Claim DO retry tokens after actor construction - #7497
Merged
Merged
Conversation
Contributor
|
LGTM |
apeacock1991
force-pushed
the
apeacock/actor-claim-after-construction
branch
2 times, most recently
from
September 24, 2026 15:50
e081705 to
eb51a8d
Compare
jqmmes
approved these changes
Sep 24, 2026
Durable Object fetch and JSRPC claimed the retry token before delivering the event, so the claim ran before the actor constructor. The claim now runs after construction and immediately before handler or method lookup. A later @retryable decorator can then inspect the constructed actor's handler or method and skip the claim. Fetch claims inside the context.run() lambda in WorkerEntrypoint::requestImpl(), after getExportedHandler() and before the handler runs. It uses the incoming request's own metrics. IoContext::getMetrics() returns the front incoming request, which can be a different request to the same actor. The plan put this claim in global-scope.c++. requestImpl() is simpler and has the right metrics. JSRPC claims in the callImpl() dispatch lambda, after getTargetInfo() and before tryGetProperty(), through a new JsRpcTargetBase::maybeClaimRetryToken() hook. EntrypointJsRpcTarget claims once per session. A rejection is sticky, so a later top-level call rethrows it. Stubs returned by a call do not claim. JsRpcSessionCustomEvent::failed() turns a claim rejection into a disconnect, so calls pipelined on the rejected call don't surface it as an application error. An in-process rejection now fails the call instead of the session, so maybeClaimRetryToken() applies the same translation. Both use disconnectRetryClaimRejection(). Claim and predecessor rejections skip logUncaughtExceptionAsync, and an output-gate failure no longer replaces them. Predecessor rejections stay marked as not delivered. Behaviour changes: - A JSRPC session with no calls no longer claims. The constructor may still run, but no requested method does. - While JSRPC claims are only observed, callers see no difference. Once enforcement is on, a rejected duplicate that lands on a fresh instance runs the constructor before it is rejected. - An in-process JSRPC claim rejection now also carries REQUEST_DELIVERED_TO_ACTOR_DETAIL_ID. The sender checks ACTOR_RETRY_CLAIM_REJECTED_DETAIL_ID first, so retries are unaffected.
apeacock1991
force-pushed
the
apeacock/actor-claim-after-construction
branch
from
September 24, 2026 18:43
eb51a8d to
354a44c
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Durable Object fetch and JSRPC calls claimed their retry token before the event was delivered, so the claim ran before the actor constructor. This moves the claim to after construction, immediately before the handler or method is looked up. A later
@retryabledecorator on a fetch handler or RPC method can then skip the claim.Fetch
The claim now runs inside the
context.run()lambda inWorkerEntrypoint::requestImpl(), aftergetExportedHandler()and before the handler. It uses the incoming request's own metrics.IoContext::getMetrics()returns the front incoming request, which can be a different request to the same actor.Claim and predecessor rejections no longer go through
logUncaughtExceptionAsync, and an output-gate failure no longer replaces them. Predecessor rejections stay marked as not delivered.JSRPC
JsRpcTargetBasehas a newmaybeClaimRetryToken()hook.callImpl()calls it aftergetTargetInfo()and beforetryGetProperty(), so no getter runs before the claim.EntrypointJsRpcTargetclaims once per session. If the claim is rejected, later top-level calls in that session rethrow the same rejection. Stubs returned from a call don't claim.The old claim in
JsRpcSessionCustomEvent::run()is gone.JsRpcSessionCustomEvent::failed()turns a claim rejection into a disconnect, so calls pipelined on the rejected call don't see it as an application error. An in-process rejection now fails the call rather than the session, somaybeClaimRetryToken()applies the same translation. Both sharedisconnectRetryClaimRejection().Behaviour changes
REQUEST_DELIVERED_TO_ACTOR_DETAIL_ID. The sender checksACTOR_RETRY_CLAIM_REJECTED_DETAIL_IDfirst, so retry decisions don't change.