185 lines
6.8 KiB
Python
185 lines
6.8 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
"""Unit tests for SharedMemoryConnector focusing on TP / CFG / metadata fallback."""
|
|
|
|
import pytest
|
|
|
|
from vllm_omni.distributed.omni_connectors.connectors.shm_connector import (
|
|
SharedMemoryConnector,
|
|
)
|
|
|
|
pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
|
|
|
|
|
|
@pytest.fixture()
|
|
def connector():
|
|
c = SharedMemoryConnector({"shm_threshold_bytes": 64})
|
|
yield c
|
|
c.close()
|
|
|
|
|
|
# ── Key-based read (the fundamental SHM path) ────────────────────────
|
|
|
|
|
|
class TestKeyBasedReadWrite:
|
|
def test_put_then_get_by_key(self, connector):
|
|
data = {"hello": "world", "n": 42}
|
|
ok, size, meta = connector.put("s0", "s1", "test_key_1", data)
|
|
assert ok
|
|
assert size > 0
|
|
assert "shm" in meta
|
|
assert "test_key_1" in connector._pending_keys
|
|
|
|
result = connector.get("s0", "s1", "test_key_1", metadata=None)
|
|
assert result is not None
|
|
obj, rsize = result
|
|
assert obj == data
|
|
assert rsize == size
|
|
assert "test_key_1" not in connector._pending_keys
|
|
|
|
def test_get_nonexistent_key_returns_none(self, connector):
|
|
result = connector.get("s0", "s1", "no_such_key_xyz", metadata=None)
|
|
assert result is None
|
|
|
|
def test_rank_aware_keys_independent(self, connector):
|
|
"""Each TP rank writes/reads its own key — simulates homogeneous TP."""
|
|
payloads = {}
|
|
for rank in range(4):
|
|
key = f"req1_s0_0_{rank}_{rank}"
|
|
data = {"rank": rank, "values": list(range(rank, rank + 3))}
|
|
ok, _, _ = connector.put("s0", "s1", key, data)
|
|
assert ok
|
|
payloads[rank] = data
|
|
|
|
for rank in range(4):
|
|
key = f"req1_s0_0_{rank}_{rank}"
|
|
result = connector.get("s0", "s1", key, metadata=None)
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == payloads[rank]
|
|
|
|
|
|
# ── Metadata fallback behaviour ──────────────────────────────────────
|
|
|
|
|
|
class TestMetadataFallback:
|
|
def test_rdma_style_metadata_falls_back_to_key(self, connector):
|
|
"""source_host/source_port metadata should be ignored; key read used."""
|
|
data = {"payload": True}
|
|
connector.put("s0", "s1", "fb_key_1", data)
|
|
|
|
rdma_meta = {"source_host": "10.0.0.1", "source_port": 12345}
|
|
result = connector.get("s0", "s1", "fb_key_1", metadata=rdma_meta)
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == data
|
|
|
|
def test_non_dict_metadata_falls_back_to_key(self, connector):
|
|
data = {"val": 99}
|
|
connector.put("s0", "s1", "fb_key_2", data)
|
|
|
|
result = connector.get("s0", "s1", "fb_key_2", metadata="not_a_dict")
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == data
|
|
|
|
def test_empty_dict_metadata_falls_back_to_key(self, connector):
|
|
data = {"x": 1}
|
|
connector.put("s0", "s1", "fb_key_3", data)
|
|
|
|
result = connector.get("s0", "s1", "fb_key_3", metadata={})
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == data
|
|
|
|
def test_shm_handle_metadata_still_works(self, connector):
|
|
"""When metadata contains a proper 'shm' handle, use it directly."""
|
|
data = {"direct": True}
|
|
ok, size, meta = connector.put("s0", "s1", "shm_direct_1", data)
|
|
assert ok
|
|
result = connector.get("s0", "s1", "shm_direct_1", metadata=meta)
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == data
|
|
|
|
def test_metadata_keyed_by_request_id(self, connector):
|
|
"""Metadata wrapped as {get_key: actual_meta} should be unwrapped."""
|
|
data = {"wrapped": True}
|
|
ok, size, meta = connector.put("s0", "s1", "wrap_key", data)
|
|
assert ok
|
|
wrapped = {"wrap_key": meta}
|
|
result = connector.get("s0", "s1", "wrap_key", metadata=wrapped)
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj == data
|
|
|
|
|
|
# ── Heterogeneous TP multi-key read ──────────────────────────────────
|
|
|
|
|
|
class TestHeteroTPMultiKey:
|
|
def test_receiver_reads_multiple_sender_keys(self, connector):
|
|
"""Simulates from_tp=2 -> to_tp=1: receiver reads 2 keys and merges."""
|
|
for sender_rank in range(2):
|
|
key = f"req1_s0_0_{sender_rank}_0"
|
|
data = {"sender": sender_rank, "shard": [sender_rank * 10]}
|
|
connector.put("s0", "s1", key, data)
|
|
|
|
shards = []
|
|
for sender_rank in range(2):
|
|
key = f"req1_s0_0_{sender_rank}_0"
|
|
result = connector.get("s0", "s1", key, metadata=None)
|
|
assert result is not None
|
|
obj, _ = result
|
|
shards.append(obj)
|
|
|
|
assert len(shards) == 2
|
|
assert shards[0]["sender"] == 0
|
|
assert shards[1]["sender"] == 1
|
|
|
|
def test_sender_writes_multiple_receiver_keys(self, connector):
|
|
"""Simulates from_tp=1 -> to_tp=2: sender writes 2 sliced keys."""
|
|
for recv_rank in range(2):
|
|
key = f"req1_s0_0_0_{recv_rank}"
|
|
data = {"target": recv_rank, "slice": list(range(recv_rank, recv_rank + 2))}
|
|
connector.put("s0", "s1", key, data)
|
|
|
|
for recv_rank in range(2):
|
|
key = f"req1_s0_0_0_{recv_rank}"
|
|
result = connector.get("s0", "s1", key, metadata=None)
|
|
assert result is not None
|
|
obj, _ = result
|
|
assert obj["target"] == recv_rank
|
|
|
|
|
|
# ── Cleanup ──────────────────────────────────────────────────────────
|
|
|
|
|
|
class TestCleanup:
|
|
def test_cleanup_removes_unconsumed_segment(self, connector):
|
|
data = {"leak": True}
|
|
connector.put("s0", "s1", "cleanup_req_42", data)
|
|
assert "cleanup_req_42" in connector._pending_keys
|
|
|
|
connector.cleanup("req_42")
|
|
assert "cleanup_req_42" not in connector._pending_keys
|
|
|
|
result = connector.get("s0", "s1", "cleanup_req_42", metadata=None)
|
|
assert result is None
|
|
|
|
def test_cleanup_noop_for_consumed_segment(self, connector):
|
|
data = {"consumed": True}
|
|
connector.put("s0", "s1", "consumed_req_99", data)
|
|
connector.get("s0", "s1", "consumed_req_99", metadata=None)
|
|
|
|
connector.cleanup("req_99")
|
|
assert "consumed_req_99" not in connector._pending_keys
|
|
|
|
def test_close_cleans_all_pending(self, connector):
|
|
for i in range(3):
|
|
connector.put("s0", "s1", f"close_test_{i}", {"i": i})
|
|
|
|
assert len(connector._pending_keys) == 3
|
|
connector.close()
|
|
assert len(connector._pending_keys) == 0
|