| 1 | #!/usr/bin/env python3 |
| 2 | """SIGKILL an actual native editing process mid-typing, reopen offline, drain |
| 3 | against real authenticated Docs handlers, and compare the server snapshot. |
| 4 | Requires the CRDT framework built by check-native.sh; never touches user data. |
| 5 | """ |
| 6 | import argparse |
| 7 | import random |
| 8 | import os |
| 9 | from pathlib import Path |
| 10 | import signal |
| 11 | import subprocess |
| 12 | import tempfile |
| 13 | import time |
| 14 | |
| 15 | parser = argparse.ArgumentParser(description=__doc__) |
| 16 | parser.add_argument("--random", action="store_true", help="kill during unpaused typing and concurrent replica writes") |
| 17 | parser.add_argument("--seed", type=int, default=None, help="reproduce the controller's random delay") |
| 18 | args = parser.parse_args() |
| 19 | seed = args.seed if args.seed is not None else random.SystemRandom().randrange(2**32) |
| 20 | kill_delay = random.Random(seed).uniform(0, .5) |
| 21 | root = Path(__file__).resolve().parents[2] |
| 22 | |
| 23 | def wait_for(path, process, timeout=300): |
| 24 | deadline = time.monotonic() + timeout |
| 25 | while not path.exists(): |
| 26 | if process.poll() is not None: |
| 27 | raise RuntimeError(f"process exited {process.returncode} before {path.name}") |
| 28 | if time.monotonic() > deadline: |
| 29 | raise TimeoutError(path.name) |
| 30 | time.sleep(.02) |
| 31 | |
| 32 | with tempfile.TemporaryDirectory(prefix="docs-k3-kill-", dir="/tmp") as tmp: |
| 33 | scratch = Path(tmp) |
| 34 | env = dict(os.environ, DOCS_KILL_DIR=tmp, DOCS_KILL_RANDOM="1" if args.random else "0", TMPDIR="/tmp") |
| 35 | server_log = (scratch / "server.log").open("w") |
| 36 | server = subprocess.Popen(["go", "test", "./internal/docs", "-run", "^TestNativeKillServer$", "-count=1", "-timeout=8m"], cwd=root, env=env, stdout=server_log, stderr=subprocess.STDOUT, start_new_session=True) |
| 37 | writer = None |
| 38 | try: |
| 39 | wait_for(scratch / "connection.json", server) |
| 40 | command = ["swift", "test", "--package-path", "native/apple", "--filter", "OfflineKillTests"] |
| 41 | with (scratch / "writer.log").open("w") as log: |
| 42 | writer = subprocess.Popen(command, cwd=root, env=dict(env, DOCS_KILL_MODE="write"), stdout=log, stderr=subprocess.STDOUT, start_new_session=True) |
| 43 | wait_for(scratch / "kill-now", writer) |
| 44 | if args.random: |
| 45 | # Only the controller waits; the editing process never pauses |
| 46 | # for us and can die during a command, enqueue, or writer tick. |
| 47 | time.sleep(kill_delay) |
| 48 | pid = int((scratch / "pid").read_text()) |
| 49 | os.kill(pid, signal.SIGKILL) |
| 50 | result = writer.wait(timeout=20) |
| 51 | assert result != 0, "typing process unexpectedly succeeded" |
| 52 | # Include assertions emitted during the controller’s random delay. |
| 53 | assert ": error:" not in (scratch / "writer.log").read_text(), "typing assertions failed before kill" |
| 54 | if args.random: |
| 55 | print(f"SIGKILL during unpaused typing + writer ticks (seed={seed}, delay={kill_delay:.6f}s)", flush=True) |
| 56 | else: |
| 57 | print("SIGKILL delivered after 12 durable edits, before coalesced replica write", flush=True) |
| 58 | recovered = subprocess.run(command, cwd=root, env=dict(env, DOCS_KILL_MODE="recover"), capture_output=True, text=True, timeout=120) |
| 59 | if recovered.returncode: |
| 60 | print(recovered.stdout) |
| 61 | print(recovered.stderr) |
| 62 | raise RuntimeError("offline recovery / real server convergence failed") |
| 63 | (scratch / "done").touch() |
| 64 | assert server.wait(timeout=15) == 0, "real server fixture failed" |
| 65 | print("PASS: offline reopen preserves recorded prefix; real server equals reopened replica; no gaps or duplicate text", flush=True) |
| 66 | except Exception: |
| 67 | # Logs contain only test-generated state. Never print connection.json. |
| 68 | for name in ("server.log", "writer.log"): |
| 69 | path = scratch / name |
| 70 | if path.exists(): |
| 71 | print(path.read_text()[-12000:]) |
| 72 | raise |
| 73 | finally: |
| 74 | for process in (writer, server): |
| 75 | if process is not None and process.poll() is None: |
| 76 | os.killpg(process.pid, signal.SIGTERM) |
| 77 | process.wait(timeout=10) |
| 78 | server_log.close() |
| 79 | |
amadan