| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | """Unit tests for the unified bidirectional P2PSession. |
| 4 | |
| 5 | A P2PSession owns a single ControlConnection and dispatches every message |
| 6 | type — both client-role (FetchMsg / TransferDoneMsg / AbortAck) and |
| 7 | server-role (FetchMsg / TransferDoneMsg / AbortAck from the peer's |
| 8 | perspective). These tests exercise both flows independently and the |
| 9 | bidirectional case where one session simultaneously serves a fetch and |
| 10 | completes its own load. |
| 11 | """ |
| 12 | |
| 13 | from __future__ import annotations |
| 14 | |
| 15 | import time |
| 16 | from collections.abc import Sequence |
| 17 | |
| 18 | import numpy as np |
| 19 | import pytest |
| 20 | |
| 21 | from vllm.v1.kv_offload.base import ( |
| 22 | LookupResult, |
| 23 | OffloadKey, |
| 24 | ReqContext, |
| 25 | RequestOffloadingContext, |
| 26 | ) |
| 27 | from vllm.v1.kv_offload.tiering.base import TransferJob |
| 28 | from vllm.v1.kv_offload.tiering.p2p.session import ( |
| 29 | LoadResult, |
| 30 | P2PSession, |
| 31 | StoreResult, |
| 32 | ) |
| 33 | from vllm.v1.kv_offload.tiering.p2p.session.client import ( |
| 34 | _ABORT_ACK_TIMEOUT_S, |
| 35 | _LOAD_TIMEOUT_S, |
| 36 | ) |
| 37 | from vllm.v1.kv_offload.tiering.p2p.session.protocol import ( |
| 38 | TYPE_KEY, |
| 39 | AbortAckMsg, |
| 40 | AbortFetchMsg, |
| 41 | ConnectAckMsg, |
| 42 | ConnectMsg, |
| 43 | DisconnectMsg, |
| 44 | FetchMsg, |
| 45 | LookupMsg, |
| 46 | LookupRespMsg, |
| 47 | TransferDoneMsg, |
| 48 | ) |
| 49 | from vllm.v1.kv_offload.tiering.p2p.session.server import ( |
| 50 | _CANCEL_DRAIN_TIMEOUT_S, |
| 51 | _InflightXfer, |
| 52 | _OutboundRequestState, |
| 53 | ) |
| 54 | from vllm.v1.kv_offload.tiering.p2p.session.session import ( |
| 55 | _MAX_CONSECUTIVE_DISPATCH_ERRORS, |
| 56 | ) |
| 57 | |
| 58 | # --------------------------------------------------------------------------- |
| 59 | # Helpers |
| 60 | # --------------------------------------------------------------------------- |
| 61 | |
| 62 | |
| 63 | # Shared PYTHONHASHSEED used by the session under test and the fake peer's |
| 64 | # ConnectMsg so the handshake succeeds unless a test overrides one side. |
| 65 | _DEFAULT_HASH_SEED = "0" |
| 66 | |
| 67 | |
| 68 | class FakeDataTransport: |
| 69 | """Minimal fake DataTransport for testing sessions.""" |
| 70 | |
| 71 | def __init__( |
| 72 | self, |
| 73 | base_addr: int = 0x1000, |
| 74 | num_blocks: int = 16, |
| 75 | block_len: int = 4096, |
| 76 | config_fingerprint: str = "", |
| 77 | ) -> None: |
| 78 | self._base_addr = base_addr |
| 79 | self._num_blocks = num_blocks |
| 80 | self._block_len = block_len |
| 81 | self._config_fingerprint = config_fingerprint |
| 82 | self._remote_peers: dict[str, dict] = {} |
| 83 | self._transfers: dict[int, tuple] = {} |
| 84 | self._next_id = 0 |
| 85 | self._poll_done: list[int] = [] |
| 86 | self._poll_failed: list[int] = [] |
| 87 | self._cancel_still_inflight: set[int] = set() |
| 88 | self._cancel_calls: list[tuple[list[int], str]] = [] |
| 89 | |
| 90 | @property |
| 91 | def base_addr(self) -> int: |
| 92 | return self._base_addr |
| 93 | |
| 94 | @property |
| 95 | def num_blocks(self) -> int: |
| 96 | return self._num_blocks |
| 97 | |
| 98 | @property |
| 99 | def block_len(self) -> int: |
| 100 | return self._block_len |
| 101 | |
| 102 | @property |
| 103 | def config_fingerprint(self) -> str: |
| 104 | return self._config_fingerprint |
| 105 | |
| 106 | def get_agent_metadata(self) -> bytes: |
| 107 | return b"fake-metadata" |
| 108 | |
| 109 | def add_remote_peer( |
| 110 | self, peer_id, agent_metadata, base_addr, num_blocks, block_len |
| 111 | ) -> None: |
| 112 | self._remote_peers[peer_id] = { |
| 113 | ConnectMsg.AGENT_METADATA: agent_metadata, |
| 114 | ConnectMsg.BASE_ADDR: base_addr, |
| 115 | "num_blocks": num_blocks, |
| 116 | "block_len": block_len, |
| 117 | } |
| 118 | |
| 119 | def remove_remote_peer(self, peer_id: str) -> None: |
| 120 | self._remote_peers.pop(peer_id, None) |
| 121 | |
| 122 | def write_blocks(self, peer_id, local_idxs, remote_idxs) -> int | None: |
| 123 | if peer_id not in self._remote_peers: |
| 124 | return None |
| 125 | tid = self._next_id |
| 126 | self._next_id += 1 |
| 127 | self._transfers[tid] = (peer_id, local_idxs, remote_idxs) |
| 128 | return tid |
| 129 | |
| 130 | def poll(self, peer_id=None): |
| 131 | from vllm.v1.kv_offload.tiering.p2p.data.base import PollResult |
| 132 | |
| 133 | result = PollResult(done=list(self._poll_done), failed=list(self._poll_failed)) |
| 134 | self._poll_done.clear() |
| 135 | self._poll_failed.clear() |
| 136 | return result |
| 137 | |
| 138 | def cancel(self, transfer_ids, mode: str = "immediate") -> list[int]: |
| 139 | ids = list(transfer_ids) |
| 140 | self._cancel_calls.append((ids, mode)) |
| 141 | if mode == "wait": |
| 142 | still: list[int] = [] |
| 143 | for tid in ids: |
| 144 | if tid in self._cancel_still_inflight: |
| 145 | still.append(tid) |
| 146 | else: |
| 147 | self._transfers.pop(tid, None) |
| 148 | return still |
| 149 | for tid in ids: |
| 150 | self._transfers.pop(tid, None) |
| 151 | self._cancel_still_inflight.discard(tid) |
| 152 | return [] |
| 153 | |
| 154 | def close(self) -> None: |
| 155 | pass |
| 156 | |
| 157 | |
| 158 | class FakeConnection: |
| 159 | """Fake ControlConnection that captures sent messages.""" |
| 160 | |
| 161 | def __init__(self, peer_id: str = "peer:8000") -> None: |
| 162 | self.peer_id = peer_id |
| 163 | self._inbox: list[dict] = [] |
| 164 | self._sent: list[dict] = [] |
| 165 | self._closed = False |
| 166 | # When True, send() raises to simulate a broken/dead connection. |
| 167 | self.fail_send = False |
| 168 | |
| 169 | @property |
| 170 | def alive(self) -> bool: |
| 171 | return not self._closed |
| 172 | |
| 173 | def send(self, msg: dict) -> None: |
| 174 | if self.fail_send: |
| 175 | raise ConnectionError("simulated dead connection") |
| 176 | self._sent.append(msg) |
| 177 | |
| 178 | def recv(self) -> list[dict]: |
| 179 | msgs = self._inbox |
| 180 | self._inbox = [] |
| 181 | return msgs |
| 182 | |
| 183 | def enqueue(self, msg: dict) -> None: |
| 184 | self._inbox.append(msg) |
| 185 | |
| 186 | def mark_dead(self) -> None: |
| 187 | self._closed = True |
| 188 | |
| 189 | def close(self) -> None: |
| 190 | self._closed = True |
| 191 | |
| 192 | |
| 193 | def _peer_connect_msg( |
| 194 | peer_id: str = "peer:8000", |
| 195 | block_len: int = 4096, |
| 196 | fingerprint: str | None = None, |
| 197 | hash_seed: str = _DEFAULT_HASH_SEED, |
| 198 | ) -> dict: |
| 199 | """Build a ConnectMsg as if the peer sent it.""" |
| 200 | msg = { |
| 201 | TYPE_KEY: ConnectMsg.TYPE, |
| 202 | ConnectMsg.PEER_ID: peer_id, |
| 203 | ConnectMsg.AGENT_METADATA: b"peer-metadata", |
| 204 | ConnectMsg.BASE_ADDR: 0x2000, |
| 205 | ConnectMsg.NUM_BLOCKS: 16, |
| 206 | ConnectMsg.BLOCK_LEN: block_len, |
| 207 | ConnectMsg.HASH_SEED: hash_seed, |
| 208 | } |
| 209 | if fingerprint is not None: |
| 210 | msg[ConnectMsg.CONFIG_FINGERPRINT] = fingerprint |
| 211 | return msg |
| 212 | |
| 213 | |
| 214 | class FakeParent: |
| 215 | """Configurable :class:`ParentManager` for server-role tests. |
| 216 | |
| 217 | ``stored`` is the dict of ready blocks (key → primary block_id). |
| 218 | ``pending`` and ``retry`` script the first lookup() result for those |
| 219 | keys; subsequent lookups behave normally (a key that promised |
| 220 | HIT_PENDING / RETRY can later be promoted to HIT by adding it to |
| 221 | ``stored`` and removing it from ``pending``/``retry``). ``calls`` |
| 222 | captures every parent invocation in order for assertions. |
| 223 | |
| 224 | Injected per-step via ``session.serve_external_requests(parent)`` — |
| 225 | not held by the session, matching how ``TieringOffloadingManager`` |
| 226 | hands the tier a handle valid only for that call. |
| 227 | """ |
| 228 | |
| 229 | def __init__( |
| 230 | self, |
| 231 | stored: dict[OffloadKey, int] | None = None, |
| 232 | pending: set[OffloadKey] | None = None, |
| 233 | retry: set[OffloadKey] | None = None, |
| 234 | ) -> None: |
| 235 | self.stored: dict[OffloadKey, int] = dict(stored or {}) |
| 236 | self.pending: set[OffloadKey] = set(pending or ()) |
| 237 | self.retry: set[OffloadKey] = set(retry or ()) |
| 238 | self._next_job_id: int = 1000 |
| 239 | self.calls: list[tuple] = [] |
| 240 | |
| 241 | def on_new_request(self, ctx: ReqContext) -> RequestOffloadingContext: |
| 242 | self.calls.append(("on_new_request", ctx.req_id)) |
| 243 | return RequestOffloadingContext() |
| 244 | |
| 245 | def lookup(self, key: OffloadKey, ctx: ReqContext) -> LookupResult: |
| 246 | self.calls.append(("lookup", key, ctx.req_id)) |
| 247 | if key in self.pending: |
| 248 | return LookupResult.HIT_PENDING |
| 249 | if key in self.retry: |
| 250 | return LookupResult.RETRY |
| 251 | if key in self.stored: |
| 252 | return LookupResult.HIT |
| 253 | return LookupResult.MISS |
| 254 | |
| 255 | def create_store_job( |
| 256 | self, |
| 257 | keys: Sequence[OffloadKey], |
| 258 | ctx: ReqContext, |
| 259 | ) -> TransferJob: |
| 260 | keys_list = list(keys) |
| 261 | self.calls.append(("create_store_job", tuple(keys_list), ctx.req_id)) |
| 262 | chunk_ids = np.array([self.stored[k] for k in keys_list], dtype=np.int32) |
| 263 | job_id = self._next_job_id |
| 264 | self._next_job_id += 1 |
| 265 | return TransferJob( |
| 266 | job_id=job_id, |
| 267 | keys=keys_list, |
| 268 | chunk_ids=chunk_ids, |
| 269 | is_promotion=False, |
| 270 | req_context=ctx, |
| 271 | ) |
| 272 | |
| 273 | def on_request_finished(self, ctx: ReqContext) -> None: |
| 274 | self.calls.append(("on_request_finished", ctx.req_id)) |
| 275 | |
| 276 | |
| 277 | def _make_session( |
| 278 | conn: FakeConnection | None = None, |
| 279 | transport: FakeDataTransport | None = None, |
| 280 | peer_id: str = "peer:8000", |
| 281 | local_id: str = "local:9000", |
| 282 | local_hash_seed: str = _DEFAULT_HASH_SEED, |
| 283 | ) -> tuple[P2PSession, FakeConnection, FakeDataTransport]: |
| 284 | if conn is None: |
| 285 | conn = FakeConnection(peer_id=peer_id) |
| 286 | if transport is None: |
| 287 | transport = FakeDataTransport() |
| 288 | session = P2PSession( |
| 289 | peer_id=peer_id, |
| 290 | local_id=local_id, |
| 291 | transport=transport, |
| 292 | local_block_len=transport.block_len, |
| 293 | local_hash_seed=local_hash_seed, |
| 294 | conn=conn, |
| 295 | ) |
| 296 | return session, conn, transport |
| 297 | |
| 298 | |
| 299 | def _serve(session: P2PSession, parent: FakeParent) -> None: |
| 300 | """Resolve enqueued inbound lookups, as the manager does each step.""" |
| 301 | session.serve_external_requests(parent) |
| 302 | |
| 303 | |
| 304 | def _activate( |
| 305 | session: P2PSession, conn: FakeConnection, peer_id: str = "peer:8000" |
| 306 | ) -> None: |
| 307 | """Drive the handshake: peer sends ConnectMsg + ConnectAckMsg.""" |
| 308 | conn.enqueue(_peer_connect_msg(peer_id=peer_id)) |
| 309 | conn.enqueue({TYPE_KEY: ConnectAckMsg.TYPE, ConnectAckMsg.PEER_ID: peer_id}) |
| 310 | session.poll() |
| 311 | |
| 312 | |
| 313 | # --- Accessors for the server role's consolidated per-request state. |
| 314 | # ServerRole keeps one _ServerRequestState per kv_request_id; entries are |
| 315 | # garbage-collected once fully idle, so "no outbound"/"no abort" reads as |
| 316 | # either a missing entry or a None field. These helpers paper over that. |
| 317 | |
| 318 | |
| 319 | def _client_load(session: P2PSession, kv_request_id: str): |
| 320 | """The single in-flight load of a kv_request_id (loads are per-round).""" |
| 321 | loads = session._client._requests[kv_request_id].loads |
| 322 | assert len(loads) == 1 |
| 323 | return next(iter(loads.values())) |
| 324 | |
| 325 | |
| 326 | def _srv_outbound(session: P2PSession, kv_request_id: str): |
| 327 | """Serve-side round for a kv_request_id, or None (idle / GC'd). |
| 328 | |
| 329 | Rounds are keyed by wire round_seq; surfaces the demanded round when |
| 330 | a fetch has bound one, else any parked supply round. |
| 331 | """ |
| 332 | st = session._server._requests.get(kv_request_id) |
| 333 | if st is None or not st.outbound: |
| 334 | return None |
| 335 | for rnd in st.outbound.values(): |
| 336 | if rnd.demand_received: |
| 337 | return rnd |
| 338 | return next(iter(st.outbound.values())) |
| 339 | |
| 340 | |
| 341 | def _srv_lookups(session: P2PSession) -> list: |
| 342 | """Every parked inbound _ActiveLookup across all requests.""" |
| 343 | return [ |
| 344 | lu for st in session._server._requests.values() for lu in st.lookups.values() |
| 345 | ] |
| 346 | |
| 347 | |
| 348 | def _srv_abort_started(session: P2PSession, kv_request_id: str) -> float | None: |
| 349 | """Pending-abort start time for a kv_request_id, or None.""" |
| 350 | for (kv, _), started in session._server._pending_aborts.items(): |
| 351 | if kv == kv_request_id: |
| 352 | return started |
| 353 | return None |
| 354 | |
| 355 | |
| 356 | def _srv_inflight_count(session: P2PSession, kv_request_id: str) -> int: |
| 357 | """Inflight-transfer count tracked for a kv_request_id (0 if idle).""" |
| 358 | st = session._server._requests.get(kv_request_id) |
| 359 | return len(st.inflight_tids) if st is not None else 0 |
| 360 | |
| 361 | |
| 362 | def _srv_total_inflight(session: P2PSession) -> int: |
| 363 | """Sum of per-request inflight counts across all requests.""" |
| 364 | return sum(len(st.inflight_tids) for st in session._server._requests.values()) |
| 365 | |
| 366 | |
| 367 | # --------------------------------------------------------------------------- |
| 368 | # Connect / handshake |
| 369 | # --------------------------------------------------------------------------- |
| 370 | |
| 371 | |
| 372 | class TestConnectHandshake: |
| 373 | def test_connect_msg_sent_on_creation(self): |
| 374 | """Session sends its own ConnectMsg on connection.""" |
| 375 | session, conn, _ = _make_session() |
| 376 | assert len(conn._sent) == 1 |
| 377 | msg = conn._sent[0] |
| 378 | assert msg[TYPE_KEY] == ConnectMsg.TYPE |
| 379 | assert msg[ConnectMsg.PEER_ID] == "local:9000" |
| 380 | assert msg[ConnectMsg.NUM_BLOCKS] == 16 |
| 381 | assert msg[ConnectMsg.BLOCK_LEN] == 4096 |
| 382 | assert ConnectMsg.AGENT_METADATA in msg |
| 383 | |
| 384 | def test_peer_connect_triggers_add_remote_and_ack(self): |
| 385 | """Receiving ConnectMsg registers peer and replies with ConnectAck.""" |
| 386 | session, conn, transport = _make_session() |
| 387 | conn.enqueue(_peer_connect_msg()) |
| 388 | session.poll() |
| 389 | assert "peer:8000" in transport._remote_peers |
| 390 | ack = next(m for m in conn._sent if m[TYPE_KEY] == ConnectAckMsg.TYPE) |
| 391 | assert ack[ConnectAckMsg.PEER_ID] == "local:9000" |
| 392 | |
| 393 | def test_connect_ack_makes_session_ready(self): |
| 394 | """Session.ready becomes True after ConnectAckMsg.""" |
| 395 | session, conn, _ = _make_session() |
| 396 | assert not session.ready |
| 397 | conn.enqueue({TYPE_KEY: ConnectAckMsg.TYPE, ConnectAckMsg.PEER_ID: "peer:8000"}) |
| 398 | session.poll() |
| 399 | assert session.ready |
| 400 | |
| 401 | def test_messages_queued_before_ack_flush_on_ack(self): |
| 402 | """Outgoing messages sent before ConnectAck are flushed after.""" |
| 403 | session, conn, _ = _make_session() |
| 404 | session.request_blocks( |
| 405 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 406 | ) |
| 407 | # Before ack: only our ConnectMsg was sent. |
| 408 | assert len(conn._sent) == 1 |
| 409 | assert conn._sent[0][TYPE_KEY] == ConnectMsg.TYPE |
| 410 | # Ack arrives. |
| 411 | conn.enqueue({TYPE_KEY: ConnectAckMsg.TYPE, ConnectAckMsg.PEER_ID: "peer:8000"}) |
| 412 | session.poll() |
| 413 | # Queued fetch is now sent. |
| 414 | assert any(m[TYPE_KEY] == FetchMsg.TYPE for m in conn._sent) |
| 415 | |
| 416 | def test_block_len_mismatch_marks_dead(self): |
| 417 | """Mismatched block_len rejects peer and marks connection dead.""" |
| 418 | session, conn, transport = _make_session() |
| 419 | conn.enqueue(_peer_connect_msg(block_len=8192)) # mismatch |
| 420 | session.poll() |
| 421 | assert "peer:8000" not in transport._remote_peers |
| 422 | assert not session.alive |
| 423 | |
| 424 | def test_config_fingerprint_mismatch_marks_dead(self): |
| 425 | """Mismatched config fingerprint rejects peer.""" |
| 426 | transport = FakeDataTransport(config_fingerprint="abc123") |
| 427 | session, conn, _ = _make_session(transport=transport) |
| 428 | conn.enqueue(_peer_connect_msg(fingerprint="different")) |
| 429 | session.poll() |
| 430 | assert "peer:8000" not in transport._remote_peers |
| 431 | assert not session.alive |
| 432 | |
| 433 | def test_config_fingerprint_match_succeeds(self): |
| 434 | """Matching fingerprints register the peer.""" |
| 435 | transport = FakeDataTransport(config_fingerprint="same_fp") |
| 436 | session, conn, _ = _make_session(transport=transport) |
| 437 | conn.enqueue(_peer_connect_msg(fingerprint="same_fp")) |
| 438 | session.poll() |
| 439 | assert "peer:8000" in transport._remote_peers |
| 440 | assert session.alive |
| 441 | |
| 442 | def test_missing_fingerprint_allowed(self): |
| 443 | """Missing fingerprint on either side is allowed.""" |
| 444 | transport = FakeDataTransport(config_fingerprint="abc123") |
| 445 | session, conn, _ = _make_session(transport=transport) |
| 446 | conn.enqueue(_peer_connect_msg()) # no fingerprint |
| 447 | session.poll() |
| 448 | assert "peer:8000" in transport._remote_peers |
| 449 | |
| 450 | def test_hash_seed_mismatch_marks_dead(self): |
| 451 | """Mismatched PYTHONHASHSEED rejects peer and marks connection dead.""" |
| 452 | session, conn, transport = _make_session(local_hash_seed="0") |
| 453 | conn.enqueue(_peer_connect_msg(hash_seed="12345")) # mismatch |
| 454 | session.poll() |
| 455 | assert "peer:8000" not in transport._remote_peers |
| 456 | assert not session.alive |
| 457 | assert not any(m[TYPE_KEY] == ConnectAckMsg.TYPE for m in conn._sent) |
| 458 | |
| 459 | def test_hash_seed_match_succeeds(self): |
| 460 | """Matching PYTHONHASHSEED registers the peer and acks.""" |
| 461 | session, conn, transport = _make_session(local_hash_seed="12345") |
| 462 | conn.enqueue(_peer_connect_msg(hash_seed="12345")) |
| 463 | session.poll() |
| 464 | assert "peer:8000" in transport._remote_peers |
| 465 | assert session.alive |
| 466 | assert any(m[TYPE_KEY] == ConnectAckMsg.TYPE for m in conn._sent) |
| 467 | |
| 468 | def test_hash_seed_advertised_in_connect_msg(self): |
| 469 | """Session advertises its own PYTHONHASHSEED in the ConnectMsg.""" |
| 470 | _, conn, _ = _make_session(local_hash_seed="777") |
| 471 | assert conn._sent[0][ConnectMsg.HASH_SEED] == "777" |
| 472 | |
| 473 | |
| 474 | # --------------------------------------------------------------------------- |
| 475 | # Client-role flows |
| 476 | # --------------------------------------------------------------------------- |
| 477 | |
| 478 | |
| 479 | class TestClientFlows: |
| 480 | def test_request_blocks_sends_fetch(self): |
| 481 | session, conn, _ = _make_session() |
| 482 | _activate(session, conn) |
| 483 | session.request_blocks( |
| 484 | job_id=1, kv_request_id="req-1", keys=[b"k1", b"k2"], block_ids=[0, 1] |
| 485 | ) |
| 486 | lookup = conn._sent[-1] |
| 487 | assert lookup[TYPE_KEY] == FetchMsg.TYPE |
| 488 | assert lookup[FetchMsg.KV_REQUEST_ID] == "req-1" |
| 489 | assert lookup[FetchMsg.KEYS] == [b"k1", b"k2"] |
| 490 | assert lookup[FetchMsg.BLOCK_INDEXES] == [0, 1] |
| 491 | |
| 492 | def test_transfer_done_success(self): |
| 493 | session, conn, _ = _make_session() |
| 494 | _activate(session, conn) |
| 495 | session.request_blocks( |
| 496 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 497 | ) |
| 498 | conn.enqueue( |
| 499 | { |
| 500 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 501 | TransferDoneMsg.ROUND_SEQ: 0, |
| 502 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 503 | TransferDoneMsg.SUCCESS: True, |
| 504 | } |
| 505 | ) |
| 506 | loads = session.poll().loads |
| 507 | assert loads == [LoadResult(job_id=1, kv_request_id="req-1", success=True)] |
| 508 | |
| 509 | def test_transfer_done_failure(self): |
| 510 | session, conn, _ = _make_session() |
| 511 | _activate(session, conn) |
| 512 | session.request_blocks( |
| 513 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 514 | ) |
| 515 | conn.enqueue( |
| 516 | { |
| 517 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 518 | TransferDoneMsg.ROUND_SEQ: 0, |
| 519 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 520 | TransferDoneMsg.SUCCESS: False, |
| 521 | } |
| 522 | ) |
| 523 | loads = session.poll().loads |
| 524 | assert loads == [LoadResult(job_id=1, kv_request_id="req-1", success=False)] |
| 525 | |
| 526 | def test_finish_request_sends_abort(self): |
| 527 | session, conn, _ = _make_session() |
| 528 | _activate(session, conn) |
| 529 | session.request_blocks( |
| 530 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 531 | ) |
| 532 | session.finish_request("req-1") |
| 533 | abort = conn._sent[-1] |
| 534 | assert abort[TYPE_KEY] == AbortFetchMsg.TYPE |
| 535 | assert abort[AbortFetchMsg.KV_REQUEST_ID] == "req-1" |
| 536 | |
| 537 | def test_finish_preserves_active_load_until_terminal_result(self): |
| 538 | session, conn, _ = _make_session() |
| 539 | _activate(session, conn) |
| 540 | session.request_blocks( |
| 541 | job_id=7, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 542 | ) |
| 543 | |
| 544 | session.finish_request("req-1") |
| 545 | |
| 546 | load = _client_load(session, "req-1") |
| 547 | assert load.job_id == 7 |
| 548 | assert load.aborted_at is not None |
| 549 | aborted_at = load.aborted_at |
| 550 | assert session._client.has_active_loads is True |
| 551 | session.finish_request("req-1") |
| 552 | assert _client_load(session, "req-1").aborted_at == aborted_at |
| 553 | aborts = [m for m in conn._sent if m[TYPE_KEY] == AbortFetchMsg.TYPE] |
| 554 | assert len(aborts) == 1 |
| 555 | assert aborts[0][AbortFetchMsg.ROUND_SEQ] == 0 |
| 556 | |
| 557 | def test_finish_abort_ack_emits_one_failure(self): |
| 558 | session, conn, _ = _make_session() |
| 559 | _activate(session, conn) |
| 560 | session.request_blocks( |
| 561 | job_id=8, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 562 | ) |
| 563 | session.finish_request("req-1") |
| 564 | |
| 565 | conn.enqueue( |
| 566 | { |
| 567 | TYPE_KEY: AbortAckMsg.TYPE, |
| 568 | AbortAckMsg.ROUND_SEQ: 0, |
| 569 | AbortAckMsg.KV_REQUEST_ID: "req-1", |
| 570 | } |
| 571 | ) |
| 572 | loads = session.poll().loads |
| 573 | assert loads == [LoadResult(job_id=8, kv_request_id="req-1", success=False)] |
| 574 | assert session._client.has_active_loads is False |
| 575 | assert "req-1" not in session._client._requests |
| 576 | assert session.poll().loads == [] |
| 577 | |
| 578 | def test_finish_abort_timeout_emits_one_failure(self): |
| 579 | session, conn, _ = _make_session() |
| 580 | _activate(session, conn) |
| 581 | session.request_blocks( |
| 582 | job_id=9, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 583 | ) |
| 584 | session.finish_request("req-1") |
| 585 | _client_load(session, "req-1").aborted_at = ( |
| 586 | time.monotonic() - _ABORT_ACK_TIMEOUT_S - 1.0 |
| 587 | ) |
| 588 | |
| 589 | loads = session.poll().loads |
| 590 | assert loads == [LoadResult(job_id=9, kv_request_id="req-1", success=False)] |
| 591 | assert session._client.has_active_loads is False |
| 592 | assert "req-1" not in session._client._requests |
| 593 | assert session.poll().loads == [] |
| 594 | |
| 595 | def test_late_transfer_done_after_abort_ack_is_ignored(self): |
| 596 | session, conn, _ = _make_session() |
| 597 | _activate(session, conn) |
| 598 | session.request_blocks( |
| 599 | job_id=10, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 600 | ) |
| 601 | session.finish_request("req-1") |
| 602 | |
| 603 | conn.enqueue( |
| 604 | { |
| 605 | TYPE_KEY: AbortAckMsg.TYPE, |
| 606 | AbortAckMsg.ROUND_SEQ: 0, |
| 607 | AbortAckMsg.KV_REQUEST_ID: "req-1", |
| 608 | } |
| 609 | ) |
| 610 | assert session.poll().loads == [ |
| 611 | LoadResult(job_id=10, kv_request_id="req-1", success=False) |
| 612 | ] |
| 613 | |
| 614 | conn.enqueue( |
| 615 | { |
| 616 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 617 | TransferDoneMsg.ROUND_SEQ: 0, |
| 618 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 619 | TransferDoneMsg.SUCCESS: True, |
| 620 | } |
| 621 | ) |
| 622 | assert session.poll().loads == [] |
| 623 | assert session._client.has_active_loads is False |
| 624 | |
| 625 | def test_active_loads_work_list_tracks_in_flight(self): |
| 626 | """collect_results / has_active_loads use the _active_loads work-list, |
| 627 | armed when a fetch is issued and discarded exactly when its load |
| 628 | clears — a probe-only request never enters it, and completion empties |
| 629 | it while the entry may briefly linger for GC.""" |
| 630 | session, conn, _ = _make_session() |
| 631 | _activate(session, conn) |
| 632 | client = session._client |
| 633 | |
| 634 | # A probe-only request has no in-flight load: not in _active_loads. |
| 635 | session.register_lookup("req-probe", b"hp") |
| 636 | assert client._active_loads == set() |
| 637 | assert client.has_active_loads is False |
| 638 | |
| 639 | # Issuing a fetch arms the work-list. |
| 640 | session.request_blocks( |
| 641 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 642 | ) |
| 643 | assert client._active_loads == {"req-1"} |
| 644 | assert client.has_active_loads is True |
| 645 | |
| 646 | # Completion clears the load and discards it from the work-list. |
| 647 | conn.enqueue( |
| 648 | { |
| 649 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 650 | TransferDoneMsg.ROUND_SEQ: 0, |
| 651 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 652 | TransferDoneMsg.SUCCESS: True, |
| 653 | } |
| 654 | ) |
| 655 | session.poll() |
| 656 | assert client._active_loads == set() |
| 657 | assert client.has_active_loads is False |
| 658 | |
| 659 | def test_load_timeout_sends_abort(self): |
| 660 | session, conn, _ = _make_session() |
| 661 | _activate(session, conn) |
| 662 | session.request_blocks( |
| 663 | job_id=1, kv_request_id="req-1", keys=[b"k"], block_ids=[0] |
| 664 | ) |
| 665 | _client_load(session, "req-1").submitted_at = time.monotonic() - 60.0 |
| 666 | session.poll() |
| 667 | abort = conn._sent[-1] |
| 668 | assert abort[TYPE_KEY] == AbortFetchMsg.TYPE |
| 669 | |
| 670 | def test_load_abort_ack_timeout_surfaces_failure(self): |
| 671 | """After load timeout sends AbortFetch, if no AbortAck arrives within |
| 672 | _ABORT_ACK_TIMEOUT_S the request is surfaced as failed and removed |
| 673 | from _requests — the engine cannot wait forever on a peer that won't |
| 674 | ack. |
| 675 | """ |
| 676 | session, conn, _ = _make_session() |
| 677 | _activate(session, conn) |
| 678 | session.request_blocks( |
| 679 | job_id=7, kv_request_id="req-7", keys=[b"k"], block_ids=[0] |
| 680 | ) |
| 681 | # 1) Trip the load timeout to send AbortFetch and stamp aborted_at. |
| 682 | _client_load(session, "req-7").submitted_at = ( |
| 683 | time.monotonic() - _LOAD_TIMEOUT_S - 1.0 |
| 684 | ) |
| 685 | loads = session.poll().loads |
| 686 | assert loads == [] |
| 687 | assert any( |
| 688 | m.get(TYPE_KEY) == AbortFetchMsg.TYPE |
| 689 | and m[AbortFetchMsg.KV_REQUEST_ID] == "req-7" |
| 690 | for m in conn._sent |
| 691 | ) |
| 692 | assert _client_load(session, "req-7").aborted_at is not None |
| 693 | |
| 694 | # 2) Now backdate aborted_at past the abort-ack timeout. No ack ever |
| 695 | # arrived from the peer. |
| 696 | _client_load(session, "req-7").aborted_at = ( |
| 697 | time.monotonic() - _ABORT_ACK_TIMEOUT_S - 1.0 |
| 698 | ) |
| 699 | loads = session.poll().loads |
| 700 | assert loads == [LoadResult(job_id=7, kv_request_id="req-7", success=False)] |
| 701 | assert "req-7" not in session._client._requests |
| 702 | |
| 703 | def test_load_abort_ack_clears_request(self): |
| 704 | """After load timeout sends AbortFetch, an arriving AbortAckMsg from |
| 705 | the peer surfaces the failure cleanly and removes the request from |
| 706 | _requests — covers the on_abort_ack arrival path.""" |
| 707 | session, conn, _ = _make_session() |
| 708 | _activate(session, conn) |
| 709 | session.request_blocks( |
| 710 | job_id=8, kv_request_id="req-8", keys=[b"k"], block_ids=[0] |
| 711 | ) |
| 712 | _client_load(session, "req-8").submitted_at = ( |
| 713 | time.monotonic() - _LOAD_TIMEOUT_S - 1.0 |
| 714 | ) |
| 715 | # First poll: AbortFetch goes out. |
| 716 | session.poll() |
| 717 | assert _client_load(session, "req-8").aborted_at is not None |
| 718 | |
| 719 | # Peer acks the abort. |
| 720 | conn.enqueue( |
| 721 | { |
| 722 | TYPE_KEY: AbortAckMsg.TYPE, |
| 723 | AbortAckMsg.ROUND_SEQ: 0, |
| 724 | AbortAckMsg.KV_REQUEST_ID: "req-8", |
| 725 | } |
| 726 | ) |
| 727 | loads = session.poll().loads |
| 728 | assert loads == [LoadResult(job_id=8, kv_request_id="req-8", success=False)] |
| 729 | assert "req-8" not in session._client._requests |
| 730 | |
| 731 | |
| 732 | # --------------------------------------------------------------------------- |
| 733 | # Symmetric-P2P lookup flow (do_p2p_fetch) |
| 734 | # --------------------------------------------------------------------------- |
| 735 | |
| 736 | |
| 737 | class TestLookupFlow: |
| 738 | """Consumer-side state machine for do_p2p_fetch lookups.""" |
| 739 | |
| 740 | def test_aggregate_flush_resolve_round_trip(self): |
| 741 | """register_lookup → flush sends one LookupMsg → response |
| 742 | resolves entries → register_lookup returns the cached bool on |
| 743 | every call, and repeat probes never re-issue a LookupMsg.""" |
| 744 | session, conn, _ = _make_session() |
| 745 | _activate(session, conn) |
| 746 | |
| 747 | # Aggregate two keys for the same kv_request_id; both return None. |
| 748 | assert session.register_lookup("req-1", b"hA") is None |
| 749 | assert session.register_lookup("req-1", b"hB") is None |
| 750 | |
| 751 | # Flush sends one LookupMsg with both keys. |
| 752 | sent_before = len(conn._sent) |
| 753 | session.flush_pending_lookups() |
| 754 | new = conn._sent[sent_before:] |
| 755 | assert len(new) == 1 |
| 756 | msg = new[0] |
| 757 | assert msg[TYPE_KEY] == LookupMsg.TYPE |
| 758 | assert msg[LookupMsg.KV_REQUEST_ID] == "req-1" |
| 759 | assert sorted(msg[LookupMsg.KEYS]) == [b"hA", b"hB"] |
| 760 | |
| 761 | # Idempotent re-flush sends nothing — the entries are now in-flight. |
| 762 | sent_before = len(conn._sent) |
| 763 | session.flush_pending_lookups() |
| 764 | assert conn._sent[sent_before:] == [] |
| 765 | |
| 766 | # While in-flight, register_lookup keeps returning None. |
| 767 | assert session.register_lookup("req-1", b"hA") is None |
| 768 | |
| 769 | # Peer answers: hA hit, hB miss. |
| 770 | conn.enqueue( |
| 771 | { |
| 772 | TYPE_KEY: LookupRespMsg.TYPE, |
| 773 | LookupRespMsg.KV_REQUEST_ID: "req-1", |
| 774 | LookupRespMsg.KEYS: [b"hA", b"hB"], |
| 775 | LookupRespMsg.HITS: [True, False], |
| 776 | } |
| 777 | ) |
| 778 | session.poll() |
| 779 | |
| 780 | # register_lookup returns the resolved bool. |
| 781 | assert session.register_lookup("req-1", b"hA") is True |
| 782 | assert session.register_lookup("req-1", b"hB") is False |
| 783 | # The entry is cached, not popped: repeat probes keep returning the |
| 784 | # same result and never re-queue the key, so a flush sends nothing. |
| 785 | assert session.register_lookup("req-1", b"hA") is True |
| 786 | assert session.register_lookup("req-1", b"hB") is False |
| 787 | sent_before = len(conn._sent) |
| 788 | session.flush_pending_lookups() |
| 789 | assert [ |
| 790 | m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE |
| 791 | ] == [] |
| 792 | |
| 793 | def test_request_blocks_clears_probe_cache(self): |
| 794 | """A resolved HIT probe is popped when its fetch is issued, so a |
| 795 | re-scheduled request re-probes instead of trusting the stale True |
| 796 | (the served block is unpinned and may have been evicted).""" |
| 797 | session, conn, _ = _make_session() |
| 798 | _activate(session, conn) |
| 799 | |
| 800 | # Probe hA, flush, and let the peer resolve it to a HIT. |
| 801 | assert session.register_lookup("req-1", b"hA") is None |
| 802 | session.flush_pending_lookups() |
| 803 | conn.enqueue( |
| 804 | { |
| 805 | TYPE_KEY: LookupRespMsg.TYPE, |
| 806 | LookupRespMsg.KV_REQUEST_ID: "req-1", |
| 807 | LookupRespMsg.KEYS: [b"hA"], |
| 808 | LookupRespMsg.HITS: [True], |
| 809 | } |
| 810 | ) |
| 811 | session.poll() |
| 812 | assert session.register_lookup("req-1", b"hA") is True |
| 813 | |
| 814 | # Fetch consumes the probe. |
| 815 | session.request_blocks( |
| 816 | job_id=1, kv_request_id="req-1", keys=[b"hA"], block_ids=[0] |
| 817 | ) |
| 818 | assert b"hA" not in session._client._requests["req-1"].probes |
| 819 | |
| 820 | # Re-scheduled probe of the same key is treated as brand-new: it |
| 821 | # returns None and re-queues, so a flush emits a fresh LookupMsg. |
| 822 | assert session.register_lookup("req-1", b"hA") is None |
| 823 | sent_before = len(conn._sent) |
| 824 | session.flush_pending_lookups() |
| 825 | fresh = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE] |
| 826 | assert len(fresh) == 1 |
| 827 | assert fresh[0][LookupMsg.KEYS] == [b"hA"] |
| 828 | |
| 829 | def test_flush_uses_work_list_not_full_scan(self): |
| 830 | """Flush drains a work-list rather than scanning every live request. |
| 831 | |
| 832 | A request with no newly-registered keys is not revisited: after a |
| 833 | flush the work-list is empty, an idle re-flush sends nothing, and a |
| 834 | subsequent register re-arms exactly the one affected id — even while |
| 835 | an unrelated request stays live in ``_requests``. |
| 836 | """ |
| 837 | session, conn, _ = _make_session() |
| 838 | _activate(session, conn) |
| 839 | client = session._client |
| 840 | |
| 841 | # Two requests register keys; both are queued for flush. |
| 842 | session.register_lookup("req-A", b"hA") |
| 843 | session.register_lookup("req-B", b"hB") |
| 844 | assert client._flush_pending == {"req-A", "req-B"} |
| 845 | |
| 846 | # Flush drains the work-list even though both requests stay live. |
| 847 | session.flush_pending_lookups() |
| 848 | assert client._flush_pending == set() |
| 849 | assert set(client._requests) == {"req-A", "req-B"} |
| 850 | |
| 851 | # An idle re-flush visits nothing and sends no LookupMsg. |
| 852 | sent_before = len(conn._sent) |
| 853 | session.flush_pending_lookups() |
| 854 | assert [ |
| 855 | m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE |
| 856 | ] == [] |
| 857 | |
| 858 | # A new register re-arms only that id. |
| 859 | session.register_lookup("req-A", b"hA2") |
| 860 | assert client._flush_pending == {"req-A"} |
| 861 | |
| 862 | def test_separate_lookup_msg_per_kv_request_id(self): |
| 863 | """Hashes for different kv_request_ids flush as separate LookupMsgs.""" |
| 864 | session, conn, _ = _make_session() |
| 865 | _activate(session, conn) |
| 866 | |
| 867 | session.register_lookup("req-A", b"h1") |
| 868 | session.register_lookup("req-B", b"h2") |
| 869 | session.register_lookup("req-A", b"h3") |
| 870 | |
| 871 | sent_before = len(conn._sent) |
| 872 | session.flush_pending_lookups() |
| 873 | sent = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE] |
| 874 | assert len(sent) == 2 |
| 875 | by_req = {m[LookupMsg.KV_REQUEST_ID]: m[LookupMsg.KEYS] for m in sent} |
| 876 | assert sorted(by_req["req-A"]) == [b"h1", b"h3"] |
| 877 | assert by_req["req-B"] == [b"h2"] |
| 878 | |
| 879 | def test_multiple_lookup_msgs_across_steps(self): |
| 880 | """A request's block set may be discovered across scheduler steps: |
| 881 | each step that registers new keys flushes its own LookupMsg for |
| 882 | the same kv_request_id, carrying only the newly-probed keys.""" |
| 883 | session, conn, _ = _make_session() |
| 884 | _activate(session, conn) |
| 885 | |
| 886 | # Step 1: probe hA, hB. |
| 887 | session.register_lookup("req-1", b"hA") |
| 888 | session.register_lookup("req-1", b"hB") |
| 889 | sent_before = len(conn._sent) |
| 890 | session.flush_pending_lookups() |
| 891 | first = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE] |
| 892 | assert len(first) == 1 |
| 893 | assert first[0][LookupMsg.KV_REQUEST_ID] == "req-1" |
| 894 | assert sorted(first[0][LookupMsg.KEYS]) == [b"hA", b"hB"] |
| 895 | |
| 896 | # Step 2: a new key is discovered for the same request. The |
| 897 | # in-flight keys from step 1 are not re-sent; a second LookupMsg |
| 898 | # goes out carrying only the newly-probed key. |
| 899 | assert session.register_lookup("req-1", b"hA") is None # in-flight no-op |
| 900 | session.register_lookup("req-1", b"hC") |
| 901 | sent_before = len(conn._sent) |
| 902 | session.flush_pending_lookups() |
| 903 | second = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupMsg.TYPE] |
| 904 | assert len(second) == 1 |
| 905 | assert second[0][LookupMsg.KV_REQUEST_ID] == "req-1" |
| 906 | assert second[0][LookupMsg.KEYS] == [b"hC"] |
| 907 | |
| 908 | def test_split_response_resolves_across_messages(self): |
| 909 | """Producer may answer one LookupMsg's keys across multiple |
| 910 | LookupRespMsgs — pairs are self-describing so each lands.""" |
| 911 | session, conn, _ = _make_session() |
| 912 | _activate(session, conn) |
| 913 | |
| 914 | session.register_lookup("req-1", b"hA") |
| 915 | session.register_lookup("req-1", b"hB") |
| 916 | session.flush_pending_lookups() |
| 917 | |
| 918 | # Two responses, each carrying one of the two keys. |
| 919 | conn.enqueue( |
| 920 | { |
| 921 | TYPE_KEY: LookupRespMsg.TYPE, |
| 922 | LookupRespMsg.KV_REQUEST_ID: "req-1", |
| 923 | LookupRespMsg.KEYS: [b"hA"], |
| 924 | LookupRespMsg.HITS: [True], |
| 925 | } |
| 926 | ) |
| 927 | conn.enqueue( |
| 928 | { |
| 929 | TYPE_KEY: LookupRespMsg.TYPE, |
| 930 | LookupRespMsg.KV_REQUEST_ID: "req-1", |
| 931 | LookupRespMsg.KEYS: [b"hB"], |
| 932 | LookupRespMsg.HITS: [False], |
| 933 | } |
| 934 | ) |
| 935 | session.poll() |
| 936 | |
| 937 | assert session.register_lookup("req-1", b"hA") is True |
| 938 | assert session.register_lookup("req-1", b"hB") is False |
| 939 | |
| 940 | def test_finish_request_cancels_pending_lookups(self): |
| 941 | """finish_request drops every pending lookup for the kv_request_id.""" |
| 942 | session, conn, _ = _make_session() |
| 943 | _activate(session, conn) |
| 944 | |
| 945 | session.register_lookup("req-1", b"hA") |
| 946 | session.register_lookup("req-1", b"hB") |
| 947 | session.register_lookup("req-2", b"hC") |
| 948 | session.finish_request("req-1") |
| 949 | |
| 950 | # req-1 entries gone, req-2 untouched. |
| 951 | assert "req-1" not in session._client._requests |
| 952 | assert b"hC" in session._client._requests["req-2"].probes |
| 953 | |
| 954 | def test_finish_after_flushed_lookup_sends_empty_fetch(self): |
| 955 | """LookupMsg flushed but no FetchMsg sent (all-miss case) → |
| 956 | finish_request emits an empty FetchMsg so the peer can drop its |
| 957 | lookup state and call parent.on_request_finished.""" |
| 958 | session, conn, _ = _make_session() |
| 959 | _activate(session, conn) |
| 960 | |
| 961 | session.register_lookup("req-1", b"hA") |
| 962 | session.flush_pending_lookups() |
| 963 | sent_before = len(conn._sent) |
| 964 | |
| 965 | session.finish_request("req-1") |
| 966 | |
| 967 | fetches = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == FetchMsg.TYPE] |
| 968 | assert len(fetches) == 1 |
| 969 | assert fetches[0][FetchMsg.KV_REQUEST_ID] == "req-1" |
| 970 | assert fetches[0][FetchMsg.KEYS] == [] |
| 971 | assert fetches[0][FetchMsg.BLOCK_INDEXES] == [] |
| 972 | |
| 973 | def test_finish_without_flushed_lookup_sends_no_fetch(self): |
| 974 | """No LookupMsg was ever sent → finish_request must not emit an |
| 975 | empty FetchMsg (the peer has no state to release).""" |
| 976 | session, conn, _ = _make_session() |
| 977 | _activate(session, conn) |
| 978 | |
| 979 | # Register but never flush. |
| 980 | session.register_lookup("req-1", b"hA") |
| 981 | sent_before = len(conn._sent) |
| 982 | |
| 983 | session.finish_request("req-1") |
| 984 | |
| 985 | fetches = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == FetchMsg.TYPE] |
| 986 | assert fetches == [] |
| 987 | |
| 988 | def test_finish_after_real_fetch_sends_no_second_fetch(self): |
| 989 | """A real FetchMsg was already sent for the id → finish_request |
| 990 | must not emit a second (empty) FetchMsg.""" |
| 991 | session, conn, _ = _make_session() |
| 992 | _activate(session, conn) |
| 993 | |
| 994 | session.register_lookup("req-1", b"hA") |
| 995 | session.flush_pending_lookups() |
| 996 | # Resolve the probe to a HIT before fetching, as the manager only |
| 997 | # loads confirmed hits (an unresolved probe yields RETRY). |
| 998 | conn.enqueue( |
| 999 | { |
| 1000 | TYPE_KEY: LookupRespMsg.TYPE, |
| 1001 | LookupRespMsg.KV_REQUEST_ID: "req-1", |
| 1002 | LookupRespMsg.KEYS: [b"hA"], |
| 1003 | LookupRespMsg.HITS: [True], |
| 1004 | } |
| 1005 | ) |
| 1006 | session.poll() |
| 1007 | session.request_blocks( |
| 1008 | job_id=1, kv_request_id="req-1", keys=[b"hA"], block_ids=[7] |
| 1009 | ) |
| 1010 | sent_before = len(conn._sent) |
| 1011 | |
| 1012 | session.finish_request("req-1") |
| 1013 | |
| 1014 | fetches = [m for m in conn._sent[sent_before:] if m[TYPE_KEY] == FetchMsg.TYPE] |
| 1015 | assert fetches == [] |
| 1016 | |
| 1017 | def test_server_lookup_deferred_until_serve_then_all_misses(self): |
| 1018 | """``poll()`` only enqueues an inbound LookupMsg — no response is |
| 1019 | sent until ``serve_external_requests``. With an all-miss parent |
| 1020 | the aggregated LookupRespMsg carries the same keys and |
| 1021 | ``hits=[False, ...]``.""" |
| 1022 | session, conn, _ = _make_session() |
| 1023 | _activate(session, conn) |
| 1024 | |
| 1025 | sent_before = len(conn._sent) |
| 1026 | conn.enqueue( |
| 1027 | { |
| 1028 | TYPE_KEY: LookupMsg.TYPE, |
| 1029 | LookupMsg.ROUND_SEQ: 0, |
| 1030 | LookupMsg.KV_REQUEST_ID: "req-1", |
| 1031 | LookupMsg.KEYS: [b"hX", b"hY", b"hZ"], |
| 1032 | } |
| 1033 | ) |
| 1034 | session.poll() |
| 1035 | |
| 1036 | # Dispatch alone must not answer — the parent handle is only valid |
| 1037 | # during serve_external_requests. |
| 1038 | assert [ |
| 1039 | m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupRespMsg.TYPE |
| 1040 | ] == [] |
| 1041 | |
| 1042 | _serve(session, FakeParent()) |
| 1043 | |
| 1044 | resps = [ |
| 1045 | m for m in conn._sent[sent_before:] if m[TYPE_KEY] == LookupRespMsg.TYPE |
| 1046 | ] |
| 1047 | assert len(resps) == 1 |
| 1048 | resp = resps[0] |
| 1049 | assert resp[LookupRespMsg.KV_REQUEST_ID] == "req-1" |
| 1050 | assert resp[LookupRespMsg.KEYS] == [b"hX", b"hY", b"hZ"] |
| 1051 | assert resp[LookupRespMsg.HITS] == [False, False, False] |
| 1052 | |
| 1053 | |
| 1054 | # --------------------------------------------------------------------------- |
| 1055 | # Server-side handling of inbound LookupMsg (ParentManager-driven) |
| 1056 | # |
| 1057 | # poll() only enqueues the LookupMsg; serve_external_requests(parent) |
| 1058 | # resolves it. Tests follow the poll() → _serve() pattern. |
| 1059 | # --------------------------------------------------------------------------- |
| 1060 | |
| 1061 | |
| 1062 | def _send_lookup(conn: FakeConnection, kv_request_id: str, keys: list[bytes]): |
| 1063 | conn.enqueue( |
| 1064 | { |
| 1065 | TYPE_KEY: LookupMsg.TYPE, |
| 1066 | LookupMsg.ROUND_SEQ: 0, |
| 1067 | LookupMsg.KV_REQUEST_ID: kv_request_id, |
| 1068 | LookupMsg.KEYS: list(keys), |
| 1069 | } |
| 1070 | ) |
| 1071 | |
| 1072 | |
| 1073 | def _lookup_resps(conn: FakeConnection, since: int = 0) -> list[dict]: |
| 1074 | return [m for m in conn._sent[since:] if m[TYPE_KEY] == LookupRespMsg.TYPE] |
| 1075 | |
| 1076 | |
| 1077 | class TestServerLookupHandling: |
| 1078 | def test_immediate_hits_create_one_store_job(self): |
| 1079 | """All-HIT batch: one create_store_job call with all keys, one |
| 1080 | LookupRespMsg with hits=[True]*N, on_request_finished fires at the |
| 1081 | end of serve, and `available` is populated for the eventual fetch.""" |
| 1082 | cb = FakeParent(stored={b"hA": 1, b"hB": 2, b"hC": 3}) |
| 1083 | session, conn, _ = _make_session() |
| 1084 | _activate(session, conn) |
| 1085 | |
| 1086 | sent_before = len(conn._sent) |
| 1087 | _send_lookup(conn, "req-1", [b"hA", b"hB", b"hC"]) |
| 1088 | session.poll() |
| 1089 | _serve(session, cb) |
| 1090 | |
| 1091 | resps = _lookup_resps(conn, sent_before) |
| 1092 | assert len(resps) == 1 |
| 1093 | assert resps[0][LookupRespMsg.KEYS] == [b"hA", b"hB", b"hC"] |
| 1094 | assert resps[0][LookupRespMsg.HITS] == [True, True, True] |
| 1095 | |
| 1096 | kinds = [c[0] for c in cb.calls] |
| 1097 | assert kinds.count("create_store_job") == 1 |
| 1098 | cs = next(c for c in cb.calls if c[0] == "create_store_job") |
| 1099 | assert cs[1] == (b"hA", b"hB", b"hC") |
| 1100 | assert cb.calls[-1][0] == "on_request_finished" |
| 1101 | |
| 1102 | # Hits are pinned in outbound state for the upcoming FetchMsg match. |
| 1103 | assert set(_srv_outbound(session, "req-1").available) == { |
| 1104 | b"hA", |
| 1105 | b"hB", |
| 1106 | b"hC", |
| 1107 | } |
| 1108 | |
| 1109 | def test_all_misses_no_store_job_finish_fires(self): |
| 1110 | """All-MISS batch: no create_store_job call; one LookupRespMsg |
| 1111 | with hits=[False]*N; on_request_finished fires at end of serve.""" |
| 1112 | cb = FakeParent() |
| 1113 | session, conn, _ = _make_session() |
| 1114 | _activate(session, conn) |
| 1115 | |
| 1116 | sent_before = len(conn._sent) |
| 1117 | _send_lookup(conn, "req-1", [b"hA", b"hB"]) |
| 1118 | session.poll() |
| 1119 | _serve(session, cb) |
| 1120 | |
| 1121 | resps = _lookup_resps(conn, sent_before) |
| 1122 | assert len(resps) == 1 |
| 1123 | assert resps[0][LookupRespMsg.HITS] == [False, False] |
| 1124 | assert all(c[0] != "create_store_job" for c in cb.calls) |
| 1125 | assert cb.calls[-1][0] == "on_request_finished" |
| 1126 | |
| 1127 | def test_mixed_hit_miss_pending_defers_response_until_aggregate(self): |
| 1128 | """HIT/MISS resolutions do not go out on first sight when any |
| 1129 | key is still HIT_PENDING / RETRY. The lookup parks until every |
| 1130 | key has settled (or the deadline fires), then one |
| 1131 | LookupRespMsg carries all keys in wire order.""" |
| 1132 | cb = FakeParent( |
| 1133 | stored={b"hA": 1}, |
| 1134 | pending={b"hB"}, |
| 1135 | retry={b"hD"}, |
| 1136 | ) |
| 1137 | session, conn, _ = _make_session() |
| 1138 | _activate(session, conn) |
| 1139 | |
| 1140 | sent_before = len(conn._sent) |
| 1141 | _send_lookup(conn, "req-1", [b"hA", b"hB", b"hC", b"hD"]) |
| 1142 | session.poll() |
| 1143 | _serve(session, cb) |
| 1144 | |
| 1145 | # No LookupRespMsg yet — hB and hD are still pending. |
| 1146 | assert _lookup_resps(conn, sent_before) == [] |
| 1147 | # HIT is still pinned immediately so the eventual FetchMsg matches. |
| 1148 | cs_calls = [c for c in cb.calls if c[0] == "create_store_job"] |
| 1149 | assert len(cs_calls) == 1 |
| 1150 | assert cs_calls[0][1] == (b"hA",) |
| 1151 | # Lookup is parked; on_request_finished not yet called. |
| 1152 | assert all(c[0] != "on_request_finished" for c in cb.calls) |
| 1153 | assert len(_srv_lookups(session)) == 1 |
| 1154 | |
| 1155 | def test_pending_resolves_then_aggregate_response_fires(self): |
| 1156 | """A HIT_PENDING key that becomes HIT on a later poll releases |
| 1157 | the deferred aggregate response: one LookupRespMsg carrying |
| 1158 | both keys in wire order, and one create_store_job call per |
| 1159 | HIT (the second HIT is pinned when it resolves, not when the |
| 1160 | response goes out).""" |
| 1161 | cb = FakeParent(stored={b"hA": 1}, pending={b"hB"}) |
| 1162 | session, conn, _ = _make_session() |
| 1163 | _activate(session, conn) |
| 1164 | |
| 1165 | sent_before = len(conn._sent) |
| 1166 | _send_lookup(conn, "req-1", [b"hA", b"hB"]) |
| 1167 | session.poll() |
| 1168 | _serve(session, cb) |
| 1169 | # No response yet — hB still pending. |
| 1170 | assert _lookup_resps(conn, sent_before) == [] |
| 1171 | |
| 1172 | # Promote hB. |
| 1173 | cb.pending.discard(b"hB") |
| 1174 | cb.stored[b"hB"] = 2 |
| 1175 | |
| 1176 | # Drive resolver via a second serve_external_requests. |
| 1177 | _serve(session, cb) |
| 1178 | |
| 1179 | resps = _lookup_resps(conn, sent_before) |
| 1180 | assert len(resps) == 1 |
| 1181 | assert resps[0][LookupRespMsg.KEYS] == [b"hA", b"hB"] |
| 1182 | assert resps[0][LookupRespMsg.HITS] == [True, True] |
| 1183 | |
| 1184 | cs_calls = [c for c in cb.calls if c[0] == "create_store_job"] |
| 1185 | assert len(cs_calls) == 2 |
| 1186 | assert cs_calls[0][1] == (b"hA",) |
| 1187 | assert cs_calls[1][1] == (b"hB",) |
| 1188 | # on_request_finished fires once after the aggregate resolve. |
| 1189 | assert sum(1 for c in cb.calls if c[0] == "on_request_finished") == 1 |
| 1190 | assert b"hA" in _srv_outbound(session, "req-1").available |
| 1191 | assert b"hB" in _srv_outbound(session, "req-1").available |
| 1192 | |
| 1193 | def test_pending_timeout_replies_miss_no_store_job(self): |
| 1194 | """A HIT_PENDING key that stays pending past the batch |
| 1195 | ``deadline`` is force-MISS and never pinned; the deferred |
| 1196 | aggregate response fires with hits=[False].""" |
| 1197 | cb = FakeParent(pending={b"hA"}) |
| 1198 | session, conn, _ = _make_session() |
| 1199 | _activate(session, conn) |
| 1200 | |
| 1201 | sent_before = len(conn._sent) |
| 1202 | _send_lookup(conn, "req-1", [b"hA"]) |
| 1203 | session.poll() |
| 1204 | _serve(session, cb) |
| 1205 | # Initial serve: nothing immediate, lookup parked, no LookupRespMsg. |
| 1206 | assert _lookup_resps(conn, sent_before) == [] |
| 1207 | |
| 1208 | # Forge the deadline into the past to trigger the timeout branch. |
| 1209 | lookup = _srv_lookups(session)[0] |
| 1210 | lookup.deadline = time.monotonic() - 0.1 |
| 1211 | |
| 1212 | _serve(session, cb) |
| 1213 | |
| 1214 | resps = _lookup_resps(conn, sent_before) |
| 1215 | assert len(resps) == 1 |
| 1216 | assert resps[0][LookupRespMsg.KEYS] == [b"hA"] |
| 1217 | assert resps[0][LookupRespMsg.HITS] == [False] |
| 1218 | assert all(c[0] != "create_store_job" for c in cb.calls) |
| 1219 | assert sum(1 for c in cb.calls if c[0] == "on_request_finished") == 1 |
| 1220 | |
| 1221 | def test_finish_request_called_per_lookup_msg_not_per_kv_request_id(self): |
| 1222 | """Two LookupMsgs for the same kv_request_id get distinct ctxs |
| 1223 | and two on_request_finished calls (one per batch).""" |
| 1224 | cb = FakeParent(stored={b"hA": 1, b"hB": 2}) |
| 1225 | session, conn, _ = _make_session() |
| 1226 | _activate(session, conn) |
| 1227 | |
| 1228 | _send_lookup(conn, "req-1", [b"hA"]) |
| 1229 | session.poll() |
| 1230 | _serve(session, cb) |
| 1231 | _send_lookup(conn, "req-1", [b"hB"]) |
| 1232 | session.poll() |
| 1233 | _serve(session, cb) |
| 1234 | |
| 1235 | finish_calls = [c for c in cb.calls if c[0] == "on_request_finished"] |
| 1236 | assert len(finish_calls) == 2 |
| 1237 | # Distinct synthetic req_ids |
| 1238 | assert finish_calls[0][1] != finish_calls[1][1] |
| 1239 | # Both namespaced under the same kv_request_id |
| 1240 | assert ":req-1:" in finish_calls[0][1] |
| 1241 | assert ":req-1:" in finish_calls[1][1] |
| 1242 | |
| 1243 | def test_close_returns_open_batch_ctxs_as_failed_serves(self): |
| 1244 | """Tearing the session down with a parked batch returns the |
| 1245 | synthetic ctx as a failed serve (no parent handle at teardown) so |
| 1246 | the manager can release the TieringManager's state on its next |
| 1247 | serve.""" |
| 1248 | cb = FakeParent(pending={b"hA"}) |
| 1249 | session, conn, _ = _make_session() |
| 1250 | _activate(session, conn) |
| 1251 | |
| 1252 | _send_lookup(conn, "req-1", [b"hA"]) |
| 1253 | session.poll() |
| 1254 | _serve(session, cb) |
| 1255 | assert len(_srv_lookups(session)) == 1 |
| 1256 | assert all(c[0] != "on_request_finished" for c in cb.calls) |
| 1257 | |
| 1258 | result = session.close() |
| 1259 | |
| 1260 | assert len(result.failed_serves) == 1 |
| 1261 | assert ":req-1:" in result.failed_serves[0].req_id |
| 1262 | # close() itself must not call the parent. |
| 1263 | assert all(c[0] != "on_request_finished" for c in cb.calls) |
| 1264 | |
| 1265 | def test_wire_finish_drops_pending_batches_for_kv_request_id(self): |
| 1266 | """``ServerRole.finish(kv_request_id)`` drops every parked batch |
| 1267 | whose kv_request_id matches and queues its ctx for the next |
| 1268 | serve's on_request_finished.""" |
| 1269 | cb = FakeParent(pending={b"hA", b"hB"}) |
| 1270 | session, conn, _ = _make_session() |
| 1271 | _activate(session, conn) |
| 1272 | |
| 1273 | _send_lookup(conn, "req-1", [b"hA"]) |
| 1274 | session.poll() |
| 1275 | _serve(session, cb) |
| 1276 | _send_lookup(conn, "req-2", [b"hB"]) |
| 1277 | session.poll() |
| 1278 | _serve(session, cb) |
| 1279 | assert len(_srv_lookups(session)) == 2 |
| 1280 | |
| 1281 | session._server.finish("req-1") |
| 1282 | |
| 1283 | # req-1 batch dropped from parked lookups; its ctx queued for release. |
| 1284 | remaining_kv_request_ids = {b.kv_request_id for b in _srv_lookups(session)} |
| 1285 | assert remaining_kv_request_ids == {"req-2"} |
| 1286 | queued = session._server._finished_lookup_ctxs |
| 1287 | assert len(queued) == 1 |
| 1288 | assert ":req-1:" in queued[0].req_id |
| 1289 | |
| 1290 | # The next serve fires on_request_finished exactly once for req-1. |
| 1291 | _serve(session, cb) |
| 1292 | finish_calls = [c for c in cb.calls if c[0] == "on_request_finished"] |
| 1293 | assert len(finish_calls) == 1 |
| 1294 | assert ":req-1:" in finish_calls[0][1] |
| 1295 | |
| 1296 | def test_incoming_fetch_drops_pending_lookups_for_kv_request_id(self): |
| 1297 | """A peer FetchMsg terminates the lookup phase for its id: parked |
| 1298 | lookups with matching kv_request_id are dropped and their |
| 1299 | ctx queued for on_request_finished; other kv_request_ids untouched.""" |
| 1300 | cb = FakeParent(pending={b"hA", b"hB"}) |
| 1301 | session, conn, _ = _make_session() |
| 1302 | _activate(session, conn) |
| 1303 | |
| 1304 | _send_lookup(conn, "req-1", [b"hA"]) |
| 1305 | session.poll() |
| 1306 | _serve(session, cb) |
| 1307 | _send_lookup(conn, "req-2", [b"hB"]) |
| 1308 | session.poll() |
| 1309 | _serve(session, cb) |
| 1310 | assert len(_srv_lookups(session)) == 2 |
| 1311 | |
| 1312 | # Empty FetchMsg: peer signals "lookup phase done" without asking |
| 1313 | # for any blocks (the all-miss case). |
| 1314 | conn.enqueue( |
| 1315 | { |
| 1316 | TYPE_KEY: FetchMsg.TYPE, |
| 1317 | FetchMsg.ROUND_SEQ: 0, |
| 1318 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1319 | FetchMsg.KEYS: [], |
| 1320 | FetchMsg.BLOCK_INDEXES: [], |
| 1321 | } |
| 1322 | ) |
| 1323 | session.poll() |
| 1324 | |
| 1325 | remaining_kv_request_ids = {lu.kv_request_id for lu in _srv_lookups(session)} |
| 1326 | assert remaining_kv_request_ids == {"req-2"} |
| 1327 | # Dispatch queues the ctx but does not call the parent yet. |
| 1328 | queued = session._server._finished_lookup_ctxs |
| 1329 | assert len(queued) == 1 |
| 1330 | assert ":req-1:" in queued[0].req_id |
| 1331 | |
| 1332 | # The next serve fires on_request_finished exactly once for req-1. |
| 1333 | _serve(session, cb) |
| 1334 | finish_calls = [c for c in cb.calls if c[0] == "on_request_finished"] |
| 1335 | assert len(finish_calls) == 1 |
| 1336 | assert ":req-1:" in finish_calls[0][1] |
| 1337 | |
| 1338 | def test_lookup_then_fetch_round_trip_emits_store_result(self): |
| 1339 | """End-to-end: lookup pins primary slots → fetch matches them → |
| 1340 | NIXL transfer completes → StoreResult surfaces with the |
| 1341 | create_store_job's job_id (the engine releases the pin).""" |
| 1342 | cb = FakeParent(stored={b"hA": 7, b"hB": 8}) |
| 1343 | session, conn, transport = _make_session() |
| 1344 | _activate(session, conn) |
| 1345 | |
| 1346 | _send_lookup(conn, "req-1", [b"hA", b"hB"]) |
| 1347 | session.poll() |
| 1348 | _serve(session, cb) |
| 1349 | cs = next(c for c in cb.calls if c[0] == "create_store_job") |
| 1350 | # FakeParent issues monotonic job_ids starting at 1000. |
| 1351 | expected_job_id = 1000 |
| 1352 | |
| 1353 | # Consumer issues FetchMsg on the resolved hits. |
| 1354 | conn.enqueue( |
| 1355 | { |
| 1356 | TYPE_KEY: FetchMsg.TYPE, |
| 1357 | FetchMsg.ROUND_SEQ: 0, |
| 1358 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1359 | FetchMsg.KEYS: [b"hA", b"hB"], |
| 1360 | FetchMsg.BLOCK_INDEXES: [20, 21], |
| 1361 | } |
| 1362 | ) |
| 1363 | session.poll() |
| 1364 | |
| 1365 | # NIXL write_blocks called with our pinned local block_ids. |
| 1366 | assert len(transport._transfers) == 1 |
| 1367 | _, (_peer, local, remote) = next(iter(transport._transfers.items())) |
| 1368 | assert local == [7, 8] |
| 1369 | assert remote == [20, 21] |
| 1370 | |
| 1371 | # Drive the transport completion. |
| 1372 | transport._poll_done.append(0) |
| 1373 | result = session.poll() |
| 1374 | |
| 1375 | store_results = [s for s in result.stores if s.success] |
| 1376 | assert any(s.job_id == expected_job_id for s in store_results) |
| 1377 | # Sanity: kv mention in synthetic ctx. |
| 1378 | assert cs[2].startswith("p2p:") |
| 1379 | |
| 1380 | |
| 1381 | # --------------------------------------------------------------------------- |
| 1382 | # Server-role flows |
| 1383 | # --------------------------------------------------------------------------- |
| 1384 | |
| 1385 | |
| 1386 | class TestServerFlows: |
| 1387 | def test_store_then_fetch_matches(self): |
| 1388 | """Blocks stored before fetch demand are matched on demand arrival.""" |
| 1389 | session, conn, transport = _make_session() |
| 1390 | _activate(session, conn) |
| 1391 | session.add_stored_blocks("req-1", [b"k1", b"k2"], [0, 1], job_id=1) |
| 1392 | conn.enqueue( |
| 1393 | { |
| 1394 | TYPE_KEY: FetchMsg.TYPE, |
| 1395 | FetchMsg.ROUND_SEQ: 0, |
| 1396 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1397 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 1398 | FetchMsg.BLOCK_INDEXES: [10, 11], |
| 1399 | } |
| 1400 | ) |
| 1401 | session.poll() |
| 1402 | assert len(transport._transfers) == 1 |
| 1403 | _, (peer, local, remote) = next(iter(transport._transfers.items())) |
| 1404 | assert local == [0, 1] |
| 1405 | assert remote == [10, 11] |
| 1406 | |
| 1407 | def test_fetch_then_store_matches(self): |
| 1408 | """Fetch demand registered before store; store fulfills it.""" |
| 1409 | session, conn, transport = _make_session() |
| 1410 | _activate(session, conn) |
| 1411 | conn.enqueue( |
| 1412 | { |
| 1413 | TYPE_KEY: FetchMsg.TYPE, |
| 1414 | FetchMsg.ROUND_SEQ: 0, |
| 1415 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1416 | FetchMsg.KEYS: [b"k1"], |
| 1417 | FetchMsg.BLOCK_INDEXES: [5], |
| 1418 | } |
| 1419 | ) |
| 1420 | session.poll() |
| 1421 | assert len(transport._transfers) == 0 |
| 1422 | session.add_stored_blocks("req-1", [b"k1"], [3], job_id=1) |
| 1423 | assert len(transport._transfers) == 1 |
| 1424 | |
| 1425 | def test_transfer_completion_emits_store_result_and_done(self): |
| 1426 | """Completed transfer reports StoreResult and sends TransferDoneMsg.""" |
| 1427 | session, conn, transport = _make_session() |
| 1428 | _activate(session, conn) |
| 1429 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1430 | conn.enqueue( |
| 1431 | { |
| 1432 | TYPE_KEY: FetchMsg.TYPE, |
| 1433 | FetchMsg.ROUND_SEQ: 0, |
| 1434 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1435 | FetchMsg.KEYS: [b"k1"], |
| 1436 | FetchMsg.BLOCK_INDEXES: [5], |
| 1437 | } |
| 1438 | ) |
| 1439 | session.poll() |
| 1440 | tid = next(iter(transport._transfers)) |
| 1441 | transport._poll_done.append(tid) |
| 1442 | stores = session.poll().stores |
| 1443 | assert StoreResult(job_id=1, success=True) in stores |
| 1444 | assert any(m[TYPE_KEY] == TransferDoneMsg.TYPE for m in conn._sent) |
| 1445 | |
| 1446 | def test_abort_fetch_replies_with_ack(self): |
| 1447 | session, conn, _ = _make_session() |
| 1448 | _activate(session, conn) |
| 1449 | conn.enqueue( |
| 1450 | { |
| 1451 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1452 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1453 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1454 | } |
| 1455 | ) |
| 1456 | session.poll() |
| 1457 | ack = next(m for m in conn._sent if m[TYPE_KEY] == AbortAckMsg.TYPE) |
| 1458 | assert ack[AbortAckMsg.KV_REQUEST_ID] == "req-1" |
| 1459 | assert _srv_abort_started(session, "req-1") is None |
| 1460 | |
| 1461 | def test_abort_fetch_defers_ack_when_cancel_pending(self): |
| 1462 | """If cancel(mode='wait') reports still-inflight tids, the ack is |
| 1463 | deferred and the abort is parked (abort_started_at set).""" |
| 1464 | session, conn, transport = _make_session() |
| 1465 | _activate(session, conn) |
| 1466 | # Seed an inflight transfer for req-1 that the transport pretends |
| 1467 | # cannot be canceled yet. |
| 1468 | tid = 42 |
| 1469 | session._server._inflight_add( |
| 1470 | tid, |
| 1471 | _InflightXfer( |
| 1472 | kv_request_id="req-1", |
| 1473 | block_count=1, |
| 1474 | job_ids={1}, |
| 1475 | round=_OutboundRequestState(inflight=1), |
| 1476 | ), |
| 1477 | ) |
| 1478 | transport._cancel_still_inflight.add(tid) |
| 1479 | |
| 1480 | conn.enqueue( |
| 1481 | { |
| 1482 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1483 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1484 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1485 | } |
| 1486 | ) |
| 1487 | session.poll() |
| 1488 | |
| 1489 | assert not any(m[TYPE_KEY] == AbortAckMsg.TYPE for m in conn._sent) |
| 1490 | assert _srv_abort_started(session, "req-1") is not None |
| 1491 | # First attempt happens inside _on_abort_fetch; the per-tick |
| 1492 | # drain runs again at the end of poll() — both are wait-mode. |
| 1493 | assert all(mode == "wait" for _, mode in transport._cancel_calls) |
| 1494 | assert tid in session._server._inflight # still tracked |
| 1495 | |
| 1496 | def test_abort_fetch_acks_after_drain(self): |
| 1497 | """Once the transport reports the tid as DONE the parked abort |
| 1498 | completes and AbortAckMsg is sent.""" |
| 1499 | session, conn, transport = _make_session() |
| 1500 | _activate(session, conn) |
| 1501 | tid = 42 |
| 1502 | session._server._inflight_add( |
| 1503 | tid, |
| 1504 | _InflightXfer( |
| 1505 | kv_request_id="req-1", |
| 1506 | block_count=1, |
| 1507 | job_ids={1}, |
| 1508 | round=_OutboundRequestState(inflight=1), |
| 1509 | ), |
| 1510 | ) |
| 1511 | transport._cancel_still_inflight.add(tid) |
| 1512 | |
| 1513 | conn.enqueue( |
| 1514 | { |
| 1515 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1516 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1517 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1518 | } |
| 1519 | ) |
| 1520 | session.poll() |
| 1521 | assert _srv_abort_started(session, "req-1") is not None |
| 1522 | |
| 1523 | # Backend finishes draining: transport.poll() will return tid as |
| 1524 | # DONE, and the next cancel(mode='wait') call sees it's gone. |
| 1525 | transport._cancel_still_inflight.discard(tid) |
| 1526 | transport._poll_done.append(tid) |
| 1527 | |
| 1528 | session.poll() |
| 1529 | |
| 1530 | ack = next(m for m in conn._sent if m[TYPE_KEY] == AbortAckMsg.TYPE) |
| 1531 | assert ack[AbortAckMsg.KV_REQUEST_ID] == "req-1" |
| 1532 | assert _srv_abort_started(session, "req-1") is None |
| 1533 | assert tid not in session._server._inflight |
| 1534 | |
| 1535 | def test_abort_fetch_force_cancels_after_timeout(self): |
| 1536 | """If wait-mode never drains, the deadline forces immediate |
| 1537 | cancel and an ack is still sent.""" |
| 1538 | session, conn, transport = _make_session() |
| 1539 | _activate(session, conn) |
| 1540 | tid = 42 |
| 1541 | session._server._inflight_add( |
| 1542 | tid, |
| 1543 | _InflightXfer( |
| 1544 | kv_request_id="req-1", |
| 1545 | block_count=1, |
| 1546 | job_ids={1}, |
| 1547 | round=_OutboundRequestState(inflight=1), |
| 1548 | ), |
| 1549 | ) |
| 1550 | transport._cancel_still_inflight.add(tid) |
| 1551 | |
| 1552 | conn.enqueue( |
| 1553 | { |
| 1554 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1555 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1556 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1557 | } |
| 1558 | ) |
| 1559 | session.poll() |
| 1560 | assert _srv_abort_started(session, "req-1") is not None |
| 1561 | # Backdate past the drain deadline. |
| 1562 | session._server._pending_aborts[("req-1", 0)] = ( |
| 1563 | time.monotonic() - _CANCEL_DRAIN_TIMEOUT_S - 1.0 |
| 1564 | ) |
| 1565 | # Even if the transport still claims it can't cancel, the |
| 1566 | # session must force-pop and ack. |
| 1567 | transport._cancel_calls.clear() |
| 1568 | |
| 1569 | session.poll() |
| 1570 | |
| 1571 | ack = next(m for m in conn._sent if m[TYPE_KEY] == AbortAckMsg.TYPE) |
| 1572 | assert ack[AbortAckMsg.KV_REQUEST_ID] == "req-1" |
| 1573 | assert _srv_abort_started(session, "req-1") is None |
| 1574 | assert tid not in session._server._inflight |
| 1575 | assert ([tid], "immediate") in transport._cancel_calls |
| 1576 | |
| 1577 | def test_abort_fetch_idempotent_while_draining(self): |
| 1578 | """Receiving AbortFetchMsg twice for the same kv_request_id |
| 1579 | keeps a single pending entry and produces a single ack.""" |
| 1580 | session, conn, transport = _make_session() |
| 1581 | _activate(session, conn) |
| 1582 | tid = 42 |
| 1583 | session._server._inflight_add( |
| 1584 | tid, |
| 1585 | _InflightXfer( |
| 1586 | kv_request_id="req-1", |
| 1587 | block_count=1, |
| 1588 | job_ids={1}, |
| 1589 | round=_OutboundRequestState(inflight=1), |
| 1590 | ), |
| 1591 | ) |
| 1592 | transport._cancel_still_inflight.add(tid) |
| 1593 | |
| 1594 | conn.enqueue( |
| 1595 | { |
| 1596 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1597 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1598 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1599 | } |
| 1600 | ) |
| 1601 | session.poll() |
| 1602 | first_started_at = _srv_abort_started(session, "req-1") |
| 1603 | |
| 1604 | # Second AbortFetchMsg for the same kv_request_id while still |
| 1605 | # draining must not reset the deadline. |
| 1606 | conn.enqueue( |
| 1607 | { |
| 1608 | TYPE_KEY: AbortFetchMsg.TYPE, |
| 1609 | AbortFetchMsg.ROUND_SEQ: 0, |
| 1610 | AbortFetchMsg.KV_REQUEST_ID: "req-1", |
| 1611 | } |
| 1612 | ) |
| 1613 | session.poll() |
| 1614 | assert _srv_abort_started(session, "req-1") == first_started_at |
| 1615 | |
| 1616 | # Now let the drain succeed and confirm exactly one ack ever. |
| 1617 | transport._cancel_still_inflight.discard(tid) |
| 1618 | transport._poll_done.append(tid) |
| 1619 | session.poll() |
| 1620 | |
| 1621 | acks = [m for m in conn._sent if m[TYPE_KEY] == AbortAckMsg.TYPE] |
| 1622 | assert len(acks) == 1 |
| 1623 | assert acks[0][AbortAckMsg.KV_REQUEST_ID] == "req-1" |
| 1624 | |
| 1625 | def test_store_timeout(self): |
| 1626 | session, conn, _ = _make_session() |
| 1627 | _activate(session, conn) |
| 1628 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1629 | # Backdate. |
| 1630 | session._server._store_jobs[1] = time.monotonic() - 60.0 |
| 1631 | stores = session.poll().stores |
| 1632 | assert StoreResult(job_id=1, success=False) in stores |
| 1633 | |
| 1634 | def test_store_timeout_then_late_completion_no_duplicate(self): |
| 1635 | """A job timed out by _timeout_pending_store_jobs must not also |
| 1636 | emit a contradictory StoreResult(success=True) when the transport |
| 1637 | later reports the same transfer as done.""" |
| 1638 | session, conn, transport = _make_session() |
| 1639 | _activate(session, conn) |
| 1640 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1641 | conn.enqueue( |
| 1642 | { |
| 1643 | TYPE_KEY: FetchMsg.TYPE, |
| 1644 | FetchMsg.ROUND_SEQ: 0, |
| 1645 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1646 | FetchMsg.KEYS: [b"k1"], |
| 1647 | FetchMsg.BLOCK_INDEXES: [5], |
| 1648 | } |
| 1649 | ) |
| 1650 | session.poll() |
| 1651 | tid = next(iter(transport._transfers)) |
| 1652 | |
| 1653 | # Backdate the store job so the next poll times it out. |
| 1654 | session._server._store_jobs[1] = time.monotonic() - 60.0 |
| 1655 | stores = session.poll().stores |
| 1656 | assert StoreResult(job_id=1, success=False) in stores |
| 1657 | assert StoreResult(job_id=1, success=True) not in stores |
| 1658 | |
| 1659 | # Transport later reports the same transfer as done — must not |
| 1660 | # emit a second (contradictory) StoreResult for job_id=1. |
| 1661 | transport._poll_done.append(tid) |
| 1662 | stores = session.poll().stores |
| 1663 | assert all(s.job_id != 1 for s in stores), ( |
| 1664 | f"unexpected duplicate StoreResult after timeout: {stores}" |
| 1665 | ) |
| 1666 | |
| 1667 | def test_store_timeout_then_late_failure_no_duplicate(self): |
| 1668 | """Symmetric guard: a timed-out job must not also emit a second |
| 1669 | StoreResult(success=False) when the transport later reports the |
| 1670 | same transfer as failed.""" |
| 1671 | session, conn, transport = _make_session() |
| 1672 | _activate(session, conn) |
| 1673 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1674 | conn.enqueue( |
| 1675 | { |
| 1676 | TYPE_KEY: FetchMsg.TYPE, |
| 1677 | FetchMsg.ROUND_SEQ: 0, |
| 1678 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1679 | FetchMsg.KEYS: [b"k1"], |
| 1680 | FetchMsg.BLOCK_INDEXES: [5], |
| 1681 | } |
| 1682 | ) |
| 1683 | session.poll() |
| 1684 | tid = next(iter(transport._transfers)) |
| 1685 | |
| 1686 | session._server._store_jobs[1] = time.monotonic() - 60.0 |
| 1687 | stores = session.poll().stores |
| 1688 | assert [s for s in stores if s.job_id == 1] == [ |
| 1689 | StoreResult(job_id=1, success=False) |
| 1690 | ] |
| 1691 | |
| 1692 | transport._poll_failed.append(tid) |
| 1693 | stores = session.poll().stores |
| 1694 | assert all(s.job_id != 1 for s in stores), ( |
| 1695 | f"unexpected duplicate StoreResult after timeout: {stores}" |
| 1696 | ) |
| 1697 | |
| 1698 | |
| 1699 | # --------------------------------------------------------------------------- |
| 1700 | # finish_request server-role early-fail flow |
| 1701 | # --------------------------------------------------------------------------- |
| 1702 | |
| 1703 | |
| 1704 | class TestFinishRequestServerSide: |
| 1705 | def _last_transfer_done(self, conn: FakeConnection) -> dict | None: |
| 1706 | for msg in reversed(conn._sent): |
| 1707 | if msg[TYPE_KEY] == TransferDoneMsg.TYPE: |
| 1708 | return msg |
| 1709 | return None |
| 1710 | |
| 1711 | def test_no_inflight_unmatched_demand_sends_failure(self): |
| 1712 | """finish_request with unmatched demand and no inflight -> |
| 1713 | immediate TransferDoneMsg(success=False); _outbound cleared.""" |
| 1714 | session, conn, _ = _make_session() |
| 1715 | _activate(session, conn) |
| 1716 | # Decoder demanded a block we never stored. |
| 1717 | conn.enqueue( |
| 1718 | { |
| 1719 | TYPE_KEY: FetchMsg.TYPE, |
| 1720 | FetchMsg.ROUND_SEQ: 0, |
| 1721 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1722 | FetchMsg.KEYS: [b"k1"], |
| 1723 | FetchMsg.BLOCK_INDEXES: [5], |
| 1724 | } |
| 1725 | ) |
| 1726 | session.poll() |
| 1727 | assert _srv_outbound(session, "req-1") is not None |
| 1728 | |
| 1729 | session.finish_request("req-1") |
| 1730 | |
| 1731 | msg = self._last_transfer_done(conn) |
| 1732 | assert msg is not None |
| 1733 | assert msg[TransferDoneMsg.KV_REQUEST_ID] == "req-1" |
| 1734 | assert msg[TransferDoneMsg.SUCCESS] is False |
| 1735 | assert _srv_outbound(session, "req-1") is None |
| 1736 | |
| 1737 | def test_with_inflight_defers_then_fires_on_last_transfer(self): |
| 1738 | """finish_request with inflight defers; last transfer fires the |
| 1739 | early-fail message and clears _outbound.""" |
| 1740 | session, conn, transport = _make_session() |
| 1741 | _activate(session, conn) |
| 1742 | # Demand 2 blocks; we store 1 (kicks one inflight transfer). |
| 1743 | conn.enqueue( |
| 1744 | { |
| 1745 | TYPE_KEY: FetchMsg.TYPE, |
| 1746 | FetchMsg.ROUND_SEQ: 0, |
| 1747 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1748 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 1749 | FetchMsg.BLOCK_INDEXES: [10, 11], |
| 1750 | } |
| 1751 | ) |
| 1752 | session.poll() |
| 1753 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1754 | assert len(transport._transfers) == 1 |
| 1755 | |
| 1756 | # finish_request while inflight: no early-fail yet. |
| 1757 | before = len(conn._sent) |
| 1758 | session.finish_request("req-1") |
| 1759 | assert len(conn._sent) == before |
| 1760 | assert _srv_outbound(session, "req-1") is not None |
| 1761 | assert _srv_outbound(session, "req-1").finishing |
| 1762 | |
| 1763 | # Last inflight settles -> early-fail fires. |
| 1764 | tid = next(iter(transport._transfers)) |
| 1765 | transport._poll_done.append(tid) |
| 1766 | session.poll() |
| 1767 | |
| 1768 | msg = self._last_transfer_done(conn) |
| 1769 | assert msg is not None |
| 1770 | assert msg[TransferDoneMsg.KV_REQUEST_ID] == "req-1" |
| 1771 | assert msg[TransferDoneMsg.SUCCESS] is False |
| 1772 | assert _srv_outbound(session, "req-1") is None |
| 1773 | |
| 1774 | def test_full_demand_satisfied_still_sends_success(self): |
| 1775 | """finish_request must not override a fully-satisfied transfer: |
| 1776 | when remaining hits 0, success=True still fires.""" |
| 1777 | session, conn, transport = _make_session() |
| 1778 | _activate(session, conn) |
| 1779 | conn.enqueue( |
| 1780 | { |
| 1781 | TYPE_KEY: FetchMsg.TYPE, |
| 1782 | FetchMsg.ROUND_SEQ: 0, |
| 1783 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1784 | FetchMsg.KEYS: [b"k1"], |
| 1785 | FetchMsg.BLOCK_INDEXES: [10], |
| 1786 | } |
| 1787 | ) |
| 1788 | session.poll() |
| 1789 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1790 | # Mark finishing (e.g., on_request_finished racing with the last |
| 1791 | # store) — but all demand is satisfied. |
| 1792 | session.finish_request("req-1") |
| 1793 | |
| 1794 | tid = next(iter(transport._transfers)) |
| 1795 | transport._poll_done.append(tid) |
| 1796 | session.poll() |
| 1797 | |
| 1798 | msg = next(m for m in conn._sent if m[TYPE_KEY] == TransferDoneMsg.TYPE) |
| 1799 | assert msg[TransferDoneMsg.SUCCESS] is True |
| 1800 | |
| 1801 | def test_prefiller_first_finish_before_fetch(self): |
| 1802 | """Prefiller-first: finish_request runs before the decoder's |
| 1803 | fetch arrives. State is held until fetch, then |
| 1804 | finalized — success=True if all demand was matched against |
| 1805 | available blocks, else success=False.""" |
| 1806 | # Case A: all demand satisfied by available blocks. |
| 1807 | session, conn, transport = _make_session() |
| 1808 | _activate(session, conn) |
| 1809 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 1810 | # finish_request first — no demand received yet -> defer. |
| 1811 | session.finish_request("req-1") |
| 1812 | assert _srv_outbound(session, "req-1") is not None |
| 1813 | # Fetch arrives now: demand fully satisfied by available. |
| 1814 | conn.enqueue( |
| 1815 | { |
| 1816 | TYPE_KEY: FetchMsg.TYPE, |
| 1817 | FetchMsg.ROUND_SEQ: 0, |
| 1818 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1819 | FetchMsg.KEYS: [b"k1"], |
| 1820 | FetchMsg.BLOCK_INDEXES: [10], |
| 1821 | } |
| 1822 | ) |
| 1823 | session.poll() |
| 1824 | # The transfer was inflight; finalize it. |
| 1825 | tid = next(iter(transport._transfers)) |
| 1826 | transport._poll_done.append(tid) |
| 1827 | session.poll() |
| 1828 | msg = next(m for m in conn._sent if m[TYPE_KEY] == TransferDoneMsg.TYPE) |
| 1829 | assert msg[TransferDoneMsg.SUCCESS] is True |
| 1830 | |
| 1831 | # Case B: demand exceeds available -> early-fail fires from fetch. |
| 1832 | session, conn, transport = _make_session() |
| 1833 | _activate(session, conn) |
| 1834 | session.add_stored_blocks("req-2", [b"k1"], [0], job_id=2) |
| 1835 | session.finish_request("req-2") |
| 1836 | conn.enqueue( |
| 1837 | { |
| 1838 | TYPE_KEY: FetchMsg.TYPE, |
| 1839 | FetchMsg.ROUND_SEQ: 0, |
| 1840 | FetchMsg.KV_REQUEST_ID: "req-2", |
| 1841 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 1842 | FetchMsg.BLOCK_INDEXES: [10, 11], |
| 1843 | } |
| 1844 | ) |
| 1845 | session.poll() |
| 1846 | # Inflight for k1 still in flight; nothing yet for the early-fail |
| 1847 | # — the same code path will fire from _collect_store_results. |
| 1848 | tid = next(iter(transport._transfers)) |
| 1849 | transport._poll_done.append(tid) |
| 1850 | session.poll() |
| 1851 | msg = next(m for m in conn._sent if m[TYPE_KEY] == TransferDoneMsg.TYPE) |
| 1852 | assert msg[TransferDoneMsg.SUCCESS] is False |
| 1853 | assert _srv_outbound(session, "req-2") is None |
| 1854 | |
| 1855 | def test_unknown_request_is_noop(self): |
| 1856 | session, conn, _ = _make_session() |
| 1857 | _activate(session, conn) |
| 1858 | before = len(conn._sent) |
| 1859 | session.finish_request("never-existed") |
| 1860 | assert len(conn._sent) == before |
| 1861 | |
| 1862 | def test_finish_request_no_inflight_emits_store_failure(self): |
| 1863 | """finish_request with a stored-but-unmatched job and no inflight -> |
| 1864 | TransferDoneMsg(success=False) AND deferred |
| 1865 | StoreResult(success=False) for the submit_store'd job, instead of |
| 1866 | the 30s _STORE_TIMEOUT_S path.""" |
| 1867 | session, conn, _ = _make_session() |
| 1868 | _activate(session, conn) |
| 1869 | # Decoder demanded b"demand"; we never stored it. |
| 1870 | conn.enqueue( |
| 1871 | { |
| 1872 | TYPE_KEY: FetchMsg.TYPE, |
| 1873 | FetchMsg.ROUND_SEQ: 0, |
| 1874 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1875 | FetchMsg.KEYS: [b"demand"], |
| 1876 | FetchMsg.BLOCK_INDEXES: [5], |
| 1877 | } |
| 1878 | ) |
| 1879 | session.poll() |
| 1880 | # We did submit_store a different block — goes to available, never |
| 1881 | # matches demand. Without the shortcut, job 42 sits in _store_jobs |
| 1882 | # for _STORE_TIMEOUT_S. |
| 1883 | session.add_stored_blocks("req-1", [b"unrelated"], [0], job_id=42) |
| 1884 | assert _srv_outbound(session, "req-1").pending_job_ids == {42} |
| 1885 | |
| 1886 | session.finish_request("req-1") |
| 1887 | |
| 1888 | # Peer notified immediately with success=False (remaining > 0). |
| 1889 | msg = self._last_transfer_done(conn) |
| 1890 | assert msg is not None |
| 1891 | assert msg[TransferDoneMsg.SUCCESS] is False |
| 1892 | assert _srv_outbound(session, "req-1") is None |
| 1893 | |
| 1894 | # Local store job surfaces on the next poll, success=False. |
| 1895 | stores = session.poll().stores |
| 1896 | assert StoreResult(job_id=42, success=False) in stores |
| 1897 | assert 42 not in session._server._store_jobs |
| 1898 | |
| 1899 | def test_finish_request_remaining_zero_emits_success_via_inflight(self): |
| 1900 | """Deferred-via-inflight path: finish_request with inflight, last |
| 1901 | transfer drains remaining to 0 -> TransferDoneMsg(success=True) |
| 1902 | AND StoreResult(success=True).""" |
| 1903 | session, conn, transport = _make_session() |
| 1904 | _activate(session, conn) |
| 1905 | conn.enqueue( |
| 1906 | { |
| 1907 | TYPE_KEY: FetchMsg.TYPE, |
| 1908 | FetchMsg.ROUND_SEQ: 0, |
| 1909 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1910 | FetchMsg.KEYS: [b"k1"], |
| 1911 | FetchMsg.BLOCK_INDEXES: [10], |
| 1912 | } |
| 1913 | ) |
| 1914 | session.poll() |
| 1915 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=7) |
| 1916 | # finish_request races with the inflight transfer. |
| 1917 | session.finish_request("req-1") |
| 1918 | assert _srv_outbound(session, "req-1") is not None # deferred |
| 1919 | |
| 1920 | # Last inflight completes -> _finalize_outbound(success=True) fires. |
| 1921 | tid = next(iter(transport._transfers)) |
| 1922 | transport._poll_done.append(tid) |
| 1923 | stores = session.poll().stores |
| 1924 | |
| 1925 | msg = self._last_transfer_done(conn) |
| 1926 | assert msg is not None |
| 1927 | assert msg[TransferDoneMsg.SUCCESS] is True |
| 1928 | assert StoreResult(job_id=7, success=True) in stores |
| 1929 | assert _srv_outbound(session, "req-1") is None |
| 1930 | |
| 1931 | def test_write_blocks_failure_finalizes_with_failure(self): |
| 1932 | """write_blocks returning None must not leave the request hanging. |
| 1933 | |
| 1934 | The matched blocks are gone from req.demanded but no inflight |
| 1935 | will satisfy them, so remaining > 0 forever. Setting finishing |
| 1936 | and calling _finalize_outbound(success=False) immediately (no |
| 1937 | other inflight) tells the peer + emits StoreResult(success=False) |
| 1938 | without waiting on _STORE_TIMEOUT_S or _LOAD_TIMEOUT_S. |
| 1939 | """ |
| 1940 | session, conn, transport = _make_session() |
| 1941 | _activate(session, conn) |
| 1942 | # Decoder demands b"k1". |
| 1943 | conn.enqueue( |
| 1944 | { |
| 1945 | TYPE_KEY: FetchMsg.TYPE, |
| 1946 | FetchMsg.ROUND_SEQ: 0, |
| 1947 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1948 | FetchMsg.KEYS: [b"k1"], |
| 1949 | FetchMsg.BLOCK_INDEXES: [10], |
| 1950 | } |
| 1951 | ) |
| 1952 | session.poll() |
| 1953 | # Force write_blocks to fail on the next call. |
| 1954 | transport.write_blocks = lambda *a, **kw: None # type: ignore[method-assign] |
| 1955 | |
| 1956 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=42) |
| 1957 | |
| 1958 | # Outbound was finalized immediately (no other inflight). |
| 1959 | assert _srv_outbound(session, "req-1") is None |
| 1960 | # Peer notified with success=False. |
| 1961 | msg = next(m for m in conn._sent if m[TYPE_KEY] == TransferDoneMsg.TYPE) |
| 1962 | assert msg[TransferDoneMsg.KV_REQUEST_ID] == "req-1" |
| 1963 | assert msg[TransferDoneMsg.SUCCESS] is False |
| 1964 | # Local store job surfaces on the next poll. |
| 1965 | stores = session.poll().stores |
| 1966 | assert StoreResult(job_id=42, success=False) in stores |
| 1967 | assert 42 not in session._server._store_jobs |
| 1968 | |
| 1969 | def test_partial_match_completes_in_two_rounds(self): |
| 1970 | """Peer demand for [k1, k2, k3]; first round only k1 is available, |
| 1971 | second round adds k2 and k3. Each round transfers what's matched |
| 1972 | and the request finalizes with success once remaining hits zero. |
| 1973 | """ |
| 1974 | session, conn, transport = _make_session() |
| 1975 | _activate(session, conn) |
| 1976 | |
| 1977 | # Peer fetches three blocks. |
| 1978 | conn.enqueue( |
| 1979 | { |
| 1980 | TYPE_KEY: FetchMsg.TYPE, |
| 1981 | FetchMsg.ROUND_SEQ: 0, |
| 1982 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 1983 | FetchMsg.KEYS: [b"k1", b"k2", b"k3"], |
| 1984 | FetchMsg.BLOCK_INDEXES: [10, 11, 12], |
| 1985 | } |
| 1986 | ) |
| 1987 | session.poll() |
| 1988 | # Demand registered, no matches yet. |
| 1989 | assert session._server._inflight == {} |
| 1990 | outbound = _srv_outbound(session, "req-1") |
| 1991 | assert outbound.remaining == 3 |
| 1992 | assert set(outbound.demanded.keys()) == {b"k1", b"k2", b"k3"} |
| 1993 | |
| 1994 | # Round 1: only k1 is stored locally. |
| 1995 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=100) |
| 1996 | # One inflight transfer for k1. |
| 1997 | assert len(session._server._inflight) == 1 |
| 1998 | tid_1 = next(iter(session._server._inflight)) |
| 1999 | assert transport._transfers[tid_1][1] == [0] |
| 2000 | assert transport._transfers[tid_1][2] == [10] |
| 2001 | # k2 and k3 still demanded. |
| 2002 | assert set(outbound.demanded.keys()) == {b"k2", b"k3"} |
| 2003 | |
| 2004 | # Transfer 1 completes. |
| 2005 | transport._poll_done.append(tid_1) |
| 2006 | stores = session.poll().stores |
| 2007 | assert StoreResult(job_id=100, success=True) in stores |
| 2008 | assert session._server._inflight == {} |
| 2009 | assert outbound.remaining == 2 |
| 2010 | # Not yet finalized — still 2 blocks demanded. |
| 2011 | assert _srv_outbound(session, "req-1") is not None |
| 2012 | |
| 2013 | # Round 2: k2 and k3 arrive together. |
| 2014 | session.add_stored_blocks("req-1", [b"k2", b"k3"], [1, 2], job_id=200) |
| 2015 | assert len(session._server._inflight) == 1 |
| 2016 | tid_2 = next(iter(session._server._inflight)) |
| 2017 | assert tid_2 != tid_1 |
| 2018 | assert sorted(transport._transfers[tid_2][1]) == [1, 2] |
| 2019 | |
| 2020 | # Transfer 2 completes — request now fully satisfied. |
| 2021 | transport._poll_done.append(tid_2) |
| 2022 | stores = session.poll().stores |
| 2023 | assert StoreResult(job_id=200, success=True) in stores |
| 2024 | # _finalize_outbound fired — request gone, peer notified with success. |
| 2025 | assert _srv_outbound(session, "req-1") is None |
| 2026 | done = next(m for m in conn._sent if m.get(TYPE_KEY) == TransferDoneMsg.TYPE) |
| 2027 | assert done[TransferDoneMsg.KV_REQUEST_ID] == "req-1" |
| 2028 | assert done[TransferDoneMsg.SUCCESS] is True |
| 2029 | |
| 2030 | def test_write_blocks_failure_finalizes_after_last_inflight_completes(self): |
| 2031 | """write_blocks returns None on a SECOND match while a first transfer |
| 2032 | is still inflight. The request should NOT finalize until the inflight |
| 2033 | completes, then the elif branch in collect_results |
| 2034 | (``finishing and not _has_inflight_for(...)``) finalizes it as failure. |
| 2035 | """ |
| 2036 | session, conn, transport = _make_session() |
| 2037 | _activate(session, conn) |
| 2038 | |
| 2039 | # Peer demands two blocks. |
| 2040 | conn.enqueue( |
| 2041 | { |
| 2042 | TYPE_KEY: FetchMsg.TYPE, |
| 2043 | FetchMsg.ROUND_SEQ: 0, |
| 2044 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 2045 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 2046 | FetchMsg.BLOCK_INDEXES: [10, 11], |
| 2047 | } |
| 2048 | ) |
| 2049 | session.poll() |
| 2050 | |
| 2051 | # Round 1: k1 transfers cleanly. |
| 2052 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=100) |
| 2053 | assert len(session._server._inflight) == 1 |
| 2054 | tid_1 = next(iter(session._server._inflight)) |
| 2055 | outbound = _srv_outbound(session, "req-1") |
| 2056 | assert outbound.remaining == 2 # decrement happens on completion |
| 2057 | assert outbound.finishing is False |
| 2058 | |
| 2059 | # Round 2: write_blocks fails for k2 while transfer_1 is still inflight. |
| 2060 | transport.write_blocks = lambda *a, **kw: None # type: ignore[method-assign] |
| 2061 | session.add_stored_blocks("req-1", [b"k2"], [1], job_id=200) |
| 2062 | # No new transfer was registered. |
| 2063 | assert list(session._server._inflight.keys()) == [tid_1] |
| 2064 | # Marked finishing, but NOT finalized yet (transfer_1 still inflight). |
| 2065 | assert outbound.finishing is True |
| 2066 | assert _srv_outbound(session, "req-1") is not None |
| 2067 | done_msgs = [m for m in conn._sent if m.get(TYPE_KEY) == TransferDoneMsg.TYPE] |
| 2068 | assert done_msgs == [] |
| 2069 | |
| 2070 | # Transfer 1 completes — now ``_has_inflight_for("req-1")`` is False |
| 2071 | # and the elif branch in collect_results fires _finalize(success=False). |
| 2072 | # The k1 success result is direct; the k2 failure result is queued |
| 2073 | # in _pending_store_results and surfaces on the NEXT poll. |
| 2074 | transport._poll_done.append(tid_1) |
| 2075 | stores_first = session.poll().stores |
| 2076 | assert StoreResult(job_id=100, success=True) in stores_first |
| 2077 | # Outbound state cleaned up; peer notified with success=False. |
| 2078 | assert _srv_outbound(session, "req-1") is None |
| 2079 | done = next(m for m in conn._sent if m.get(TYPE_KEY) == TransferDoneMsg.TYPE) |
| 2080 | assert done[TransferDoneMsg.KV_REQUEST_ID] == "req-1" |
| 2081 | assert done[TransferDoneMsg.SUCCESS] is False |
| 2082 | |
| 2083 | # Next poll drains the queued failure. |
| 2084 | stores_second = session.poll().stores |
| 2085 | assert StoreResult(job_id=200, success=False) in stores_second |
| 2086 | |
| 2087 | |
| 2088 | # --------------------------------------------------------------------------- |
| 2089 | # Bidirectional — the case the unification is meant to fix |
| 2090 | # --------------------------------------------------------------------------- |
| 2091 | |
| 2092 | |
| 2093 | class TestBidirectional: |
| 2094 | def test_session_handles_both_roles_concurrently(self): |
| 2095 | """Single session simultaneously serves a fetch and completes a load. |
| 2096 | |
| 2097 | This is the regression test for the unification: with the old |
| 2098 | split design the inbound FetchMsg would be dispatched to a |
| 2099 | client-only session and dropped (or a server-only session would |
| 2100 | miss the TransferDoneMsg). One unified session handles both. |
| 2101 | """ |
| 2102 | session, conn, transport = _make_session() |
| 2103 | _activate(session, conn) |
| 2104 | |
| 2105 | # Server role: the peer fetches a block from us. |
| 2106 | session.add_stored_blocks("req-srv", [b"served"], [0], job_id=100) |
| 2107 | conn.enqueue( |
| 2108 | { |
| 2109 | TYPE_KEY: FetchMsg.TYPE, |
| 2110 | FetchMsg.ROUND_SEQ: 0, |
| 2111 | FetchMsg.KV_REQUEST_ID: "req-srv", |
| 2112 | FetchMsg.KEYS: [b"served"], |
| 2113 | FetchMsg.BLOCK_INDEXES: [7], |
| 2114 | } |
| 2115 | ) |
| 2116 | |
| 2117 | # Client role: we ask the peer for a different block. |
| 2118 | session.request_blocks( |
| 2119 | job_id=200, kv_request_id="req-cli", keys=[b"loaded"], block_ids=[3] |
| 2120 | ) |
| 2121 | |
| 2122 | # Both flows progress in the same poll. |
| 2123 | session.poll() |
| 2124 | |
| 2125 | # Server side: write_blocks was submitted. |
| 2126 | assert len(transport._transfers) == 1 |
| 2127 | # Client side: the lookup was sent. |
| 2128 | assert any( |
| 2129 | m[TYPE_KEY] == FetchMsg.TYPE and m[FetchMsg.KV_REQUEST_ID] == "req-cli" |
| 2130 | for m in conn._sent |
| 2131 | ) |
| 2132 | |
| 2133 | # Peer now signals: server-side transfer completes AND a |
| 2134 | # TransferDoneMsg arrives for our client-side request, all in |
| 2135 | # one batch on the same connection. |
| 2136 | tid = next(iter(transport._transfers)) |
| 2137 | transport._poll_done.append(tid) |
| 2138 | conn.enqueue( |
| 2139 | { |
| 2140 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 2141 | TransferDoneMsg.ROUND_SEQ: 0, |
| 2142 | TransferDoneMsg.KV_REQUEST_ID: "req-cli", |
| 2143 | TransferDoneMsg.SUCCESS: True, |
| 2144 | } |
| 2145 | ) |
| 2146 | result_ = session.poll() |
| 2147 | loads = result_.loads |
| 2148 | stores = result_.stores |
| 2149 | |
| 2150 | assert LoadResult(job_id=200, kv_request_id="req-cli", success=True) in loads |
| 2151 | assert StoreResult(job_id=100, success=True) in stores |
| 2152 | |
| 2153 | |
| 2154 | # --------------------------------------------------------------------------- |
| 2155 | # Pending sessions (no connection yet) |
| 2156 | # --------------------------------------------------------------------------- |
| 2157 | |
| 2158 | |
| 2159 | class TestPendingSession: |
| 2160 | def test_pending_session_buffers_stored_blocks(self): |
| 2161 | """Pending session accepts add_stored_blocks but cannot send.""" |
| 2162 | transport = FakeDataTransport() |
| 2163 | session = P2PSession( |
| 2164 | peer_id="peer:8000", |
| 2165 | local_id="local:9000", |
| 2166 | transport=transport, |
| 2167 | local_block_len=4096, |
| 2168 | local_hash_seed=_DEFAULT_HASH_SEED, |
| 2169 | conn=None, |
| 2170 | ) |
| 2171 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 2172 | result_ = session.poll() |
| 2173 | loads = result_.loads |
| 2174 | stores = result_.stores |
| 2175 | assert loads == [] |
| 2176 | assert stores == [] |
| 2177 | assert not session.connected |
| 2178 | assert session.alive |
| 2179 | |
| 2180 | def test_attach_connection_sends_connect(self): |
| 2181 | """attach_connection triggers our ConnectMsg send.""" |
| 2182 | transport = FakeDataTransport() |
| 2183 | session = P2PSession( |
| 2184 | peer_id="peer:8000", |
| 2185 | local_id="local:9000", |
| 2186 | transport=transport, |
| 2187 | local_block_len=4096, |
| 2188 | local_hash_seed=_DEFAULT_HASH_SEED, |
| 2189 | conn=None, |
| 2190 | ) |
| 2191 | conn = FakeConnection(peer_id="peer:8000") |
| 2192 | session.attach_connection(conn) |
| 2193 | assert conn._sent |
| 2194 | assert conn._sent[0][TYPE_KEY] == ConnectMsg.TYPE |
| 2195 | |
| 2196 | def test_attach_connection_twice_raises(self): |
| 2197 | """attach_connection on an already-connected session raises.""" |
| 2198 | session, conn, _ = _make_session() |
| 2199 | with pytest.raises(ValueError, match="already connected"): |
| 2200 | session.attach_connection(FakeConnection()) |
| 2201 | |
| 2202 | def test_pending_close_returns_pending_stores(self): |
| 2203 | """Closing a pending session reports buffered stores as failed.""" |
| 2204 | transport = FakeDataTransport() |
| 2205 | session = P2PSession( |
| 2206 | peer_id="peer:8000", |
| 2207 | local_id="local:9000", |
| 2208 | transport=transport, |
| 2209 | local_block_len=4096, |
| 2210 | local_hash_seed=_DEFAULT_HASH_SEED, |
| 2211 | conn=None, |
| 2212 | ) |
| 2213 | session.add_stored_blocks("req-1", [b"k1"], [0], job_id=1) |
| 2214 | session.add_stored_blocks("req-2", [b"k2"], [1], job_id=2) |
| 2215 | result = session.close() |
| 2216 | assert result.failed_jobs == [] |
| 2217 | assert result.failed_req_ids == [] |
| 2218 | assert set(result.failed_stores) == {1, 2} |
| 2219 | assert result.failed_serves == [] |
| 2220 | |
| 2221 | |
| 2222 | # --------------------------------------------------------------------------- |
| 2223 | # Disconnect |
| 2224 | # --------------------------------------------------------------------------- |
| 2225 | |
| 2226 | |
| 2227 | class TestDisconnect: |
| 2228 | def test_disconnect_marks_session_dead(self): |
| 2229 | session, conn, _ = _make_session() |
| 2230 | _activate(session, conn) |
| 2231 | conn.enqueue({TYPE_KEY: DisconnectMsg.TYPE}) |
| 2232 | session.poll() |
| 2233 | assert not session.alive |
| 2234 | |
| 2235 | def test_close_returns_pending_loads_and_stores(self): |
| 2236 | session, conn, _ = _make_session() |
| 2237 | _activate(session, conn) |
| 2238 | session.request_blocks(1, "req-1", [b"k"], [0]) |
| 2239 | session.request_blocks(2, "req-2", [b"k"], [0]) |
| 2240 | session.add_stored_blocks("req-srv", [b"k"], [0], job_id=10) |
| 2241 | result = session.close() |
| 2242 | assert set(result.failed_jobs) == {1, 2} |
| 2243 | assert set(result.failed_req_ids) == {"req-1", "req-2"} |
| 2244 | assert set(result.failed_stores) == {10} |
| 2245 | assert result.failed_serves == [] |
| 2246 | |
| 2247 | def test_send_failure_marks_connection_dead(self): |
| 2248 | """A raising send must mark the connection dead, not silently drop |
| 2249 | the message — otherwise the session lingers alive, is never reaped, |
| 2250 | and in-flight lookups/loads toward the dead peer hang forever.""" |
| 2251 | session, conn, _ = _make_session() |
| 2252 | _activate(session, conn) |
| 2253 | assert session.alive |
| 2254 | |
| 2255 | conn.fail_send = True |
| 2256 | # request_blocks flushes a FetchMsg synchronously via _do_send. |
| 2257 | session.request_blocks(1, "req-1", [b"k"], [0]) |
| 2258 | |
| 2259 | assert not session.alive |
| 2260 | |
| 2261 | def test_close_surfaces_inflight_lookups(self): |
| 2262 | """close() reports kv_request_ids whose symmetric-P2P probe is still |
| 2263 | unresolved; resolved probes are not reported (their answer is in).""" |
| 2264 | session, conn, _ = _make_session() |
| 2265 | _activate(session, conn) |
| 2266 | |
| 2267 | session.register_lookup("req-hit", b"hA") |
| 2268 | session.register_lookup("req-inflight", b"hB") |
| 2269 | session.flush_pending_lookups() |
| 2270 | |
| 2271 | # Only req-hit is answered; req-inflight stays in flight. |
| 2272 | conn.enqueue( |
| 2273 | { |
| 2274 | TYPE_KEY: LookupRespMsg.TYPE, |
| 2275 | LookupRespMsg.KV_REQUEST_ID: "req-hit", |
| 2276 | LookupRespMsg.KEYS: [b"hA"], |
| 2277 | LookupRespMsg.HITS: [True], |
| 2278 | } |
| 2279 | ) |
| 2280 | session.poll() |
| 2281 | |
| 2282 | result = session.close() |
| 2283 | assert result.failed_jobs == [] |
| 2284 | assert result.failed_req_ids == ["req-inflight"] |
| 2285 | |
| 2286 | |
| 2287 | # --------------------------------------------------------------------------- |
| 2288 | # Adversarial / malformed messages |
| 2289 | # --------------------------------------------------------------------------- |
| 2290 | |
| 2291 | |
| 2292 | class TestAdversarial: |
| 2293 | def test_unknown_message_type_logged(self): |
| 2294 | session, conn, _ = _make_session() |
| 2295 | _activate(session, conn) |
| 2296 | conn.enqueue({TYPE_KEY: "evil_command"}) |
| 2297 | result_ = session.poll() |
| 2298 | loads = result_.loads |
| 2299 | stores = result_.stores |
| 2300 | assert loads == [] |
| 2301 | assert stores == [] |
| 2302 | |
| 2303 | def test_empty_message(self): |
| 2304 | session, conn, _ = _make_session() |
| 2305 | _activate(session, conn) |
| 2306 | conn.enqueue({}) |
| 2307 | result_ = session.poll() |
| 2308 | loads = result_.loads |
| 2309 | stores = result_.stores |
| 2310 | assert loads == [] |
| 2311 | assert stores == [] |
| 2312 | |
| 2313 | def test_non_dict_message(self): |
| 2314 | session, conn, _ = _make_session() |
| 2315 | _activate(session, conn) |
| 2316 | conn._inbox.append(42) |
| 2317 | result_ = session.poll() |
| 2318 | loads = result_.loads |
| 2319 | stores = result_.stores |
| 2320 | assert loads == [] |
| 2321 | assert stores == [] |
| 2322 | |
| 2323 | def test_fetch_mismatched_lengths(self): |
| 2324 | session, conn, transport = _make_session() |
| 2325 | _activate(session, conn) |
| 2326 | conn.enqueue( |
| 2327 | { |
| 2328 | TYPE_KEY: FetchMsg.TYPE, |
| 2329 | FetchMsg.ROUND_SEQ: 0, |
| 2330 | FetchMsg.KV_REQUEST_ID: "req-bad", |
| 2331 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 2332 | FetchMsg.BLOCK_INDEXES: [1], |
| 2333 | } |
| 2334 | ) |
| 2335 | session.poll() |
| 2336 | assert len(transport._transfers) == 0 |
| 2337 | # Protocol violation: session disconnects immediately so the peer |
| 2338 | # can't keep wedging us with malformed traffic. |
| 2339 | assert not session.alive |
| 2340 | assert any(m[TYPE_KEY] == DisconnectMsg.TYPE for m in conn._sent) |
| 2341 | |
| 2342 | def test_transfer_done_missing_kv_request_id(self): |
| 2343 | session, conn, _ = _make_session() |
| 2344 | _activate(session, conn) |
| 2345 | conn.enqueue({TYPE_KEY: TransferDoneMsg.TYPE, TransferDoneMsg.SUCCESS: True}) |
| 2346 | loads = session.poll().loads |
| 2347 | assert loads == [] |
| 2348 | assert not session.alive |
| 2349 | assert any(m[TYPE_KEY] == DisconnectMsg.TYPE for m in conn._sent) |
| 2350 | |
| 2351 | def test_duplicate_connect_ack(self): |
| 2352 | session, conn, _ = _make_session() |
| 2353 | _activate(session, conn) |
| 2354 | conn.enqueue({TYPE_KEY: ConnectAckMsg.TYPE, ConnectAckMsg.PEER_ID: "peer:8000"}) |
| 2355 | session.poll() |
| 2356 | assert session.ready |
| 2357 | |
| 2358 | |
| 2359 | class TestDispatchErrorHandling: |
| 2360 | """Errors raised by message handlers split into two classes: |
| 2361 | |
| 2362 | - Protocol-contract violations from the peer (ValueError) → disconnect |
| 2363 | on the first occurrence; retrying won't help and may corrupt state. |
| 2364 | - Anything else is treated as an internal bug: log loudly, count, and |
| 2365 | only disconnect once errors arrive in a tight burst. A successful |
| 2366 | dispatch in between resets the counter. |
| 2367 | """ |
| 2368 | |
| 2369 | def test_value_error_disconnects_on_first_occurrence(self): |
| 2370 | """A FetchMsg that fails validate() raises ValueError and must |
| 2371 | terminate the session immediately, with a DisconnectMsg sent.""" |
| 2372 | session, conn, _ = _make_session() |
| 2373 | _activate(session, conn) |
| 2374 | # length mismatch → FetchMsg.validate raises ValueError |
| 2375 | conn.enqueue( |
| 2376 | { |
| 2377 | TYPE_KEY: FetchMsg.TYPE, |
| 2378 | FetchMsg.ROUND_SEQ: 0, |
| 2379 | FetchMsg.KV_REQUEST_ID: "req-bad", |
| 2380 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 2381 | FetchMsg.BLOCK_INDEXES: [1], |
| 2382 | } |
| 2383 | ) |
| 2384 | session.poll() |
| 2385 | assert not session.alive |
| 2386 | assert any(m[TYPE_KEY] == DisconnectMsg.TYPE for m in conn._sent) |
| 2387 | |
| 2388 | def test_transfer_done_missing_field_disconnects(self): |
| 2389 | """Same contract for a malformed TransferDoneMsg from the peer.""" |
| 2390 | session, conn, _ = _make_session() |
| 2391 | _activate(session, conn) |
| 2392 | conn.enqueue({TYPE_KEY: TransferDoneMsg.TYPE, TransferDoneMsg.SUCCESS: True}) |
| 2393 | session.poll() |
| 2394 | assert not session.alive |
| 2395 | |
| 2396 | def test_internal_error_does_not_disconnect_once(self): |
| 2397 | """A non-ValueError raised by a handler is treated as an internal |
| 2398 | bug: counter increments, session stays alive on a single hit.""" |
| 2399 | session, conn, _ = _make_session() |
| 2400 | _activate(session, conn) |
| 2401 | |
| 2402 | def _boom(*args, **kwargs): |
| 2403 | raise RuntimeError("simulated internal bug") |
| 2404 | |
| 2405 | session._server.on_fetch = _boom |
| 2406 | conn.enqueue( |
| 2407 | { |
| 2408 | TYPE_KEY: FetchMsg.TYPE, |
| 2409 | FetchMsg.ROUND_SEQ: 0, |
| 2410 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 2411 | FetchMsg.KEYS: [b"k1"], |
| 2412 | FetchMsg.BLOCK_INDEXES: [0], |
| 2413 | } |
| 2414 | ) |
| 2415 | session.poll() |
| 2416 | assert session.alive |
| 2417 | assert session._dispatch_error_count == 1 |
| 2418 | |
| 2419 | def test_internal_error_threshold_disconnects(self): |
| 2420 | """Once consecutive non-protocol errors hit the threshold, the |
| 2421 | session tears down via _protocol_error.""" |
| 2422 | session, conn, _ = _make_session() |
| 2423 | _activate(session, conn) |
| 2424 | |
| 2425 | def _boom(*args, **kwargs): |
| 2426 | raise RuntimeError("simulated internal bug") |
| 2427 | |
| 2428 | session._server.on_fetch = _boom |
| 2429 | for _ in range(_MAX_CONSECUTIVE_DISPATCH_ERRORS): |
| 2430 | conn.enqueue( |
| 2431 | { |
| 2432 | TYPE_KEY: FetchMsg.TYPE, |
| 2433 | FetchMsg.ROUND_SEQ: 0, |
| 2434 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 2435 | FetchMsg.KEYS: [b"k1"], |
| 2436 | FetchMsg.BLOCK_INDEXES: [0], |
| 2437 | } |
| 2438 | ) |
| 2439 | session.poll() |
| 2440 | assert not session.alive |
| 2441 | assert any(m[TYPE_KEY] == DisconnectMsg.TYPE for m in conn._sent) |
| 2442 | |
| 2443 | def test_internal_error_counter_resets_on_success(self): |
| 2444 | """A successful dispatch between errors prevents the threshold |
| 2445 | from being reached.""" |
| 2446 | session, conn, _ = _make_session() |
| 2447 | _activate(session, conn) |
| 2448 | |
| 2449 | original_on_fetch = session._server.on_fetch |
| 2450 | |
| 2451 | def _boom(*args, **kwargs): |
| 2452 | raise RuntimeError("simulated internal bug") |
| 2453 | |
| 2454 | # Alternate (boom, success) (_MAX-1) times: counter rises to 1 |
| 2455 | # then resets to 0 each cycle, never reaching the threshold. |
| 2456 | for _ in range(_MAX_CONSECUTIVE_DISPATCH_ERRORS - 1): |
| 2457 | session._server.on_fetch = _boom |
| 2458 | conn.enqueue( |
| 2459 | { |
| 2460 | TYPE_KEY: FetchMsg.TYPE, |
| 2461 | FetchMsg.ROUND_SEQ: 0, |
| 2462 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 2463 | FetchMsg.KEYS: [b"k1"], |
| 2464 | FetchMsg.BLOCK_INDEXES: [0], |
| 2465 | } |
| 2466 | ) |
| 2467 | session.poll() |
| 2468 | session._server.on_fetch = original_on_fetch |
| 2469 | # A benign no-op message (unknown type) dispatches cleanly |
| 2470 | # and resets the consecutive-error counter. |
| 2471 | conn.enqueue({TYPE_KEY: "unknown_for_test"}) |
| 2472 | session.poll() |
| 2473 | |
| 2474 | assert session.alive |
| 2475 | assert session._dispatch_error_count == 0 |
| 2476 | |
| 2477 | |
| 2478 | class TestInflightPerReqInvariant: |
| 2479 | """Per-request `inflight_tids` is the O(1) replacement for the |
| 2480 | previous O(N) scan in `_has_inflight_for`. These tests check that |
| 2481 | every mutation site keeps the set in sync with `_inflight` and |
| 2482 | that the lookup is correct under high fan-out. |
| 2483 | """ |
| 2484 | |
| 2485 | def test_invariant_holds_through_lifecycle(self): |
| 2486 | """Run a full submit→complete sequence for two concurrent |
| 2487 | kv_request_ids and assert the counter matches `_inflight` at |
| 2488 | every observable step, including the empty-after-finish case.""" |
| 2489 | session, conn, transport = _make_session() |
| 2490 | _activate(session, conn) |
| 2491 | |
| 2492 | def _invariant_holds() -> bool: |
| 2493 | return _srv_total_inflight(session) == len(session._server._inflight) |
| 2494 | |
| 2495 | assert _invariant_holds() |
| 2496 | |
| 2497 | # Two requests, two blocks each, all dispatched in one batch. |
| 2498 | session.add_stored_blocks("req-A", [b"a1", b"a2"], [0, 1], job_id=10) |
| 2499 | session.add_stored_blocks("req-B", [b"b1", b"b2"], [2, 3], job_id=11) |
| 2500 | for kv_id, keys, indexes in ( |
| 2501 | ("req-A", [b"a1", b"a2"], [100, 101]), |
| 2502 | ("req-B", [b"b1", b"b2"], [102, 103]), |
| 2503 | ): |
| 2504 | conn.enqueue( |
| 2505 | { |
| 2506 | TYPE_KEY: FetchMsg.TYPE, |
| 2507 | FetchMsg.ROUND_SEQ: 0, |
| 2508 | FetchMsg.KV_REQUEST_ID: kv_id, |
| 2509 | FetchMsg.KEYS: keys, |
| 2510 | FetchMsg.BLOCK_INDEXES: indexes, |
| 2511 | } |
| 2512 | ) |
| 2513 | session.poll() |
| 2514 | |
| 2515 | assert _invariant_holds() |
| 2516 | assert session._server._has_inflight_for("req-A") |
| 2517 | assert session._server._has_inflight_for("req-B") |
| 2518 | assert not session._server._has_inflight_for("req-C") |
| 2519 | |
| 2520 | # Complete req-A's transfer first; req-B should still be inflight. |
| 2521 | a_tids = [ |
| 2522 | tid |
| 2523 | for tid, x in session._server._inflight.items() |
| 2524 | if x.kv_request_id == "req-A" |
| 2525 | ] |
| 2526 | for tid in a_tids: |
| 2527 | transport._poll_done.append(tid) |
| 2528 | session.poll() |
| 2529 | |
| 2530 | assert _invariant_holds() |
| 2531 | assert not session._server._has_inflight_for("req-A") |
| 2532 | assert _srv_inflight_count(session, "req-A") == 0 # entry drained |
| 2533 | assert session._server._has_inflight_for("req-B") |
| 2534 | |
| 2535 | # Complete req-B; counter must drain to empty. |
| 2536 | b_tids = [ |
| 2537 | tid |
| 2538 | for tid, x in session._server._inflight.items() |
| 2539 | if x.kv_request_id == "req-B" |
| 2540 | ] |
| 2541 | for tid in b_tids: |
| 2542 | transport._poll_done.append(tid) |
| 2543 | session.poll() |
| 2544 | |
| 2545 | assert _invariant_holds() |
| 2546 | assert session._server._inflight == {} |
| 2547 | assert _srv_total_inflight(session) == 0 |
| 2548 | |
| 2549 | def test_has_inflight_for_correct_with_many_requests(self): |
| 2550 | """Populate many inflight xfers across many ids; lookup must |
| 2551 | match the actual presence in `_inflight` for both hits and |
| 2552 | misses. The whole point of the counter is that this lookup is |
| 2553 | constant-time, but we assert correctness, not timing.""" |
| 2554 | session, _, _ = _make_session() |
| 2555 | for kv_id_idx in range(100): |
| 2556 | kv_id = f"req-{kv_id_idx}" |
| 2557 | for j in range(10): |
| 2558 | tid = kv_id_idx * 10 + j |
| 2559 | session._server._inflight_add( |
| 2560 | tid, |
| 2561 | _InflightXfer( |
| 2562 | kv_request_id=kv_id, |
| 2563 | block_count=1, |
| 2564 | job_ids={tid}, |
| 2565 | round=_OutboundRequestState(inflight=1), |
| 2566 | ), |
| 2567 | ) |
| 2568 | assert _srv_total_inflight(session) == len(session._server._inflight) |
| 2569 | assert session._server._has_inflight_for("req-0") |
| 2570 | assert session._server._has_inflight_for("req-99") |
| 2571 | assert not session._server._has_inflight_for("req-missing") |
| 2572 | |
| 2573 | # Drain all entries for req-50 and confirm the entry disappears. |
| 2574 | tids_50 = [ |
| 2575 | tid |
| 2576 | for tid, x in session._server._inflight.items() |
| 2577 | if x.kv_request_id == "req-50" |
| 2578 | ] |
| 2579 | for tid in tids_50: |
| 2580 | session._server._inflight_pop(tid) |
| 2581 | assert _srv_inflight_count(session, "req-50") == 0 |
| 2582 | assert not session._server._has_inflight_for("req-50") |
| 2583 | # Other ids unaffected. |
| 2584 | assert session._server._has_inflight_for("req-49") |
| 2585 | |
| 2586 | |
| 2587 | # --------------------------------------------------------------------------- |
| 2588 | # Protocol validation tests (unchanged from the prior file) |
| 2589 | # --------------------------------------------------------------------------- |
| 2590 | |
| 2591 | |
| 2592 | class TestConnectMsgValidation: |
| 2593 | def _valid_msg(self) -> dict: |
| 2594 | return { |
| 2595 | TYPE_KEY: ConnectMsg.TYPE, |
| 2596 | ConnectMsg.PEER_ID: "peer:1", |
| 2597 | ConnectMsg.AGENT_METADATA: b"meta", |
| 2598 | ConnectMsg.BASE_ADDR: 0x1000, |
| 2599 | ConnectMsg.NUM_BLOCKS: 8, |
| 2600 | ConnectMsg.BLOCK_LEN: 4096, |
| 2601 | ConnectMsg.HASH_SEED: "0", |
| 2602 | } |
| 2603 | |
| 2604 | def test_valid_message_passes(self): |
| 2605 | ConnectMsg.validate(self._valid_msg()) |
| 2606 | |
| 2607 | def test_missing_peer_id(self): |
| 2608 | msg = self._valid_msg() |
| 2609 | del msg[ConnectMsg.PEER_ID] |
| 2610 | with pytest.raises(ValueError, match="peer_id"): |
| 2611 | ConnectMsg.validate(msg) |
| 2612 | |
| 2613 | def test_missing_agent_metadata(self): |
| 2614 | msg = self._valid_msg() |
| 2615 | del msg[ConnectMsg.AGENT_METADATA] |
| 2616 | with pytest.raises(ValueError, match="agent_metadata"): |
| 2617 | ConnectMsg.validate(msg) |
| 2618 | |
| 2619 | def test_agent_metadata_wrong_type(self): |
| 2620 | msg = self._valid_msg() |
| 2621 | msg[ConnectMsg.AGENT_METADATA] = "not bytes" |
| 2622 | with pytest.raises(ValueError, match="agent_metadata"): |
| 2623 | ConnectMsg.validate(msg) |
| 2624 | |
| 2625 | def test_base_addr_negative(self): |
| 2626 | msg = self._valid_msg() |
| 2627 | msg[ConnectMsg.BASE_ADDR] = -1 |
| 2628 | with pytest.raises(ValueError, match="base_addr"): |
| 2629 | ConnectMsg.validate(msg) |
| 2630 | |
| 2631 | def test_num_blocks_zero(self): |
| 2632 | msg = self._valid_msg() |
| 2633 | msg[ConnectMsg.NUM_BLOCKS] = 0 |
| 2634 | with pytest.raises(ValueError, match="num_blocks"): |
| 2635 | ConnectMsg.validate(msg) |
| 2636 | |
| 2637 | def test_block_len_zero(self): |
| 2638 | msg = self._valid_msg() |
| 2639 | msg[ConnectMsg.BLOCK_LEN] = 0 |
| 2640 | with pytest.raises(ValueError, match="block_len"): |
| 2641 | ConnectMsg.validate(msg) |
| 2642 | |
| 2643 | def test_missing_hash_seed(self): |
| 2644 | msg = self._valid_msg() |
| 2645 | del msg[ConnectMsg.HASH_SEED] |
| 2646 | with pytest.raises(ValueError, match="hash_seed"): |
| 2647 | ConnectMsg.validate(msg) |
| 2648 | |
| 2649 | def test_hash_seed_wrong_type(self): |
| 2650 | msg = self._valid_msg() |
| 2651 | msg[ConnectMsg.HASH_SEED] = 12345 # int, not str |
| 2652 | with pytest.raises(ValueError, match="hash_seed"): |
| 2653 | ConnectMsg.validate(msg) |
| 2654 | |
| 2655 | |
| 2656 | class TestFetchMsgValidation: |
| 2657 | def _valid_msg(self) -> dict: |
| 2658 | return { |
| 2659 | TYPE_KEY: FetchMsg.TYPE, |
| 2660 | FetchMsg.ROUND_SEQ: 0, |
| 2661 | FetchMsg.KV_REQUEST_ID: "req-1", |
| 2662 | FetchMsg.KEYS: [b"k1", b"k2"], |
| 2663 | FetchMsg.BLOCK_INDEXES: [0, 1], |
| 2664 | } |
| 2665 | |
| 2666 | def test_valid_message_passes(self): |
| 2667 | FetchMsg.validate(self._valid_msg()) |
| 2668 | |
| 2669 | def test_length_mismatch(self): |
| 2670 | msg = self._valid_msg() |
| 2671 | msg[FetchMsg.BLOCK_INDEXES] = [0] |
| 2672 | with pytest.raises(ValueError, match="length mismatch"): |
| 2673 | FetchMsg.validate(msg) |
| 2674 | |
| 2675 | def test_negative_index(self): |
| 2676 | msg = self._valid_msg() |
| 2677 | msg[FetchMsg.BLOCK_INDEXES] = [0, -1] |
| 2678 | with pytest.raises(ValueError, match="invalid index"): |
| 2679 | FetchMsg.validate(msg) |
| 2680 | |
| 2681 | |
| 2682 | class TestTransferDoneMsgValidation: |
| 2683 | def test_valid_message_passes(self): |
| 2684 | msg = { |
| 2685 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 2686 | TransferDoneMsg.ROUND_SEQ: 0, |
| 2687 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 2688 | TransferDoneMsg.SUCCESS: True, |
| 2689 | } |
| 2690 | TransferDoneMsg.validate(msg) |
| 2691 | |
| 2692 | def test_success_wrong_type(self): |
| 2693 | msg = { |
| 2694 | TYPE_KEY: TransferDoneMsg.TYPE, |
| 2695 | TransferDoneMsg.ROUND_SEQ: 0, |
| 2696 | TransferDoneMsg.KV_REQUEST_ID: "req-1", |
| 2697 | TransferDoneMsg.SUCCESS: 1, |
| 2698 | } |
| 2699 | with pytest.raises(ValueError, match="success"): |
| 2700 | TransferDoneMsg.validate(msg) |