Waits for textures to be loaded for Isaac RTX renderer#4956
Waits for textures to be loaded for Isaac RTX renderer#4956bareya wants to merge 11 commits intoisaac-sim:developfrom
Conversation
bareya
changed the title
Greptile SummaryThis PR replaces a long-standing workaround (running 5 fixed simulation steps to allow textures to load) with a proper event-driven mechanism that subscribes to the Carb Key changes:
Issues found:
Confidence Score: 3/5
Important Files Changed
Sequence DiagramsequenceDiagram
participant Cam as Camera/TiledCamera
participant Utils as isaac_rtx_renderer_utils
participant Carb as carb.eventdispatcher
participant App as omni.kit.app
Cam->>Utils: ensure_isaac_rtx_render_update()
Utils->>Utils: check dedup key (id(sim), step_count)
Utils->>Carb: _ensure_streaming_subscription() [idempotent]
Carb-->>Utils: subscription handle
Utils->>App: app.update() [trigger RTX render]
App-->>Carb: fires streaming_status(isBusy=True)
Carb-->>Utils: _on_streaming_status_event → _streaming_is_busy=True
Utils->>Utils: if _streaming_is_busy → _wait_for_streaming_complete()
loop Until idle or 30s timeout
Utils->>App: app.update()
App-->>Carb: fires streaming_status(isBusy=False)
Carb-->>Utils: _on_streaming_status_event → _streaming_is_busy=False
end
Utils->>Utils: update _last_render_update_key
Utils-->>Cam: return (textures ready)
Last reviewed commit: fc40398 |
![greptile-apps[bot]](https://avatars.githubusercontent.com/in/867647?s=60&v=4){kind=small}
| _streaming_is_busy: bool = False | ||
| _streaming_subscription = None | ||
| _streaming_subscribed: bool = False |
There was a problem hiding this comment.
Stale streaming state across SimulationContext instances
_streaming_is_busy and _streaming_subscribed are module-level globals that are never reset when a new SimulationContext is created. The render-update dedup key (_last_render_update_key) correctly invalidates itself per-context via id(sim), but the streaming state has no equivalent invalidation.
The problematic scenario: if a previous simulation ends while _streaming_is_busy = True (i.e., the 30 s timeout was hit), the next test's first call to ensure_isaac_rtx_render_update() will skip _ensure_streaming_subscription() (already subscribed) then immediately enter _wait_for_streaming_complete() because the stale True flag is never cleared between contexts. This could add an unnecessary second full timeout wait at the start of every subsequent test in a test suite.
Consider resetting the streaming state alongside the dedup key when a new SimulationContext is detected:
# at the top of ensure_isaac_rtx_render_update(), after computing `key`
if key[0] != _last_render_update_key[0]: # new SimulationContext
global _streaming_is_busy, _streaming_subscribed, _streaming_subscription
_streaming_is_busy = False
_streaming_subscribed = False
_streaming_subscription = NoneThere was a problem hiding this comment.
Good point! Thanks AI!
| def _ensure_streaming_subscription() -> None: | ||
| """Subscribe to RTX streaming status events (idempotent).""" | ||
| global _streaming_subscription, _streaming_subscribed | ||
| if _streaming_subscribed: | ||
| return | ||
|
|
||
| _streaming_subscribed = True | ||
| try: | ||
| from carb.eventdispatcher import get_eventdispatcher | ||
|
|
||
| dispatcher = get_eventdispatcher() | ||
| if dispatcher is not None: | ||
| _streaming_subscription = dispatcher.observe_event( | ||
| observer_name="isaaclab_rtx_streaming_wait", | ||
| event_name=_RTX_STREAMING_STATUS_EVENT, | ||
| on_event=_on_streaming_status_event, | ||
| ) | ||
| except (ImportError, AttributeError): | ||
| pass |
There was a problem hiding this comment.
_streaming_subscribed guard set before subscription succeeds
_streaming_subscribed = True is written on line 52 before the try block verifies that a dispatcher is available and the subscription actually succeeds. If get_eventdispatcher() returns None (or raises and is silently caught), _streaming_subscribed is permanently True, meaning no retry will ever be attempted — even though _streaming_subscription remains None and no events will ever arrive.
The result is that the texture-wait feature silently degrades to a no-op without any log message, making it hard to diagnose why textures are still black in environments where the event dispatcher is slow to initialise.
Consider either:
- Moving
_streaming_subscribed = Trueto after a successfulobserve_event()call, or - Logging a warning when the dispatcher is unavailable.
| def _wait_for_streaming_complete() -> None: | ||
| """Pump ``app.update()`` until RTX streaming reports idle or timeout.""" | ||
| import omni.kit.app | ||
|
|
||
| start = time.monotonic() | ||
| while _streaming_is_busy and (time.monotonic() - start) < _STREAMING_WAIT_TIMEOUT_S: | ||
| omni.kit.app.get_app().update() |
There was a problem hiding this comment.
No rate-limiting in the busy-wait loop
_wait_for_streaming_complete() calls omni.kit.app.get_app().update() in a tight loop with no time.sleep() between iterations. While each app.update() call does real work, it can return quickly in headless/test configurations, turning this into a CPU-intensive spin for up to 30 seconds in the timeout scenario.
Consider adding a small sleep to yield the CPU between pumps:
while _streaming_is_busy and (time.monotonic() - start) < _STREAMING_WAIT_TIMEOUT_S:
omni.kit.app.get_app().update()
time.sleep(0.005) # ~200 Hz max pump rate
There was a problem hiding this comment.
i don't think we need to add overhead of sleep here, even small. if something is timing out no need to optimize the exit IMO
nvsekkin
force-pushed
the
pbarejko/rtx-renderer-load-textures
branch
from
5b51550 to
8704770
Compare
kellyguo11
changed the title
3 participants
Wait for all textures to load in IsaacRTX Renderer for Camera and Tiled Camera
Please include a summary of the change and which issue is fixed. Please also include relevant motivation and context.
List any dependencies that are required for this change.
Fixes # (issue)
Type of change
Screenshots
Please attach before and after screenshots of the change if applicable.
Checklist
pre-commitchecks with./isaaclab.sh --formatconfig/extension.tomlfileCONTRIBUTORS.mdor my name already exists there