Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
OpenDAS
vllm_cscc
Commits
699d62e6
Unverified
Commit
699d62e6
authored
Oct 24, 2025
by
Chendi.Xue
Committed by
GitHub
Oct 24, 2025
Browse files
[NIXL][BUGFIX] delay done_recving queue cleanup to bottom of get_finished (#27297)
Signed-off-by:
Chendi Xue
<
chendi.xue@intel.com
>
parent
cd390b60
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
8 additions
and
11 deletions
+8
-11
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
...distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+8
-11
No files found.
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
View file @
699d62e6
...
@@ -1452,11 +1452,17 @@ class NixlConnectorWorker:
...
@@ -1452,11 +1452,17 @@ class NixlConnectorWorker:
len
(
done_recving
),
len
(
done_recving
),
)
)
# clean up metadata for completed requests
block_ids_to_permute
=
[]
for
req_id
in
done_recving
:
for
req_id
in
done_recving
:
# clean up metadata for completed requests
meta
=
self
.
_recving_metadata
.
pop
(
req_id
,
None
)
meta
=
self
.
_recving_metadata
.
pop
(
req_id
,
None
)
if
self
.
use_host_buffer
and
meta
:
assert
meta
is
not
None
,
f
"
{
req_id
}
not found in recving_metadata list"
if
self
.
use_host_buffer
:
self
.
sync_recved_kv_to_device
(
req_id
,
meta
)
self
.
sync_recved_kv_to_device
(
req_id
,
meta
)
if
self
.
enable_permute_local_kv
:
block_ids_to_permute
+=
meta
.
local_block_ids
if
len
(
block_ids_to_permute
)
>
0
:
self
.
permute_device_kv
(
block_ids_to_permute
)
# Handle timeout to avoid stranding blocks on remote.
# Handle timeout to avoid stranding blocks on remote.
now
=
time
.
perf_counter
()
now
=
time
.
perf_counter
()
...
@@ -1477,15 +1483,6 @@ class NixlConnectorWorker:
...
@@ -1477,15 +1483,6 @@ class NixlConnectorWorker:
del
self
.
_reqs_to_send
[
req_id
]
del
self
.
_reqs_to_send
[
req_id
]
done_sending
.
add
(
req_id
)
done_sending
.
add
(
req_id
)
if
self
.
enable_permute_local_kv
and
len
(
done_recving
)
>
0
:
block_ids
=
[]
for
req_id
in
done_recving
:
meta
=
self
.
_recving_metadata
.
pop
(
req_id
)
assert
meta
,
f
"
{
req_id
}
not found in recving_metadata list"
block_ids
+=
meta
.
local_block_ids
self
.
permute_device_kv
(
block_ids
)
return
done_sending
,
done_recving
return
done_sending
,
done_recving
def
_get_new_notifs
(
self
)
->
set
[
str
]:
def
_get_new_notifs
(
self
)
->
set
[
str
]:
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment