Skip to content
Merged
Show file tree
Hide file tree
Changes from 109 commits
Commits
Show all changes
112 commits
Select commit Hold shift + click to select a range
2430058
resources: publish snapshot hooks so a step can defer its publication
merceod Oct 7, 2026
91cae50
runner: publish_snapshot and publish_from_snapshots sweeps
merceod Oct 7, 2026
68d7c4d
kv: snapshot the publication per step and finish it on demand
merceod Oct 7, 2026
18d8e17
rope: snapshot the publication per step and finish it on demand
merceod Oct 7, 2026
e84b7c2
engine: finalize_batch snapshots the publication, materialize_publish…
merceod Oct 7, 2026
99593f1
worker: finish publications only for stopped rids and rids that send …
merceod Oct 7, 2026
258e27a
test: lazy publication
merceod Oct 7, 2026
b36967b
submodule_base: a stop buffer that is a row output shares the row clone
merceod Oct 7, 2026
52b171e
engine: collect_outputs fast path for row-only outputs
merceod Oct 7, 2026
9272246
test: collect_outputs fast path
merceod Oct 7, 2026
c05f8bb
micro_scheduler: check backlog readiness only up to a step's worth
merceod Oct 7, 2026
2191035
test: bounded backlog readiness checks
merceod Oct 7, 2026
f5a7327
resources: keys_prefix_chains says whether a prefix cache is open
merceod Oct 7, 2026
9e38073
kv: report an open prefix cache through keys_prefix_chains
merceod Oct 7, 2026
f1b0be5
runner: prefix_chain_keys, and extend_prefix_chains over a given key …
merceod Oct 7, 2026
5ef8de3
engine: skip the per-request prefix-chain sweep when no cache is open
merceod Oct 7, 2026
bdf055e
rust frames: RESULT_TOKENS, one frame per step for inline client values
merceod Oct 7, 2026
77d32ec
rust runtime: route an inline client signal without staging, refs or …
merceod Oct 7, 2026
77aab0c
graph runtime: RouteInput.inline_signal/inline_values and supports_in…
merceod Oct 7, 2026
db4def8
rust runtime wrapper: pass inline values, report supports_inline_emit
merceod Oct 7, 2026
95006f7
api_server: ResultTokens message type
merceod Oct 7, 2026
6661a71
wire: register result_tokens
merceod Oct 7, 2026
f6e5939
api_server: dispatch result_tokens frames to the data worker
merceod Oct 7, 2026
27322e4
data worker: detokenize inline tokens in arrival order, no read and n…
merceod Oct 7, 2026
810cf24
submodule_base: inline_client_signals hook
merceod Oct 7, 2026
1761cde
qwen3_5: the decode loop's client token rides inline
merceod Oct 7, 2026
02a6af9
engine: inline_client_signals passthrough
merceod Oct 7, 2026
d57681c
worker: hand the stop check's host tokens to the route as inline values
merceod Oct 7, 2026
21fb696
test: inline token path (wire, data worker, worker)
merceod Oct 7, 2026
a31c952
test: rust runtime inline emit
merceod Oct 7, 2026
b61b1b5
qwen3_5: decode capture buckets up to 128 rows
merceod Oct 7, 2026
51edd0c
api_server: wake the streaming handler on events instead of polling p…
merceod Oct 7, 2026
8d21448
test: a chunk from the message thread wakes the stream
merceod Oct 7, 2026
85c5367
submodule_base: step_is_input_free hook
merceod Oct 7, 2026
ba56be0
qwen3_5: decode steps declare from spans alone
merceod Oct 7, 2026
2ce4dac
engine: run the pre-plan's declared step for input-free walks
merceod Oct 7, 2026
6557266
docs: MSTAR_LAZY_PUBLISH
merceod Oct 7, 2026
080065c
worker: build and pre-plan N+2 while N+1 runs, between N's routing an…
merceod Oct 7, 2026
65ffc1a
docs: MSTAR_EARLY_SPEC
merceod Oct 7, 2026
7b21494
sampler: skip the static gather for a steady batch while other reques…
merceod Oct 8, 2026
6be4dd1
kv plan: build the paged indptr tensors through numpy
merceod Oct 8, 2026
ab3c97a
test: static gather skip with queued requests
merceod Oct 8, 2026
1025fd5
ci: list the new cpu tests
merceod Oct 8, 2026
7359e67
batched output: carry the row views the collect fast path built
merceod Oct 8, 2026
13e7943
engine: hand the row views to the collected output
merceod Oct 8, 2026
5a057b6
tensor store: batch the ownership maps, dtype index helper
merceod Oct 8, 2026
cb3d7e6
tensors: row store for a step whose outputs are row views of one batc…
merceod Oct 8, 2026
f19248a
worker: store decode row outputs through the row store
merceod Oct 8, 2026
129311f
test: merge_per_rid returns the row views
merceod Oct 8, 2026
c104c99
test: row store against the general path
merceod Oct 8, 2026
28d739d
ci: list the row store test
merceod Oct 8, 2026
760bde4
worker: build the stop check's per-rid host dicts only when read
merceod Oct 8, 2026
4e50fb7
engine: skip the unpack pass when nothing is packed
merceod Oct 8, 2026
9163255
test: lazy per-rid host dicts
merceod Oct 8, 2026
5e2b57c
ci: list the lazy per-rid test
merceod Oct 8, 2026
083cc0f
qwen3_5: MSTAR_QWEN35_DECODE_BUCKETS overrides the decode capture ladder
merceod Oct 8, 2026
5025706
docs: MSTAR_QWEN35_DECODE_BUCKETS
merceod Oct 8, 2026
a025fac
step context: constant-time padding and real-row tests
merceod Oct 8, 2026
b3a01b6
kv: batched publish snapshot under one lock, real-row test in the pag…
merceod Oct 8, 2026
91e4d28
resources: publish_snapshot_batch hook
merceod Oct 8, 2026
b2f60e6
runner: snapshot the publication per resource for the whole batch
merceod Oct 8, 2026
cb6bed1
sampler: one graph sampler per bucket and slot
merceod Oct 8, 2026
1dce205
test: step context row membership
merceod Oct 8, 2026
6cb5601
test: batched publish snapshot
merceod Oct 8, 2026
d2dcb38
test: sampler_for cache
merceod Oct 8, 2026
9f222e4
ci: list the step, snapshot and sampler tests
merceod Oct 8, 2026
fb04771
kv admit: skip the allocation call for rows whose pages already cover…
merceod Oct 8, 2026
a2d1155
engine: hoist the per-row constants out of prepare_inputs, look up th…
merceod Oct 8, 2026
dbbc572
arena: batched cleanup of collectable tensors
merceod Oct 8, 2026
50baf43
worker: send only the request entries that carry counts or stream con…
merceod Oct 8, 2026
5404251
engine: direct properties for the forward-pass wrapper's hot attributes
merceod Oct 8, 2026
bcfe673
test: batched stream consumption report
merceod Oct 8, 2026
75f5ea4
ci: list the send prep test
merceod Oct 8, 2026
c72d6af
kv commit: per-row constants hoisted, index offer only with a chain
merceod Oct 8, 2026
3a31d8b
api_server: wake the streaming handlers once per message pass
merceod Oct 8, 2026
b861dcf
test: batched handler wake-up
merceod Oct 8, 2026
f5f5303
rust runtime: env-gated timing of send_outputs (body vs GIL reacquire)
merceod Oct 8, 2026
6de4423
worker: MSTAR_SPEC_YIELD_HOLD_MS holds the decode loop's yield so arr…
merceod Oct 8, 2026
2b422b9
docs: MSTAR_SPEC_YIELD_HOLD_MS
merceod Oct 8, 2026
720587d
rust runtime: flat export of the dynamic loop iterations
merceod Oct 8, 2026
d47c22f
graph runtime: update_dynamic_loop_iters hook
merceod Oct 8, 2026
880cdbe
rust runtime wrapper: refresh loop iterations from the flat export
merceod Oct 8, 2026
cb0b9ce
worker: refresh loop iterations through the runtime hook
merceod Oct 8, 2026
aa5c467
test: update_dynamic_loop_iters matches the per-rid form
merceod Oct 8, 2026
92bed84
kv: batched prefix-chain extension off the stop check's host rows
merceod Oct 8, 2026
cb338d8
runner: batched chain extension; sweep resources that override the ho…
merceod Oct 8, 2026
da6df5b
resources: document when the chain hook is swept
merceod Oct 8, 2026
61ea439
engine: extend prefix chains from the host rows
merceod Oct 8, 2026
f66cb9b
worker: hand the host rows to the prefix-chain extension
merceod Oct 8, 2026
a490f5b
test: batched prefix-chain extension
merceod Oct 8, 2026
b369e75
ci: list the prefix batch test
merceod Oct 8, 2026
60c0ab2
engine: read the keyed walks leniently in prepare_inputs
merceod Oct 8, 2026
d35c66b
zmq communicator: non-blocking try_send
merceod Oct 8, 2026
22278c5
rust runtime: send frames with the GIL held, release it only for a pe…
merceod Oct 8, 2026
04a8f36
docs: MSTAR_RUST_SEND_TIMING
merceod Oct 8, 2026
5db372f
data worker: sort the import block
merceod Oct 8, 2026
5912708
rust runtime: the inline signal keys of a route input are optional
merceod Oct 8, 2026
4406b62
test: row store tests skip without the columnar tensor store
merceod Oct 8, 2026
e3c54f5
engine: a reset pre-plan drops its kept step
merceod Oct 12, 2026
f6274b1
test: a partly dropped batch does not keep its pre-planned step
merceod Oct 12, 2026
eef4b85
kv: a snapshot of a stream reset since publishes the stream as it stands
merceod Oct 12, 2026
57ebad9
test: publish from a snapshot after a reset
merceod Oct 12, 2026
b3e3b55
worker: drop the spec yield hold, note why two slots suffice, note th…
merceod Oct 12, 2026
734972e
docs: MSTAR_SPEC_YIELD_HOLD_MS removed, MSTAR_DECODE_BUCKETS added
merceod Oct 12, 2026
5b54fc7
api: APIServerMessageBody names the message body union
merceod Oct 12, 2026
5838168
tensor store: ColumnarTensorInfo.add_row
merceod Oct 12, 2026
37f9613
tensors: the row fast path fills the columns through add_row
merceod Oct 12, 2026
3a814eb
qwen3_5: decode buckets from MSTAR_DECODE_BUCKETS, the model knob ove…
merceod Oct 12, 2026
cef920e
nvtx had _postprocess_batch_stops being closed at start of _postproce…
stephen-dwq Oct 12, 2026
e7d109f
revert N+2 speculation
stephen-dwq Oct 14, 2026
7729610
CI fix + merge
stephen-dwq Oct 14, 2026
b759735
Merge remote-tracking branch 'origin/main' into perf/host-overhead
stephen-dwq Oct 14, 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
32 changes: 32 additions & 0 deletions docs/environment_variables.rst
Original file line number Diff line number Diff line change
Expand Up @@ -295,10 +295,42 @@ Worker scheduling
- Pre-plan the speculative batch's attention on a dedicated thread
while the previous replay runs. ``0`` plans inline on the GPU
thread.
* - ``MSTAR_DECODE_BUCKETS``
- ``1,2,4,8,16,32,48,64,96,128``
- The decode batch sizes a model captures as CUDA graphs (Qwen3.5 reads
it; a model with its own ladder reads it the same way). A decode step
runs at the smallest bucket that holds its rows, so the ladder bounds
both the rows per step and the graph memory; narrow it (for example
``1,2,4,8,16,32``) on a memory-tight GPU or for an A/B against the
former ladder. ``MSTAR_QWEN35_DECODE_BUCKETS`` overrides it for
Qwen3.5 alone.
* - ``MSTAR_EARLY_SPEC``
- ``1``
- Build and pre-plan step N+2 while N+1 runs, between the stop and
routing halves of step N's post-processing, so the pre-plan overlaps
work the main thread does anyway instead of the next launch. ``0``
builds each speculation at the top of the next iteration, after the
whole post-processing, which left the plan thread idle during it.
Non-parallel (TP1) nodes only; a tensor-parallel node keeps the old
order so the followers see heads in the order they settle them.
* - ``MSTAR_LAZY_PUBLISH``
- ``1``
- Publish resource state (KV pages, positions) only for the requests
whose frame or completion carries it: ``finalize_batch`` keeps a
cheap per-request snapshot and the worker finishes it on demand.
``0`` builds the full publication for every request on every step,
the old behaviour; it was a quarter of the GPU thread's host time at
batch 32 in a decode loop that never reads it.
* - ``MSTAR_MAX_CONSECUTIVE_SPEC_STEPS``
- ``1024``
- Cap on back-to-back speculative steps before the leader yields to
other ready work.
* - ``MSTAR_RUST_SEND_TIMING``

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Nit: this environment variable is framed as a diagnostic for this host-overhead work (aka this PR); if you want to keep the switch in production, can you frame it as a being for generic performance debugging?

- unset
- ``1`` makes the Rust runtime print, every 500 calls, how long the
worker's per-step send spends in its own work against the whole call
(which includes taking the GIL back when the send had to release it).
A diagnostic for the host-overhead work; off by default.
* - ``MSTAR_SPEC_PEEK_FOR_FAIRNESS``
- ``1``
- Yield the speculation chain only when another (node, walk) is
Expand Down
63 changes: 63 additions & 0 deletions mstar/api_server/data_worker.py
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,15 @@ def _video_frame_metadata(
"frame_index": frame_index,
"frame_count": frame_count,
}
@dataclass
class InlineToken:
"""One request's value out of a RESULT_TOKENS frame, queued behind the
tensor results so sequences keep arrival order."""
request_id: str
value: int
loop_indices: NestedLoopIndices | None
modality: str



def _preprocess_loop(**kwargs):
Expand Down Expand Up @@ -221,6 +230,24 @@ def new_result_tensors(self, input: ResultTensors):
)
self.result_tensor_input_queue.put(input)

def new_result_token(
self, request_id: str, value: int,
loop_indices: NestedLoopIndices | None, signal: str, modality: str,
):
"""A value that rode inline: same accounting as a result tensor, no
transport read and nothing to ack."""
if request_id not in self.output_loop_idxs:
logger.debug("Late token for cleaned-up request %s dropped", request_id)
return
if loop_indices is not None:
self.output_loop_idxs[request_id][signal] = loop_indices.max(
self.output_loop_idxs[request_id].get(signal, None)
)
self.per_request_reading_tensors[request_id] += 1
self.result_tensor_input_queue.put(
InlineToken(request_id, value, loop_indices, modality)
)

def discard_result_tensors(self, input: ResultTensors):
"""Ack and drop result tensors for an already-removed request.

Expand Down Expand Up @@ -663,6 +690,36 @@ def _read_result_tensor(
state.order[tensor_info.uuid] = (state.next_sequence, result.loop_indices)
state.next_sequence += 1

def _emit_inline_token(self, item: InlineToken) -> None:
"""Detokenize a value that arrived inline and queue its chunk."""
request_id = item.request_id
state = self.request_output_state.setdefault(
request_id, RequestOutputState()
)
sequence = state.next_sequence
state.next_sequence += 1
try:
tensor = torch.tensor([item.value], dtype=torch.int64)
postprocessed = self.model.postprocess(
tensor, item.modality,
request_kwargs=self.request_model_kwargs.get(request_id),
)
self._queue_completed_output(
request_id,
sequence,
ResultChunk(
request_id=request_id,
modality=item.modality,
data=postprocessed,
metadata={},
),
)
except Exception as exc: # noqa: BLE001 — must reach the client
self._fail_request(
request_id, exc, f"{item.modality} output postprocessing",
sequence=sequence,
)

def _queue_completed_output(
self,
request_id: str,
Expand Down Expand Up @@ -888,6 +945,12 @@ def run(self):
while not self.result_tensor_queue.empty():
did_work = True
result = self.result_tensor_queue.get()
if isinstance(result, InlineToken):
# Nothing to read or ack: postprocess in place, in
# arrival order with the tensor results.
if result.request_id not in self._draining_rids:
self._emit_inline_token(result)
continue
# Draining for teardown: don't start new reads — ack the
# tensors back so the producing worker can free its buffers.
if result.request_id in self._draining_rids:
Expand Down
86 changes: 85 additions & 1 deletion mstar/api_server/entrypoint.py
Original file line number Diff line number Diff line change
Expand Up @@ -184,6 +184,47 @@ class PendingRequest:
consumed_chunks: int = 0
error: Any | None = None
error_status: int = 500
# Set by the streaming handler on its event loop; the message thread wakes
# it through these when a chunk or the completion lands, so the handler
# sleeps between chunks instead of polling every millisecond per request.
wake: Any | None = None
loop: Any | None = None

def notify(self) -> None:
"""Wake the streaming handler, from any thread."""
loop, wake = self.loop, self.wake
if loop is None or wake is None:
return
try:
loop.call_soon_threadsafe(wake.set)
except RuntimeError:
pass # the loop is closed: the client is gone


def _set_events(wakes: list) -> None:
for wake in wakes:
wake.set()


def notify_many(reqs) -> None:
"""``PendingRequest.notify`` for a batch: one loop wake-up (a self-pipe
write) per event loop rather than one per chunk. A step's RESULT_TOKENS
frame lands one chunk per running request, so the message thread was
waking the loop once per token."""
by_loop: dict[int, tuple[Any, list]] = {}
for req in reqs:
loop, wake = req.loop, req.wake
if loop is None or wake is None:
continue
entry = by_loop.get(id(loop))
if entry is None:
entry = by_loop[id(loop)] = (loop, [])
entry[1].append(wake)
for loop, wakes in by_loop.values():
try:
loop.call_soon_threadsafe(_set_events, wakes)
except RuntimeError:
pass # the loop is closed: the client is gone


def _chunk_to_ndjson_payload(chunk: ResultChunk) -> str:
Expand Down Expand Up @@ -361,6 +402,7 @@ def _fail_pending_for_dead_conductor(self, exited: str) -> None:
req.error = message
req.error_status = 503
req.event.set()
req.notify()
on_fatal = self.on_fatal
if on_fatal is not None:
on_fatal()
Expand Down Expand Up @@ -491,6 +533,7 @@ def _prune_recently_completed(self) -> None:
)
req.error_status = 500
req.event.set()
req.notify()
# Snapshot the data worker's tx/rx now: the request is done (all
# final chunks received), so the worker thread is no longer mutating
# this rid's transport state, and we must read it before the hard
Expand Down Expand Up @@ -531,6 +574,24 @@ def _process_messages(self) -> None:
logger.warning("Unexpected message type: %s", type(message))
continue

if message.message_type == "result_tokens":
# One frame for the step: every request's value.
# A request that is gone needs nothing acked back.
body = message.body
with self.request_lock:
for i, rid in enumerate(body.request_ids):
if rid in self.pending_requests:
self.preprocess_worker.new_result_token(
rid, body.values[i],
body.loop_indices[i],
body.signal, body.modality,
)
elif rid not in self.recently_completed:
logger.warning(
"Token for unknown request %s dropped", rid,
)
continue

rid = message.body.request_id

with self.request_lock:
Expand Down Expand Up @@ -559,6 +620,7 @@ def _process_messages(self) -> None:
# result is coming, so the alternative is the
# blanket request timeout.
req.event.set()
req.notify()
# The conductor has already dropped this rid,
# so no abort is needed — but the data worker
# still holds its transport state. Parking the
Expand Down Expand Up @@ -617,6 +679,9 @@ def _process_messages(self) -> None:
update.preprocess_finish_time
req.profile.inputs = update.inputs

# the handlers are woken once per pass, after every chunk of
# the pass has been appended, not once per chunk
to_wake: dict[int, PendingRequest] = {}
for result_chunk in self.preprocess_worker.get_result_chunks():
logger.debug(
"Got result chunk of %s modality for request %s",
Expand All @@ -638,6 +703,7 @@ def _process_messages(self) -> None:
req.profile.timing.first_chunk_time = now
req.profile.timing.last_chunk_time = now
req.chunks.append(result_chunk)
to_wake[id(req)] = req

if result_chunk.modality == "error":
# The data worker failed this request (preprocess,
Expand All @@ -654,6 +720,8 @@ def _process_messages(self) -> None:
# the data worker's per-request state once the
# client lets go of the request.
self.recently_completed[rid] = time.time()
if to_wake:
notify_many(to_wake.values())
except Exception:
if self.running:
logger.exception("Error in message processing loop")
Expand All @@ -676,11 +744,22 @@ async def iter_result_chunks(self, request_id: str):
"""
start = time.time()
finished = False
# Woken by the message thread (PendingRequest.notify) when a chunk or
# the completion lands; the timeout below is only a backstop.
wake = asyncio.Event()
with self.request_lock:
req = self.pending_requests.get(request_id)
if req is not None:
req.loop = asyncio.get_running_loop()
req.wake = wake
try:
while True:
if time.time() - start > self.timeout_seconds:
raise HTTPException(status_code=500, detail="Request timed out")

# Cleared before the read: a notification that lands after
# this point stays set and the wait below returns at once.
wake.clear()
new_chunks: list[ResultChunk] = []
done = False
with self.request_lock:
Comment thread
NSagan271 marked this conversation as resolved.
Resolved
Expand Down Expand Up @@ -729,7 +808,12 @@ async def iter_result_chunks(self, request_id: str):
finished = True
break

await asyncio.sleep(0.001)
if new_chunks:
continue # more may already be waiting
try:
await asyncio.wait_for(wake.wait(), timeout=0.05)
except asyncio.TimeoutError:
pass
finally:
if not finished:
self.abort_request(request_id)
Expand Down
22 changes: 20 additions & 2 deletions mstar/api_server/request_types.py
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,20 @@ class ResultTensors:
metadata: dict = field(default_factory=dict)


@dataclass
class ResultTokens:
"""One step's client-facing scalars (sampled tokens), every request in one
frame. Replaces one ResultTensors per request for an output whose value
the worker already holds on the host: no transport tensor, no ack."""
request_ids: list[str]
values: list[int]
# per request, the loop context of the step that produced the value
loop_indices: list[NestedLoopIndices | None]
signal: str
modality: str
metadata: dict = field(default_factory=dict)


@dataclass
class RequestComplete:
"""Signals that a request has finished processing."""
Expand Down Expand Up @@ -52,12 +66,16 @@ class RequestFailed:
status: int = 500


# what an ``APIServerMessage`` carries; None for ``setup_done``
APIServerMessageBody = ResultTensors | ResultTokens | RequestComplete | RequestFailed | None


@dataclass
class APIServerMessage:
"""Envelope for messages received by the API server."""
# "result_tensors" | "request_complete" | "request_failed" | "setup_done"
# "result_tensors" | "result_tokens" | "request_complete" | "request_failed" | "setup_done"
message_type: str
body: ResultTensors | RequestComplete | RequestFailed | None = None # None for setup_done
body: APIServerMessageBody = None


@dataclass
Expand Down
41 changes: 41 additions & 0 deletions mstar/communication/arena.py
Original file line number Diff line number Diff line change
Expand Up @@ -925,6 +925,47 @@ def close(self) -> None:
self._wake_q.put(None)
self._wake_q = None

def cleanup_collectable(
self, collectable: list[int], registered: list[bool],
):
"""``_cleanup_by_uuid`` for a batch, inlined: a decode step frees one
tensor per request here, and the three-level call chain per uuid was
most of the cost. Same steps in the same order as the per-uuid form."""
if not collectable:
return
del registered # no transfer engine to unregister from
store = self.tensor_store
store.mark_forgotten(collectable)
shard_dims = self.uuid_to_shard_dim
edge_names = self.uuid_to_edge_name
arena_ts = self._arena_ts
arena_locs = self._arena_locs
shm_files = self._shm_files
present = store.check_uuid_presence
remove = store.remove_tensor
for uuid in collectable:
try:
shard_dims.pop(uuid, None)
edge_names.pop(uuid, None)
if present(uuid):
remove(uuid)
else:
logger.warning(
"Trying to cleanup tensor %s, but uuid not found", uuid,
)
arena_ts.pop(uuid, None)
if (loc := arena_locs.pop(uuid, None)) is not None:
self._arena.free(*loc)
if (path := shm_files.pop(uuid, None)) is not None:
try:
os.unlink(path)
except FileNotFoundError:
pass
except Exception as e:
logger.error(
"Error cleaning up tensor uuid %d: %s, skipping.", uuid, e,
)

def _cleanup_by_uuid(self, uuid: int, registered: bool | None = None):
# Grandparent cleanup (refcounts): skip the file manager's unlink.
super(SharedMemoryCommunicationManager, self)._cleanup_by_uuid(
Expand Down
Loading
Loading