Automate browser based workflows with AI
Python aiapiautomationbrowserbrowser-automationcomputergptllmplaywrightpowerautomatepuppeteerpythonrpaseleniumvisionworkflow
fix(SKY-11153): land adopted-session downloads run-side via eager save + url re-fetch (#6679)
Andrew Neilson3010c0ab51df
diff --git a/skyvern/webeye/actions/handler.py b/skyvern/webeye/actions/handler.pyindex 8f7f57c59..3fdcc6755 100644--- a/skyvern/webeye/actions/handler.py+++ b/skyvern/webeye/actions/handler.py@@ -158,6 +158,66 @@ def _download_target_path(download_dir: Path, suggested_filename: str | None) ->return download_dir / f"{uuid.uuid4()}-{stem or 'download'}{suffix}"+async def _save_adopted_session_download(+ download: Download,+ page: Page,+ download_dir: Path,+ workflow_run_id: str | None = None,+) -> Path | None:+ """Land an adopted-session download's bytes into download_dir, returning the file path or None.++ Eager save_as is the only protection against the worker pod tearing the shared browser down before a+ deferred save_as runs.+ """+ download_target = _download_target_path(download_dir, download.suggested_filename)+ try:+ await download.save_as(download_target)+ if download_target.exists() and download_target.stat().st_size > 0:+ return download_target+ download_target.unlink(missing_ok=True)+ LOG.warning(+ "Adopted-session eager save_as produced an empty file; re-fetching download url",+ download_dir=str(download_dir),+ workflow_run_id=workflow_run_id,+ )+ except Exception:+ download_target.unlink(missing_ok=True)+ LOG.warning(+ "Adopted-session eager save_as failed; re-fetching download url",+ download_dir=str(download_dir),+ workflow_run_id=workflow_run_id,+ exc_info=True,+ )++ try:+ response = await page.context.request.get(download.url)+ if response.status != 200:+ LOG.error(+ "Adopted-session download url re-fetch returned non-200 status",+ status=response.status,+ workflow_run_id=workflow_run_id,+ )+ return None+ # APIResponse.body() has no streaming variant, so a large download peaks at 2x its size in RSS.+ body = await response.body()+ if not body:+ LOG.error(+ "Adopted-session download url re-fetch returned an empty body",+ workflow_run_id=workflow_run_id,+ )+ return None+ download_target.write_bytes(body)+ return download_target+ except Exception:+ LOG.error(+ "Adopted-session download url re-fetch failed",+ download_dir=str(download_dir),+ workflow_run_id=workflow_run_id,+ exc_info=True,+ )+ return None++def _remove_download_listener(page: Page, callback: Callable[[Download], None]) -> None:off = getattr(page, "off", None)if callable(off):@@ -831,6 +891,36 @@ class ActionHandler:download_signal_poll_iterations=download_signal_poll_iterations,)+ if (+ task.browser_session_id+ and captured_download is not None+ and not download_event_fallback_attempted+ ):+ download_event_fallback_attempted = True+ saved_path = await _save_adopted_session_download(+ captured_download,+ page,+ download_dir,+ workflow_run_id=task.workflow_run_id,+ )+ if saved_path is not None:+ download_event_fallback_used = True+ download_triggered = True+ LOG.info(+ "Saved adopted-session download to active run directory",+ download_dir=download_dir,+ download_target=str(saved_path),+ workflow_run_id=task.workflow_run_id,+ )+ break+ download_event_fallback_failed = True+ LOG.warning(+ "Adopted-session download could not be saved or re-fetched",+ download_dir=download_dir,+ workflow_run_id=task.workflow_run_id,+ )+ break+list_files_after = await _list_observed_download_files()if len(list_files_after) > len(list_files_before):@@ -847,7 +937,8 @@ class ActionHandler:breakif (- captured_download is not None+ not task.browser_session_id+ and captured_download is not Noneand download_event_captured_at is not Noneand not download_event_fallback_attemptedand time.monotonic() - download_event_captured_at >= _download_event_grace_secondsdiff --git a/skyvern/webeye/browser_factory.py b/skyvern/webeye/browser_factory.pyindex f1e0b8cfc..b5de11402 100644--- a/skyvern/webeye/browser_factory.py+++ b/skyvern/webeye/browser_factory.py@@ -198,6 +198,14 @@ def set_popup_video_listener(browser_context: BrowserContext, browser_artifacts:asyncio.ensure_future(_on_page(page))+def _redact_url_query(url: str) -> str:+ # Download URLs are often S3 presigned, carrying an X-Amz signature in the query — drop it before logging.+ try:+ return urlparse(url)._replace(query="").geturl()+ except Exception:+ return "<redacted>"++def set_download_file_listener(browser_context: BrowserContext, download_timeout: float | None = None, **kwargs: Any) -> None:@@ -208,6 +216,16 @@ def set_download_file_listener(try:async with asyncio.timeout(download_timeout or BROWSER_DOWNLOAD_TIMEOUT):file_path = await download.path()+ if not file_path.exists():+ # On an adopted persistent session the bytes live on the run connection, not+ # this worker connection; saving is the run side's job, so skip rather than crash.+ LOG.warning(+ "Download artifact absent on this connection; skipping worker-side rename",+ workflow_run_id=workflow_run_id,+ task_id=task_id,+ suggested_filename=download.suggested_filename,+ )+ returnif file_path.suffix:return@@ -216,7 +234,7 @@ def set_download_file_listener(workflow_run_id=workflow_run_id,task_id=task_id,suggested_filename=download.suggested_filename,- url=download.url,+ url=_redact_url_query(download.url),)suffix = Path(download.suggested_filename).suffixif suffix:diff --git a/tests/unit/test_adopted_session_download_save.py b/tests/unit/test_adopted_session_download_save.pynew file mode 100644index 000000000..230c764c6--- /dev/null+++ b/tests/unit/test_adopted_session_download_save.py@@ -0,0 +1,139 @@+"""Eager-save-then-refetch for adopted persistent-session downloads.++On an adopted session the run connection owns the download artifact, but in prod the+worker pod can tear the shared browser down before a deferred save_as runs. The helper+saves eagerly and, when save_as raises (TargetClosedError) or yields a 0-byte file,+re-fetches the replayable download url through the run page's request context.+"""++from pathlib import Path+from unittest.mock import AsyncMock, MagicMock++import pytest++from skyvern.webeye.actions.handler import _save_adopted_session_download++PDF_BODY = b"%PDF-1.4\n" + b"x" * 830+++def _download(suggested: str = "153743777.pdf", url: str = "https://example.com/download") -> MagicMock:+ download = MagicMock()+ download.suggested_filename = suggested+ download.url = url+ download.save_as = AsyncMock()+ return download+++def _page_with_refetch(status: int = 200, body: bytes = PDF_BODY) -> MagicMock:+ response = MagicMock()+ response.status = status+ response.body = AsyncMock(return_value=body)+ page = MagicMock()+ page.context.request.get = AsyncMock(return_value=response)+ return page+++async def test_happy_path_eager_save_writes_bytes(tmp_path) -> None:+ download = _download()++ async def _save(target: object) -> None:+ Path(str(target)).write_bytes(PDF_BODY)++ download.save_as.side_effect = _save+ page = _page_with_refetch()++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is not None and saved.exists()+ assert saved.read_bytes() == PDF_BODY+ page.context.request.get.assert_not_awaited()+++async def test_save_as_raises_target_closed_falls_back_to_refetch(tmp_path) -> None:+ download = _download()+ download.save_as.side_effect = Exception("Target page, context or browser has been closed")+ page = _page_with_refetch()++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is not None and saved.exists()+ assert saved.read_bytes() == PDF_BODY+ page.context.request.get.assert_awaited_once_with(download.url)+++async def test_zero_byte_save_as_falls_back_to_refetch(tmp_path) -> None:+ download = _download()++ async def _save_empty(target: object) -> None:+ Path(str(target)).write_bytes(b"")++ download.save_as.side_effect = _save_empty+ page = _page_with_refetch()++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is not None and saved.exists()+ assert saved.read_bytes() == PDF_BODY+ page.context.request.get.assert_awaited_once_with(download.url)+ # the empty placeholder must not survive alongside the recovered file+ assert sorted(p.name for p in tmp_path.iterdir()) == [saved.name]+++async def test_refetch_non_200_returns_none(tmp_path) -> None:+ download = _download()+ download.save_as.side_effect = Exception("closed")+ page = _page_with_refetch(status=403, body=b"forbidden")++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is None+ assert list(tmp_path.iterdir()) == []+++async def test_refetch_empty_body_returns_none(tmp_path) -> None:+ download = _download()+ download.save_as.side_effect = Exception("closed")+ page = _page_with_refetch(status=200, body=b"")++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is None+ assert list(tmp_path.iterdir()) == []+++async def test_refetch_raises_returns_none(tmp_path) -> None:+ download = _download()+ download.save_as.side_effect = Exception("closed")+ page = MagicMock()+ page.context.request.get = AsyncMock(side_effect=Exception("connection gone"))++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is None+ assert list(tmp_path.iterdir()) == []+++async def test_partial_save_as_then_failed_refetch_leaves_no_orphan(tmp_path) -> None:+ """A partial (non-empty) write followed by a save_as raise must not orphan a corrupt file+ when the subsequent re-fetch also fails."""+ download = _download()++ async def _save_partial_then_raise(target: object) -> None:+ Path(str(target)).write_bytes(b"%PDF-1.4 truncated")+ raise Exception("Target page, context or browser has been closed")++ download.save_as.side_effect = _save_partial_then_raise+ page = MagicMock()+ page.context.request.get = AsyncMock(side_effect=Exception("connection gone"))++ saved = await _save_adopted_session_download(download, page, tmp_path, workflow_run_id="wr")++ assert saved is None+ assert list(tmp_path.iterdir()) == []diff --git a/tests/unit/test_download_file_action_handler.py b/tests/unit/test_download_file_action_handler.pyindex 85c1af812..ff9175475 100644--- a/tests/unit/test_download_file_action_handler.py+++ b/tests/unit/test_download_file_action_handler.py@@ -3,6 +3,7 @@ import tempfileimport timefrom collections.abc import Callablefrom datetime import UTC, datetime+from pathlib import Pathfrom unittest.mock import AsyncMock, MagicMock, patchimport pytest@@ -1654,3 +1655,206 @@ async def test_handle_action_logs_warning_when_late_native_appears_after_xhr_falassert kwargs["xhr_fallback_files"] == ["report.zip"]assert kwargs["post_settle_extra_file_count"] == 1assert kwargs["post_settle_extra_files"] == ["native-late.zip"]+++async def test_handle_action_adopted_session_lands_download_via_eager_save(+ span_exporter: InMemorySpanExporter,+) -> None:+ """On an adopted persistent session the run connection saves the download eagerly at event+ time and the bytes land in the run dir."""+ now = datetime.now(UTC)+ organization = make_organization(now)+ task = make_task(+ now,+ organization,+ workflow_run_id="wr-1",+ browser_session_id="bs-1",+ download_timeout=30.0,+ )+ step = make_step(now, task, step_id="step-1", status=StepStatus.created, order=0, output=None)++ page = MagicMock()+ page.url = "https://example.com/download"+ page.context.browser = None+ page.evaluate = AsyncMock()+ page.expose_binding = AsyncMock()+ download_callbacks: dict[str, Callable[[object], None]] = {}+ page.on.side_effect = lambda event, callback: download_callbacks.__setitem__(event, callback)++ browser_state = MagicMock()+ browser_state.list_valid_pages = AsyncMock(return_value=[page])++ scraped_page = ScrapedPage(+ elements=[],+ element_tree=[],+ element_tree_trimmed=[],+ _browser_state=browser_state,+ _clean_up_func=AsyncMock(return_value=[]),+ _scrape_exclude=None,+ )++ action = ClickAction(+ element_id="download-link",+ download=True,+ organization_id=task.organization_id,+ task_id=task.task_id,+ step_id=step.step_id,+ )++ download = MagicMock()+ download.suggested_filename = "report.pdf"+ download.url = "https://example.com/presigned/report.pdf"++ async def save_download(target_path: str | os.PathLike[str]) -> None:+ with open(target_path, "wb") as f:+ f.write(b"%PDF-1.4 report bytes")++ download.save_as = AsyncMock(side_effect=save_download)++ async def mock_inner_handle_action(*args: object, **kwargs: object) -> list[ActionSuccess]:+ download_callbacks["download"](download)+ return [ActionSuccess()]++ with tempfile.TemporaryDirectory() as temp_root:+ primary_dir = os.path.join(temp_root, "bs-1")+ os.makedirs(primary_dir)++ mock_app = MagicMock()+ mock_app.BROWSER_MANAGER.get_for_task.return_value = browser_state+ mock_app.DATABASE.workflow_params.create_action = AsyncMock(return_value=action)+ mock_app.STORAGE.list_downloaded_files_in_browser_session = AsyncMock(return_value=[])+ wait_for_downloads = AsyncMock()++ with (+ patch.object(ActionHandler, "_handle_action", side_effect=mock_inner_handle_action),+ patch("skyvern.webeye.actions.handler.get_download_dir", return_value=primary_dir),+ patch("skyvern.webeye.actions.handler.skyvern_context.current", return_value=MagicMock(run_id="bs-1")),+ patch(+ "skyvern.webeye.actions.handler.check_downloading_files_and_wait_for_download_to_complete",+ new=wait_for_downloads,+ ),+ patch("skyvern.webeye.actions.handler.app", mock_app),+ ):+ results = await ActionHandler.handle_action(+ scraped_page=scraped_page,+ task=task,+ step=step,+ page=page,+ action=action,+ )++ landed_files = sorted(os.listdir(primary_dir))++ assert results[-1].download_triggered is True+ assert action.download_triggered is True+ assert len(landed_files) == 1 and landed_files[0].endswith("-report.pdf")+ assert results[-1].downloaded_files == landed_files+ download.save_as.assert_awaited_once()+ # eager save wins; the url re-fetch is not needed when save_as succeeds+ page.context.request.get.assert_not_called()+ span_attrs = _download_wait_span_attrs(span_exporter)+ assert span_attrs["download_event_fallback_used"] is True+++async def test_handle_action_adopted_session_refetches_when_save_as_target_closed(+ span_exporter: InMemorySpanExporter,+) -> None:+ """When the worker tears the shared browser down and the run's save_as raises while the run+ request context is still alive, the adopted-session path re-fetches the replayable url and the+ bytes still land in the run dir."""+ now = datetime.now(UTC)+ organization = make_organization(now)+ task = make_task(+ now,+ organization,+ workflow_run_id="wr-1",+ browser_session_id="bs-1",+ download_timeout=30.0,+ )+ step = make_step(now, task, step_id="step-1", status=StepStatus.created, order=0, output=None)++ page = MagicMock()+ page.url = "https://example.com/download"+ page.context.browser = None+ page.evaluate = AsyncMock()+ page.expose_binding = AsyncMock()+ download_callbacks: dict[str, Callable[[object], None]] = {}+ page.on.side_effect = lambda event, callback: download_callbacks.__setitem__(event, callback)++ refetched_bytes = b"%PDF-1.4 refetched report bytes"+ refetch_response = MagicMock()+ refetch_response.status = 200+ refetch_response.body = AsyncMock(return_value=refetched_bytes)+ page.context.request.get = AsyncMock(return_value=refetch_response)++ browser_state = MagicMock()+ browser_state.list_valid_pages = AsyncMock(return_value=[page])++ scraped_page = ScrapedPage(+ elements=[],+ element_tree=[],+ element_tree_trimmed=[],+ _browser_state=browser_state,+ _clean_up_func=AsyncMock(return_value=[]),+ _scrape_exclude=None,+ )++ action = ClickAction(+ element_id="download-link",+ download=True,+ organization_id=task.organization_id,+ task_id=task.task_id,+ step_id=step.step_id,+ )++ download = MagicMock()+ download.suggested_filename = "report.pdf"+ download.url = "https://example.com/presigned/report.pdf"+ download.save_as = AsyncMock(side_effect=Exception("Target page, context or browser has been closed"))++ async def mock_inner_handle_action(*args: object, **kwargs: object) -> list[ActionSuccess]:+ download_callbacks["download"](download)+ return [ActionSuccess()]++ with tempfile.TemporaryDirectory() as temp_root:+ primary_dir = os.path.join(temp_root, "bs-1")+ os.makedirs(primary_dir)++ mock_app = MagicMock()+ mock_app.BROWSER_MANAGER.get_for_task.return_value = browser_state+ mock_app.DATABASE.workflow_params.create_action = AsyncMock(return_value=action)+ mock_app.STORAGE.list_downloaded_files_in_browser_session = AsyncMock(return_value=[])+ wait_for_downloads = AsyncMock()++ with (+ patch.object(ActionHandler, "_handle_action", side_effect=mock_inner_handle_action),+ patch("skyvern.webeye.actions.handler.get_download_dir", return_value=primary_dir),+ patch("skyvern.webeye.actions.handler.skyvern_context.current", return_value=MagicMock(run_id="bs-1")),+ patch(+ "skyvern.webeye.actions.handler.check_downloading_files_and_wait_for_download_to_complete",+ new=wait_for_downloads,+ ),+ patch("skyvern.webeye.actions.handler.app", mock_app),+ ):+ results = await ActionHandler.handle_action(+ scraped_page=scraped_page,+ task=task,+ step=step,+ page=page,+ action=action,+ )++ landed = sorted(os.listdir(primary_dir))+ landed_bytes = Path(primary_dir, landed[0]).read_bytes() if landed else b""++ assert results[-1].download_triggered is True+ assert action.download_triggered is True+ assert len(landed) == 1 and landed[0].endswith("-report.pdf")+ assert landed_bytes == refetched_bytes+ assert results[-1].downloaded_files == landed+ download.save_as.assert_awaited_once()+ page.context.request.get.assert_awaited_once_with(download.url)+ span_attrs = _download_wait_span_attrs(span_exporter)+ assert span_attrs["download_event_fallback_used"] is Truediff --git a/tests/unit/test_proxy_url_custom.py b/tests/unit/test_proxy_url_custom.pyindex fd32b0ae8..50e16f8ab 100644--- a/tests/unit/test_proxy_url_custom.py+++ b/tests/unit/test_proxy_url_custom.py@@ -3,7 +3,23 @@ import pytestfrom skyvern.config import settingsfrom skyvern.forge.sdk.db.utils import deserialize_proxy_location, serialize_proxy_locationfrom skyvern.schemas.runs import GeoTarget, ProxyLocation-from skyvern.webeye.browser_factory import BrowserContextFactory, _redact_proxy_url+from skyvern.webeye.browser_factory import BrowserContextFactory, _redact_proxy_url, _redact_url_query+++def test_redact_url_query_strips_presigned_signature() -> None:+ redacted = _redact_url_query(+ "https://bucket.s3.amazonaws.com/docs/report?X-Amz-Signature=deadbeef&X-Amz-Credential=AKIA"+ )+ assert redacted == "https://bucket.s3.amazonaws.com/docs/report"+ assert "X-Amz-Signature" not in redacted+++def test_redact_url_query_keeps_url_without_query() -> None:+ assert _redact_url_query("https://example.com/docs/report.pdf") == "https://example.com/docs/report.pdf"+++def test_redact_url_query_plain_string_passes_through() -> None:+ assert _redact_url_query("not-a-url") == "not-a-url"def test_redact_proxy_url_strips_password() -> None: