problem hiding this comment.
On second thought, if the user can't change/do anything about it, do we need to explain this here at all?
There was a problem hiding this comment.
I have the same question.
There was a problem hiding this comment.
So, the two topics are deduplication of framework-level retries, and deduplication of messages that are duplicates according to an application-domain definition of equivalence.
We could get rid of it. The reason I put it there is because I have personally confused them in the past, which leads me to suspect others will. Put differently, whenever request IDs are discussed, idempotency is also. If we tell people that they are on the hook for signal idempotency then I think many people will confuse that with a statement that Temporal doesn't use request IDs for signal.
| #### Message IDs and handling Continue-As-New {#exactly-once-message-processing} | ||
|
|
||
| Many developers want their message handlers to run exactly once--to be idempotent--in cases where the same Signal or Update is delivered twice or sent by two different call sites. Temporal deduplicates messages for you on the server, but there is one important case when you need to think about this yourself when authoring a Workflow, and one when sending Signals and Updates. | ||
| Usually, you'll want your message handlers to run exactly once--to be idempotent--in cases where the same Signal or Update is delivered twice. For Updates, Temporal handles this for you on the server, by deduplicating according to the Update ID. For Signals, you should use a custom idempotency key that you send as part of your own signal inputs, implementing the deduplication in your Workflow code. The Update ID is set automatically to a UUID, but you can set it yourself. |
There was a problem hiding this comment.
I'd move the last sentence about UUID next to the Update sentence. The Signal sentence breaks the flow there in the middle.
There was a problem hiding this comment.
It may be simpler to just ask people to check idempotency. And then explain the reasoning that it's necessary for signals and update plus CaN .
That way, if they add CaN later, they won't cause a bug. And I don't think there is much harm to a redundant check.
| See the links below for how to ensure handlers are finished in your SDK. | ||
|
|
||
| #### Ensuring your messages are processed exactly once {#exactly-once-message-processing} | ||
| #### Message IDs and handling Continue-As-New {#exactly-once-message-processing} |
There was a problem hiding this comment.
I like that this is focused on CaN now. The part about the same Update/Signal coming in from different clients is now only implied. That might be fine.
What does this PR do?
Clarifies the section on idempotency keys and CAN. There were reports that for the following sentence
it was not obvious that it referred to the situation when the workflow is using CAN.