| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Tests for DataTransport ABC and NixlDataTransport.""" |
| 4 | |
| 5 | from unittest.mock import MagicMock, patch |
| 6 | |
| 7 | import pytest |
| 8 | |
| 9 | from vllm.distributed.ec_transfer.ec_connector.cpu.data.base import ( |
| 10 | DataTransport, |
| 11 | ) |
| 12 | from vllm.distributed.ec_transfer.ec_connector.cpu.utils import ( |
| 13 | serialize_mem_descriptor, |
| 14 | ) |
| 15 | |
| 16 | _NIXL_PATH = "vllm.distributed.ec_transfer.ec_connector.cpu.data.nixl" |
| 17 | |
| 18 | |
| 19 | # ── helpers ─────────────────────────────────────────────────────────────────── |
| 20 | |
| 21 | |
| 22 | def _make_nixl_mock(prep_side_effect=None) -> MagicMock: |
| 23 | nixl = MagicMock() |
| 24 | nixl.get_agent_metadata.return_value = b"agent-meta" |
| 25 | nixl.get_reg_descs.return_value = MagicMock() |
| 26 | nixl.get_xfer_descs.return_value = MagicMock() |
| 27 | nixl.prep_xfer_dlist.side_effect = prep_side_effect or [77, 88] |
| 28 | nixl.add_remote_agent.return_value = "remote-agent-1" |
| 29 | nixl.get_new_notifs.return_value = {} |
| 30 | nixl.check_xfer_state.return_value = "PROC" |
| 31 | return nixl |
| 32 | |
| 33 | |
| 34 | def _make_transport(nixl_mock=None): |
| 35 | from vllm.distributed.ec_transfer.ec_connector.cpu.data.nixl import ( |
| 36 | NixlDataTransport, |
| 37 | ) |
| 38 | |
| 39 | mock = nixl_mock or _make_nixl_mock() |
| 40 | with ( |
| 41 | patch(f"{_NIXL_PATH}.NixlWrapper", return_value=mock), |
| 42 | patch(f"{_NIXL_PATH}.nixl_agent_config", return_value=MagicMock()), |
| 43 | ): |
| 44 | t = NixlDataTransport("agent", 0, 4, 64, 256) |
| 45 | t._nixl = mock |
| 46 | return t |
| 47 | |
| 48 | |
| 49 | # ── ABC ─────────────────────────────────────────────────────────────────────── |
| 50 | |
| 51 | |
| 52 | def test_data_transport_abc_cannot_be_instantiated(): |
| 53 | with pytest.raises(TypeError): |
| 54 | DataTransport() # type: ignore[abstract] |
| 55 | |
| 56 | |
| 57 | # ── NixlDataTransport construction ─────────────────────────────────────────── |
| 58 | |
| 59 | |
| 60 | def test_nixl_data_transport_registers_memory_on_init(): |
| 61 | nixl = _make_nixl_mock() |
| 62 | _make_transport(nixl) |
| 63 | nixl.register_memory.assert_called_once() |
| 64 | |
| 65 | |
| 66 | def test_nixl_data_transport_unwinds_registration_if_init_fails(): |
| 67 | """A failure after register_memory must not leave the region pinned. |
| 68 | |
| 69 | __init__ propagates, so the caller never gets a transport to deregister. |
| 70 | """ |
| 71 | nixl = _make_nixl_mock(prep_side_effect=RuntimeError("no dlist")) |
| 72 | from vllm.distributed.ec_transfer.ec_connector.cpu.data.nixl import ( |
| 73 | NixlDataTransport, |
| 74 | ) |
| 75 | |
| 76 | with ( |
| 77 | patch(f"{_NIXL_PATH}.NixlWrapper", return_value=nixl), |
| 78 | patch(f"{_NIXL_PATH}.nixl_agent_config", return_value=MagicMock()), |
| 79 | pytest.raises(RuntimeError, match="no dlist"), |
| 80 | ): |
| 81 | NixlDataTransport("agent", 0, 4, 64, 256) |
| 82 | |
| 83 | nixl.deregister_memory.assert_called_once_with(nixl.get_reg_descs.return_value) |
| 84 | |
| 85 | |
| 86 | def test_nixl_data_transport_raises_if_nixl_unavailable(): |
| 87 | from vllm.distributed.ec_transfer.ec_connector.cpu.data.nixl import ( |
| 88 | NixlDataTransport, |
| 89 | ) |
| 90 | |
| 91 | with ( |
| 92 | patch(f"{_NIXL_PATH}.NixlWrapper", None), |
| 93 | patch(f"{_NIXL_PATH}.nixl_agent_config", None), |
| 94 | pytest.raises(RuntimeError, match="requires NIXL"), |
| 95 | ): |
| 96 | NixlDataTransport("a", 0, 1, 64, 64) |
| 97 | |
| 98 | |
| 99 | # ── public accessors ────────────────────────────────────────────────────────── |
| 100 | |
| 101 | |
| 102 | def test_get_agent_metadata_returns_nixl_blob(): |
| 103 | assert _make_transport().get_agent_metadata() == b"agent-meta" |
| 104 | |
| 105 | |
| 106 | def test_get_mem_descriptor_roundtrips_block_descs(): |
| 107 | t = _make_transport() |
| 108 | raw = t.get_mem_descriptor() |
| 109 | assert isinstance(raw, bytes) and len(raw) > 0 |
| 110 | |
| 111 | |
| 112 | # ── add_remote_peer ─────────────────────────────────────────────────────────── |
| 113 | |
| 114 | |
| 115 | def test_add_remote_peer_returns_agent_name_and_stores_handle_internally(): |
| 116 | nixl = _make_nixl_mock(prep_side_effect=[77, 88]) |
| 117 | t = _make_transport(nixl) |
| 118 | mem_desc = serialize_mem_descriptor([(0, 64, 0)]) |
| 119 | agent_name = t.add_remote_peer(b"meta", mem_desc) |
| 120 | assert agent_name == "remote-agent-1" |
| 121 | # Dlist handle stored internally; accessible via _peer_handles. |
| 122 | assert t._peer_handles["remote-agent-1"] == 88 |
| 123 | |
| 124 | |
| 125 | def test_add_remote_peer_registers_remote_agent(): |
| 126 | nixl = _make_nixl_mock(prep_side_effect=[77, 88]) |
| 127 | t = _make_transport(nixl) |
| 128 | mem_desc = serialize_mem_descriptor([(0, 64, 0)]) |
| 129 | t.add_remote_peer(b"fresh-meta", mem_desc) |
| 130 | nixl.add_remote_agent.assert_called_once_with(b"fresh-meta") |
| 131 | |
| 132 | |
| 133 | # ── remove_remote_peer ──────────────────────────────────────────────────────── |
| 134 | |
| 135 | |
| 136 | def test_remove_remote_peer_delegates_to_nixl(): |
| 137 | t = _make_transport() |
| 138 | t.remove_remote_peer("agent-x") |
| 139 | t._nixl.remove_remote_agent.assert_called_once_with("agent-x") |
| 140 | |
| 141 | |
| 142 | def test_remove_remote_peer_swallows_exceptions(): |
| 143 | t = _make_transport() |
| 144 | t._nixl.remove_remote_agent.side_effect = RuntimeError("gone") |
| 145 | t.remove_remote_peer("agent-x") # must not raise |
| 146 | |
| 147 | |
| 148 | # ── post_read ───────────────────────────────────────────────────────────────── |
| 149 | |
| 150 | |
| 151 | def test_post_read_raises_on_index_count_mismatch(): |
| 152 | nixl = _make_nixl_mock(prep_side_effect=[77, 88]) |
| 153 | t = _make_transport(nixl) |
| 154 | mem_desc = serialize_mem_descriptor([(0, 64, 0)]) |
| 155 | agent = t.add_remote_peer(b"meta", mem_desc) |
| 156 | with pytest.raises(ValueError, match="mismatch"): |
| 157 | t.post_read([0, 1], agent, [5], notif_msg=b"sid") |
| 158 | |
| 159 | |
| 160 | def test_post_read_calls_make_prepped_xfer_and_transfer(): |
| 161 | nixl = _make_nixl_mock(prep_side_effect=[77, 88]) |
| 162 | t = _make_transport(nixl) |
| 163 | mem_desc = serialize_mem_descriptor([(0, 64, 0)]) |
| 164 | agent = t.add_remote_peer(b"meta", mem_desc) |
| 165 | nixl.make_prepped_xfer.return_value = 999 |
| 166 | handle = t.post_read([0, 1], agent, [5, 6], notif_msg=b"sid") |
| 167 | nixl.make_prepped_xfer.assert_called_once() |
| 168 | nixl.transfer.assert_called_once_with(999) |
| 169 | assert handle == 999 |
| 170 | |
| 171 | |
| 172 | # ── release_xfer_handle / deregister ───────────────────────────────────────── |
| 173 | |
| 174 | |
| 175 | def test_release_xfer_handle_swallows_exceptions(): |
| 176 | t = _make_transport() |
| 177 | t._nixl.release_xfer_handle.side_effect = RuntimeError("stale") |
| 178 | t.release_xfer_handle(42) # must not raise |
| 179 | |
| 180 | |
| 181 | def test_deregister_swallows_exceptions(): |
| 182 | t = _make_transport() |
| 183 | t._nixl.deregister_memory.side_effect = RuntimeError("already gone") |
| 184 | t.deregister() # must not raise |