Skip to content

[https://nvbugs/6342844][fix] Prevent disaggregated KV transfer stalls #15737

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
chienchunhung merged 4 commits into from
Jul 10, 2026
Merged

[https://nvbugs/6342844][fix] Prevent disaggregated KV transfer stalls #15737

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
3cc5c5b
[https://nvbugs/6342844][fix] Prevent disaggregated KV transfer stalls
chienchunhung Jul 7, 2026
17fabf8
[https://nvbugs/6342844][test] Add synchronous KV transfer stress cov…
chienchunhung Jul 7, 2026
30fcf36
[https://nvbugs/6342844][test] Update disaggregated serving qualifica…
chienchunhung Jul 7, 2026
be8dcdc
[https://nvbugs/6342844][fix] Drain synchronous KV transfer batches
chienchunhung Jul 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 19 additions & 0 deletions cpp/tensorrt_llm/batch_manager/cacheTransceiver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -600,10 +600,29 @@ void CacheTransceiver::respondAndSendLayerWise(
void CacheTransceiver::requestAndReceiveSync(std::shared_ptr<LlmRequest> llmRequest)
{
TLLM_CHECK(llmRequest && llmRequest->isGenerationOnlyRequest());
auto const requestId = llmRequest->mRequestId;
auto const contextRequestId = llmRequest->getContextPhaseParams().value().getReqId();
TLLM_LOG_DEBUG("Synchronous KV cache receive request %zu, context request %zu waiting for native completion.",
requestId, contextRequestId);
try
{
auto future = mCacheReceiver->receiveAsync(llmRequest);
future.get();
}
catch (std::exception const& err)
{
llmRequest->setState(LlmRequestState::kDISAGG_TRANS_ERROR);
llmRequest->setKvCacheTransferEnd(LlmRequest::getSteadyClockNow());
TLLM_LOG_ERROR("Synchronous KV cache receive request %zu, context request %zu failed: %s", requestId,
contextRequestId, err.what());
return;
}
if (llmRequest->getState() == LlmRequestState::kDISAGG_TRANS_ERROR)
{
TLLM_LOG_ERROR("Synchronous KV cache receive request %zu, context request %zu completed with an error state.",
requestId, contextRequestId);
return;
}
llmRequest->setState(LlmRequestState::kDISAGG_GENERATION_TRANS_COMPLETE);
}

Expand Down
Loading
Loading