Skip to content

fix(sdk): show interrupts raised after a passive thread rejoin - #2813

Merged
Elior Nataf Lackritz (eliornl) merged 3 commits into
mainfrom
fix/sdk-passive-rejoin-interrupts
Sep 9, 2026
Merged

Elior Nataf Lackritz (eliornl) merged 3 commits into
mainfrom
fix/sdk-passive-rejoin-interrupts

Conversation

@eliornl

@eliornl Elior Nataf Lackritz (eliornl) commented Sep 9, 2026

Copy link
Copy Markdown
Contributor

Summary

Fixes LG-661. After a page refresh (or a second tab) while a run is in progress, an interrupt the run then raises never shows up in useStream until the next reload. isLoading still drops to false, so the agent looks stuck. Reproduced on a real server with a slow-then-interrupt graph: the original tab shows the interrupt, the refreshed tab shows interrupts: [].

Why it happens

After hydrate, the controller treats every input.requested it does not already know from state.tasks[].interrupts as replayed history, because the SSE replays the whole thread buffer on connect and the stream has no marker for where replay ends. #2807 parks those unknown interrupts until a checkpoints event carrying the hydrated checkpoint id arrives, then releases everything newer.

That event never arrives:

  • LangGraph Python emits the checkpoints stream mode as a get_state()-shaped snapshot. The server session only forwards a checkpoint envelope with a top-level string id, so nothing is forwarded. Zero checkpoints events on the wire in the reproduction, matching the customer's HAR.
  • LangGraph.js has no checkpoints emission at all (libs/langgraph/src).
  • The Redis replay buffer trims by age (120s), so on a long node the matching event is gone before the refresh even when it did exist.

So the parked interrupts wait forever.

The fix

Decide parked interrupts with the server's thread state instead of a stream marker.

  • Unknown interrupts are parked as before, now whether or not hydrate returned a checkpoint id (the release no longer needs it).
  • When a root terminal lifecycle (interrupted / completed / failed) arrives with interrupts parked, #settleParkedInterrupts runs the existing #reconcilePendingInterruptsFromServer: ids the server lists as pending are shown, the rest are dropped as history. Payloads come from state.tasks[].interrupts, the same source hydrate uses, so a refreshed tab and a fresh load agree by construction.
  • The server commits the interrupt to thread state slightly after it streams the event (the server has its own 5s settle poll for this), so the fetch retries every 500ms for up to 5s. One loop runs at a time, so a replay burst with many old terminals costs one loop, not one per event. The common path (no parked interrupts) makes no extra calls.
  • Rebased on revert(sdk): appropriately track persisted seq for stream replay (#2807) #2814, which reverts the checkpoints barrier from fix(sdk): appropriately track persisted seq for stream replay #2807. This PR carries the parking list itself; the barrier is gone, so the settle against server state is the only release path.

Tests

controller.test.ts:

  • shows an interrupt raised after a passive rejoin without a checkpoints event
  • waits for the server to commit the interrupt before showing it (second fetch returns it)
  • drops replayed interrupts the server no longer lists

All three fail on main. The two #2807 tests still pass. Full libs/sdk unit suite (794) and libs/sdk-react browser suite (244) pass.

Real-stack check (langgraph-api inmem, langgraph 1.2.6): submit a run, create a second StreamController on the same thread 1s in, let the run interrupt at 4s. Before: refreshed tab interrupts: []. After: refreshed tab shows the interrupt, respond() from that tab resumes the run, and it completes.

Follow-ups (not in this PR)

  • The checkpoints channel is dead on the wire for every current deployment; the server never forwards it because no runtime produces the {id, parent_id, step, source} envelope.
  • onCompleted not firing after a refresh (also reported in the thread): #runLifecycleListener requires isLoading to be true, and the buffer's running event is trimmed on long runs.

Changeset: @langchain/langgraph-sdk patch.

@changeset-bot

changeset-bot Bot commented Sep 9, 2026

Copy link
Copy Markdown

🦋 Changeset detected

Latest commit: 105ba45

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 5 packages
Name Type
@langchain/langgraph-sdk Patch
@langchain/angular Patch
@langchain/react Patch
@langchain/svelte Patch
@langchain/vue Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

@pkg-pr-new

pkg-pr-new Bot commented Sep 9, 2026

Copy link
Copy Markdown

Open in StackBlitz

@langchain/langgraph-checkpoint

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint@2813

@langchain/langgraph-checkpoint-mongodb

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint-mongodb@2813

@langchain/langgraph-checkpoint-postgres

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint-postgres@2813

@langchain/langgraph-checkpoint-redis

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint-redis@2813

@langchain/langgraph-checkpoint-sqlite

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint-sqlite@2813

@langchain/langgraph-checkpoint-validation

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-checkpoint-validation@2813

create-langgraph

npm i https://pkg.pr.new/langchain-ai/langgraphjs/create-langgraph@2813

@langchain/langgraph-api

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-api@2813

@langchain/langgraph-cli

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-cli@2813

@langchain/langgraph

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph@2813

@langchain/langgraph-cua

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-cua@2813

@langchain/langgraph-supervisor

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-supervisor@2813

@langchain/langgraph-swarm

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-swarm@2813

@langchain/langgraph-ui

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-ui@2813

@langchain/langgraph-sdk

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/langgraph-sdk@2813

@langchain/angular

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/angular@2813

@langchain/react

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/react@2813

@langchain/svelte

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/svelte@2813

@langchain/vue

npm i https://pkg.pr.new/langchain-ai/langgraphjs/@langchain/vue@2813

commit: 105ba45

@open-swe open-swe Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Open SWE Review found 1 potential issue.

Open in WebView Open SWE trace

Comment thread libs/sdk/src/stream/controller.ts Outdated
After a page refresh mid-run, the controller treated every input.requested
it did not already know from the hydrated thread state as replayed history
and dropped it. The SSE replays the thread buffer on connect with no marker
for where the replay ends, so the interrupt the run then raised never
reached the UI until the next reload, while isLoading still dropped to
false.

Unknown interrupts are now parked, and settled against the server's thread
state when a root terminal lifecycle arrives: ids the server lists in
tasks[].interrupts are shown, the rest are dropped as history. The state is
written slightly after the event streams, so the fetch retries every 500ms
for up to 5s, coalesced across a replay burst. A failed fetch keeps
retrying until the deadline; only disposal, a thread swap, or a new submit
stop the loop.
@eliornl
Elior Nataf Lackritz (eliornl) merged commit 4fc118f into main Sep 9, 2026
30 checks passed
@eliornl
Elior Nataf Lackritz (eliornl) deleted the fix/sdk-passive-rejoin-interrupts branch September 9, 2026 17:38
Hunter Lovell (hntrl) pushed a commit that referenced this pull request Sep 9, 2026
This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to main, this PR will
be updated.

⚠️⚠️⚠️⚠️⚠️⚠️

`main` is currently in **pre mode** so this branch has prereleases
rather than normal releases. If you want to exit prereleases, run
`changeset pre exit` on `main`.

⚠️⚠️⚠️⚠️⚠️⚠️

# Releases
## @langchain/langgraph-sdk@1.10.3-rc.2

### Patch Changes

- [#2813](#2813)
[`4fc118f`](4fc118f)
Thanks [@eliornl](https://github.com/eliornl)! - fix(sdk): show
interrupts raised after a passive thread rejoin
  
After a page refresh mid-run, `useStream` filtered every interrupt it
did not
already know from the hydrated thread state as replayed history, and
waited
for a `checkpoints` event to lift that filter. Current runtimes never
emit
that event and the replay buffer trims it on long runs, so interrupts
raised
after the refresh never appeared until the next reload. Unknown
interrupts are
now settled against the server's thread state when the run reaches a
terminal
lifecycle: the ones the server lists as pending are shown, the rest are
  dropped as history.

- [#2812](#2812)
[`db4bdad`](db4bdad)
Thanks [@eliornl](https://github.com/eliornl)! - fix(sdk): recover from
a server-side thread stream drop instead of freezing
  
The protocol SSE transport now reconnects when the server closes the
event
stream cleanly. The thread stream is open-ended, so a clean close only
happens
when the server's own upstream consumer died or it is restarting;
before, the
client treated it as the end of the thread and the UI froze mid-run with
no
error. A connection that delivered events also resets the reconnect
budget, so
long-lived pages survive repeated deploys. `maxReconnectAttempts: 0`
keeps the
  old end-on-close behavior.
  
Unsolicited server error frames (no command id) and a shared stream that
gives
up reconnecting now reach `stream.error`: `ThreadStream.onError` exposes
them,
`useStream` sets `error`, clears `isLoading`, and settles the in-flight
  `submit()` as failed.
## @langchain/angular@1.0.36-rc.2

### Patch Changes

- Updated dependencies
[[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad)]:
  - @langchain/langgraph-sdk@1.10.3-rc.2
## @langchain/react@1.0.36-rc.2

### Patch Changes

- Updated dependencies
[[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad)]:
  - @langchain/langgraph-sdk@1.10.3-rc.2
## @langchain/svelte@1.0.36-rc.2

### Patch Changes

- Updated dependencies
[[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad)]:
  - @langchain/langgraph-sdk@1.10.3-rc.2
## @langchain/vue@1.0.36-rc.2

### Patch Changes

- Updated dependencies
[[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad)]:
  - @langchain/langgraph-sdk@1.10.3-rc.2

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Hunter Lovell (hntrl) pushed a commit that referenced this pull request Sep 12, 2026
This PR was opened by the [Changesets
release](https://github.com/changesets/action) GitHub action. When
you're ready to do a release, you can merge this and the packages will
be published to npm automatically. If you're not ready to do a release
yet, that's fine, whenever you add more changesets to main, this PR will
be updated.


# Releases
## @langchain/langgraph-sdk@1.11.0

### Minor Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Expose
connection lifecycle callbacks for built-in streaming transports.
  
`onConnected` runs after the initial SSE or WebSocket connection becomes
usable and after every successful reconnect. Its payload distinguishes
an `initial` connection from a `reconnected` connection and includes the
reconnect attempt number.
  
`onReconnect` now also receives the scheduled `delayMs`, allowing
applications to display accurate retry state before the next connection
attempt. React, Vue, Svelte, and Angular stream hooks forward both
callbacks.

### Patch Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Recover
cleanly when a thread's root event stream terminates unexpectedly.
  
If the stream fails or closes while a run is active, the controller now
records the transport error and clears `isLoading` rather than leaving
the UI in a permanently running state. Once the failed pump settles, a
later submission can start a fresh root subscription without recreating
the thread stream or replaying the failed command.

- [#2809](#2809)
[`11a4535`](11a4535)
Thanks [@hntrl](https://github.com/hntrl)! - fix(sdk): appropriately
track persisted seq for stream replay
  
Sequences weren't being appropriately attributed when rehydrating the
page (e.g. on refresh). This meant we'd lose stream information on
`useStream` on reloads. This has been fixed by adding a lookup step to
determine what the most appropriate sequence index is to track in the
event stream.:x

- [#2762](#2762)
[`2fab6fd`](2fab6fd)
Thanks [@JessYanCoding](https://github.com/JessYanCoding)! - Send
`checkpoint_id` in the `runs.stream()` request body, so a `checkpointId`
passed to `client.runs.stream()` forks from the requested checkpoint
instead of being silently dropped. Matches `runs.create()` and
`runs.wait()`, which already send it.

- [#2813](#2813)
[`4fc118f`](4fc118f)
Thanks [@eliornl](https://github.com/eliornl)! - fix(sdk): show
interrupts raised after a passive thread rejoin
  
After a page refresh mid-run, `useStream` filtered every interrupt it
did not
already know from the hydrated thread state as replayed history, and
waited
for a `checkpoints` event to lift that filter. Current runtimes never
emit
that event and the replay buffer trims it on long runs, so interrupts
raised
after the refresh never appeared until the next reload. Unknown
interrupts are
now settled against the server's thread state when the run reaches a
terminal
lifecycle: the ones the server lists as pending are shown, the rest are
  dropped as history.

- [#2812](#2812)
[`db4bdad`](db4bdad)
Thanks [@eliornl](https://github.com/eliornl)! - fix(sdk): recover from
a server-side thread stream drop instead of freezing
  
The protocol SSE transport now reconnects when the server closes the
event
stream cleanly. The thread stream is open-ended, so a clean close only
happens
when the server's own upstream consumer died or it is restarting;
before, the
client treated it as the end of the thread and the UI froze mid-run with
no
error. A connection that delivered events also resets the reconnect
budget, so
long-lived pages survive repeated deploys. `maxReconnectAttempts: 0`
keeps the
  old end-on-close behavior.
  
Unsolicited server error frames (no command id) and a shared stream that
gives
up reconnecting now reach `stream.error`: `ThreadStream.onError` exposes
them,
`useStream` sets `error`, clears `isLoading`, and settles the in-flight
  `submit()` as failed.

- [#2808](#2808)
[`55fa26b`](55fa26b)
Thanks [@hntrl](https://github.com/hntrl)! - fix(sdk): coalesce locally
resolved interrupts
  
when resolving interrupts using `useStream`, there was a case where we
prioritized the remote state values (which we lookup in React Strict
mode on every page transition) over the local interrupt responses we
know we've responded with.
  
This has since been fixed to first prioritize the local cache of
interrupts, resolved against the remote state values when a run hits a
terminal event
## @langchain/angular@1.1.0

### Minor Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Expose
connection lifecycle callbacks for built-in streaming transports.
  
`onConnected` runs after the initial SSE or WebSocket connection becomes
usable and after every successful reconnect. Its payload distinguishes
an `initial` connection from a `reconnected` connection and includes the
reconnect attempt number.
  
`onReconnect` now also receives the scheduled `delayMs`, allowing
applications to display accurate retry state before the next connection
attempt. React, Vue, Svelte, and Angular stream hooks forward both
callbacks.

### Patch Changes

- Updated dependencies
[[`3234c69`](3234c69),
[`3234c69`](3234c69),
[`11a4535`](11a4535),
[`2fab6fd`](2fab6fd),
[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad),
[`55fa26b`](55fa26b)]:
  - @langchain/langgraph-sdk@1.11.0
## @langchain/react@1.1.0

### Minor Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Expose
connection lifecycle callbacks for built-in streaming transports.
  
`onConnected` runs after the initial SSE or WebSocket connection becomes
usable and after every successful reconnect. Its payload distinguishes
an `initial` connection from a `reconnected` connection and includes the
reconnect attempt number.
  
`onReconnect` now also receives the scheduled `delayMs`, allowing
applications to display accurate retry state before the next connection
attempt. React, Vue, Svelte, and Angular stream hooks forward both
callbacks.

### Patch Changes

- Updated dependencies
[[`3234c69`](3234c69),
[`3234c69`](3234c69),
[`11a4535`](11a4535),
[`2fab6fd`](2fab6fd),
[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad),
[`55fa26b`](55fa26b)]:
  - @langchain/langgraph-sdk@1.11.0
## @langchain/svelte@1.1.0

### Minor Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Expose
connection lifecycle callbacks for built-in streaming transports.
  
`onConnected` runs after the initial SSE or WebSocket connection becomes
usable and after every successful reconnect. Its payload distinguishes
an `initial` connection from a `reconnected` connection and includes the
reconnect attempt number.
  
`onReconnect` now also receives the scheduled `delayMs`, allowing
applications to display accurate retry state before the next connection
attempt. React, Vue, Svelte, and Angular stream hooks forward both
callbacks.

### Patch Changes

- Updated dependencies
[[`3234c69`](3234c69),
[`3234c69`](3234c69),
[`11a4535`](11a4535),
[`2fab6fd`](2fab6fd),
[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad),
[`55fa26b`](55fa26b)]:
  - @langchain/langgraph-sdk@1.11.0
## @langchain/vue@1.1.0

### Minor Changes

- [#2831](#2831)
[`3234c69`](3234c69)
Thanks [@ramon-langchain](https://github.com/ramon-langchain)! - Expose
connection lifecycle callbacks for built-in streaming transports.
  
`onConnected` runs after the initial SSE or WebSocket connection becomes
usable and after every successful reconnect. Its payload distinguishes
an `initial` connection from a `reconnected` connection and includes the
reconnect attempt number.
  
`onReconnect` now also receives the scheduled `delayMs`, allowing
applications to display accurate retry state before the next connection
attempt. React, Vue, Svelte, and Angular stream hooks forward both
callbacks.

### Patch Changes

- Updated dependencies
[[`3234c69`](3234c69),
[`3234c69`](3234c69),
[`11a4535`](11a4535),
[`2fab6fd`](2fab6fd),
[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad),
[`55fa26b`](55fa26b)]:
  - @langchain/langgraph-sdk@1.11.0
## @langchain/langgraph-api@1.4.6

### Patch Changes

- [#2751](#2751)
[`0a9611a`](0a9611a)
Thanks [@t3s7r](https://github.com/t3s7r)! - fix(api): `run.start` with
input on a cancelled thread no longer folds the input into
`Command(resume)`. A cancelled run shares the "interrupted" status with
a genuine `interrupt()` pause, but has no pending interrupt to consume
the resume value, so the submitted message was silently dropped. The
input-vs-resume decision now keys on whether the thread actually has
pending interrupts (both in the protocol service and the embed
protocol).
- Updated dependencies []:
  - @langchain/langgraph-ui@1.4.6
## @langchain/langgraph-cli@1.4.6

### Patch Changes

- Updated dependencies
[[`0a9611a`](0a9611a)]:
  - @langchain/langgraph-api@1.4.6
## @langchain/langgraph@1.4.15

### Patch Changes

- [#2794](#2794)
[`83a4b62`](83a4b62)
Thanks [@hntrl](https://github.com/hntrl)! - feat(langgraph): add
per-node `tracePolicy` input/output processors and `omitPayload`
  
Transform the payloads recorded on a node's own trace run while
retaining its span and timing. Processors receive raw values and fall
back to the original payload if they throw. Graph state, root runs, and
child runs remain unchanged when processors do not mutate their
arguments.
  
Matches Python's callback-level behavior: transforms also affect chain
events and message streaming, so omitting outputs can suppress messages
returned directly by nodes and omitting inputs can affect message
deduplication.
- Updated dependencies
[[`3234c69`](3234c69),
[`3234c69`](3234c69),
[`11a4535`](11a4535),
[`2fab6fd`](2fab6fd),
[`4fc118f`](4fc118f),
[`db4bdad`](db4bdad),
[`55fa26b`](55fa26b)]:
  - @langchain/langgraph-sdk@1.11.0
## @langchain/langgraph-cua@1.0.3

No changes in this release.
## @langchain/langgraph-supervisor@1.1.2

No changes in this release.
## @langchain/langgraph-swarm@1.0.3

No changes in this release.
## @langchain/langgraph-ui@1.4.6

No changes in this release.

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants