mirror of
https://github.com/kvcache-ai/Mooncake.git
synced 2026-11-05 01:17:55 +00:00
[GH-ISSUE #682] [Bug]: TcpTransport::startTransfer encountered an ASIO exception.Exception: connect: Connection timed out #221
Labels
No labels
ascend
auto-closed
bug
bug
CI/Build
Common
documentation
enhancement
feature request
good first issue
help wanted
high priority
MetaX
nightly-failure
pull-request
PyTorch Backend
question
RFC
Roadmap
stale
Store
Tests
Transfer Engine
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
kvcache-ai/Mooncake#221
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @Marovlo on GitHub (Jul 28, 2025).
Original GitHub issue: https://github.com/kvcache-ai/Mooncake/issues/682
Bug Report
question
I have two server called as 16 (30.221.162.16) and 198 (30.221.161.198) in short for vllm+lmcache+mooncakeStore PD disaggregation. The mooncake_master and mooncake_http_metadata_server are running at 16. During experiment when decoder try to use store.get_buffer() to get kv_cache, it always encounters a TimeoutError, while the store.is_exists() operation returns True successfully before store.get_buffer(), which means that decoder can't establish connection with prefiller using mooncake.
I write a simple example to reproduce the problem as below:
In server 16 (prefiller-test)
In server 198 (decoder-test)
outputs
master
metadata_server
prefiller-test
decoder-test
I am wondering whether the multi net interface is the cause. As we can see that in the prefiller-test output, which says the "Transfer Engine RPC using new RPC mapping, listening on 30.221.239.158:16137" that is interface "bond8", but I prefer to use "bond1" for both prefiller(30.221.162.16) and deocder(30.221.161.198).
I don't kown how it works specificlly, could you please help me out? Thanks a lot.
Before submitting...
@xiaguan commented on GitHub (Jul 28, 2025):
maybe you could try set
MC_TCP_BIND_ADDRESS@Marovlo commented on GitHub (Jul 28, 2025):
thanks! This helps a lot.
@xiaguan commented on GitHub (Jul 28, 2025):
By the way, we recommend using RDMA for optimal performance.
@Marovlo commented on GitHub (Jul 28, 2025):
Yeah,we should try rdma once the whole process can be completed by using tcp as an example. Thanks again!