browser sessions v2 - backend (#4515)

Signed-off-by: Benji Visser <benji@093b.org>
This commit is contained in:
Benji Visser
2026-01-21 22:27:16 -05:00
committed by GitHub
parent f781a6f0ef
commit b5ff547a3a
15 changed files with 273 additions and 35 deletions

View File

@@ -281,6 +281,43 @@ async def ask_for_clipboard(vnc_channel: VncChannel) -> None:
LOG.exception(f"{class_name} Failed to ask for clipboard via CDP", **vnc_channel.identity)
# TODO(benji): I hate this function. It's messy and gross. Once we remove v1,
# we should clean this up.
def _build_vnc_url_from_browser_address(browser_address: str) -> str | None:
"""
Build a routed VNC URL from a V2 K8s routed browser_address.
V2 K8s routed browser_address format:
wss://{domain}/{session_id}/{token}/devtools/browser/{browser_id}
Returns VNC URL in format:
wss://{domain}/vnc/{session_id}/{token}
Returns None if browser_address is not a V2 routed URL.
"""
if not browser_address:
return None
parsed = urlparse(browser_address)
# Check if this looks like a V2 routed URL (wss:// with token in path)
if parsed.scheme not in ("wss", "ws"):
return None
# Parse path: /{session_id}/{token}/devtools/browser/{browser_id}
path_parts = parsed.path.strip("/").split("/")
if len(path_parts) < 4 or path_parts[2] != "devtools":
return None
session_id = path_parts[0]
token = path_parts[1]
domain = parsed.netloc
# Build VNC URL with same domain and token
scheme = "wss" if parsed.scheme == "wss" else "ws"
return f"{scheme}://{domain}/vnc/{session_id}/{token}"
async def loop_stream_vnc(vnc_channel: VncChannel) -> None:
"""
Actually stream the VNC data between a frontend and a browser.
@@ -292,24 +329,28 @@ async def loop_stream_vnc(vnc_channel: VncChannel) -> None:
browser_session = vnc_channel.browser_session
class_name = vnc_channel.class_name
if browser_session:
if browser_session.ip_address:
if ":" in browser_session.ip_address:
ip, _ = browser_session.ip_address.split(":")
vnc_url = f"ws://{ip}:{vnc_channel.vnc_port}"
else:
vnc_url = f"ws://{browser_session.ip_address}:{vnc_channel.vnc_port}"
else:
browser_address = browser_session.browser_address
parsed_browser_address = urlparse(browser_address)
host = parsed_browser_address.hostname
vnc_url = f"ws://{host}:{vnc_channel.vnc_port}"
else:
if not browser_session:
raise Exception(f"{class_name} No browser session associated with vnc channel.")
# NOTE(jdo:streaming-local-dev)
# vnc_url = "ws://localhost:6080"
# First, check if this is a V2 K8s routed session by examining browser_address
# V2 sessions have browser_address like: wss://{domain}/{session_id}/{token}/devtools/...
# For these, we need to route VNC through the same nginx proxy
routed_vnc_url = _build_vnc_url_from_browser_address(browser_session.browser_address)
if routed_vnc_url:
vnc_url = routed_vnc_url
elif browser_session.ip_address:
# V1 ECS sessions: Direct IP connection (ip_address is a public/reachable IP)
if ":" in browser_session.ip_address:
ip, _ = browser_session.ip_address.split(":")
vnc_url = f"ws://{ip}:{vnc_channel.vnc_port}"
else:
vnc_url = f"ws://{browser_session.ip_address}:{vnc_channel.vnc_port}"
else:
# Last resort: parse browser_address hostname
browser_address = browser_session.browser_address
parsed_browser_address = urlparse(browser_address)
host = parsed_browser_address.hostname
vnc_url = f"ws://{host}:{vnc_channel.vnc_port}"
LOG.info(
f"{class_name} Connecting to vnc url.",
@@ -317,7 +358,12 @@ async def loop_stream_vnc(vnc_channel: VncChannel) -> None:
**vnc_channel.identity,
)
async with websockets.connect(vnc_url) as novnc_ws:
# For routed VNC URLs (wss://), we need to pass the x-api-key header for authentication
extra_headers: dict[str, str] = {}
if vnc_url.startswith("wss://") and vnc_channel.x_api_key:
extra_headers["x-api-key"] = vnc_channel.x_api_key
async with websockets.connect(vnc_url, additional_headers=extra_headers) as novnc_ws:
async def frontend_to_browser() -> None:
nonlocal class_name

View File

@@ -5,8 +5,10 @@ Provides WS endpoints for streaming messages to/from our frontend application.
import structlog
from fastapi import WebSocket
from skyvern.config import settings
from skyvern.forge.sdk.routes.routers import base_router, legacy_base_router
from skyvern.forge.sdk.routes.streaming.auth import auth
from skyvern.forge.sdk.routes.streaming.auth import _auth as local_auth
from skyvern.forge.sdk.routes.streaming.auth import auth as real_auth
from skyvern.forge.sdk.routes.streaming.channels.message import (
Loops,
MessageChannel,
@@ -60,6 +62,7 @@ async def messages(
client_id: str | None = None,
token: str | None = None,
) -> None:
auth = local_auth if settings.ENV == "local" else real_auth
organization_id = await auth(apikey=apikey, token=token, websocket=websocket)
if not organization_id:
@@ -98,7 +101,7 @@ async def messages(
)
else:
LOG.error(
"Message channel: no browser_session_id or workflow_run_id provided.",
"[WS] messages: no browser_session_id or workflow_run_id provided",
client_id=client_id,
organization_id=organization_id,
)
@@ -139,5 +142,4 @@ async def messages(
workflow_run_id=workflow_run_id,
organization_id=organization_id,
)
await message_channel.close(reason="message-stream-closed")

View File

@@ -45,12 +45,12 @@ async def verify_browser_session(
"""
Verify the browser session exists, and is usable.
"""
if settings.ENV == "local":
dummy_browser_session = AddressablePersistentBrowserSession(
persistent_browser_session_id=browser_session_id,
organization_id=organization_id,
browser_address="0.0.0.0:9223",
ip_address="localhost",
created_at=datetime.now(),
modified_at=datetime.now(),
)
@@ -199,6 +199,7 @@ async def verify_workflow_run(
persistent_browser_session_id=workflow_run_id,
organization_id=organization_id,
browser_address="0.0.0.0:9223",
ip_address="localhost",
created_at=datetime.now(),
modified_at=datetime.now(),
)

View File

@@ -12,8 +12,10 @@ NOTE(jdo:streaming-local-dev)
import structlog
from fastapi import WebSocket
from skyvern.config import settings
from skyvern.forge.sdk.routes.routers import base_router, legacy_base_router
from skyvern.forge.sdk.routes.streaming.auth import auth
from skyvern.forge.sdk.routes.streaming.auth import _auth as local_auth
from skyvern.forge.sdk.routes.streaming.auth import auth as real_auth
from skyvern.forge.sdk.routes.streaming.channels.vnc import (
Loops,
VncChannel,
@@ -86,6 +88,7 @@ async def stream(
workflow_run_id=workflow_run_id,
)
auth = local_auth if settings.ENV == "local" else real_auth
organization_id = await auth(apikey=apikey, token=token, websocket=websocket)
if not organization_id: