Hey, I am running a gRPC server using Rust SDK's

serve_grpc_opts

function. I pass

MemoryLimit::parse("2GB")

(also tried with

::from_bytes

and the byte count), but my rerun server process goes wild with memory usage. It basically continues to accumulate memory until the OS kills it or hangs (see screenshot it's already at around 25Gbyte). I run it in a Docker container on Ubuntu, if that matters. The web-viewer recording itself hovers around ~2Gbyte, so that is good. But if the server runs OOM that doesn't help much. Is there something I misunderstand or am missing about the configurable memory limits? On code level, this is all I am doing:

let rec = RecordingStreamBuilder::new("test")
        .recording_id("test-recording")
        .serve_grpc_opts("0.0.0.0", 9876, MemoryLimit::parse("2GB")?)?;

loop {
    std::thread::park();
}

https://cdn.discordapp.com/attachments/1408534249694302229/1408534249858007162/CleanShot_2025-08-22_at_12.25.53.png?ex=68aa173e&is=68a8c5be&hm=4d3f24a0470516d847807d07eab5868c7d21a2a8be971f681b61543a5d73c42b& https://cdn.discordapp.com/attachments/1408534249694302229/1408534250340483142/CleanShot_2025-08-22_at_12.26.022x.png?ex=68aa173e&is=68a8c5be&hm=a266a0240ef24cea32cad8e490a29efd4226c8dd394920136df51301c65c7f83&

I'm trying to create a Transform3D with xyz and rpy (roll, pitch, yaw) data, but I can't find any mention of euler angles in any of the rerun docs. Is there really no easy way to create a Rotation3D object out of euler angles using the rerun API?

Hi, Rerun Team! Firstly, thanks for building awesome software:) In our project, we want to switch from using ROS and rviz to Rerun. Could you please tell if Rerun has an equivalent to ROS topics that can be used as a communication bus between threads?

I know it can be set from the GUI clicking through settings, but it would be nice if we could make this the default for our app so we don't need every user to change this setting on their own.

I'm using v0.24.1 with the web viewer.

LinkAxis.Independent

works well, but

LinkAxis.LinkGlobal

doesn't log any data. Attached a screenshot for Independent vs Global link. Is there anything wrong with the way I am logging the data? Here's how I have the data logged: For each entity: - Fetch timestamps and values using fetch_data(entity_path, start_time, end_time) - Send data to Rerun:

rr.send_columns(
            entity_path,
            indexes=[rr.TimeColumn("time", timestamp=timestamps)],
            columns=rr.Scalars.columns(scalars=values),
        )

Build a list of views: - For each panel:

rrb.TimeSeriesView(
            origin="/",
            name=panel["name"],
            contents=entity_paths,
            axis_x=rrb.archetypes.TimeAxis(
                link=rrb.components.LinkAxis.LinkGlobal
            ),
        )

Arrange views into a grid layout:

grid = rrb.Grid(*views, grid_columns=1)

Send the blueprint to Rerun:

rr.send_blueprint(
        rrb.Blueprint(
            grid,
            collapse_panels=True,
        ),
        make_active=True,
    )

https://cdn.discordapp.com/attachments/1405995563413143694/1405995564037968035/Screenshot_2025-08-15_at_3.22.02_PM.png?ex=68a0dae8&is=689f8968&hm=164bd9c87ac6d25e2db3d7de055eab4b591ce43b43f1429e9c3f6b466ba0ff8f& https://cdn.discordapp.com/attachments/1405995563413143694/1405995564444946473/Screenshot_2025-08-15_at_3.24.03_PM.png?ex=68a0dae8&is=689f8968&hm=89724618fee1ed6db1659b575c4df51270ba77b34865be98b282ae792325e7ae&

Is there a tutorial for logging with rerun on a remote machine, which I am SSHd to with port forwarding, and viewing the data on my local machine? There are lots of random questions about it but I still can't make it work. Would appreciate a clean working example hosted on the rerun docs! Right now I reverted to just trying to get the web viewer to work locally (no remote) with the code provided on the website

"""Demonstrates how to log data to a gRPC server and connect the web viewer to it."""

import time

import rerun as rr

rr.init("rerun_example_serve_web_viewer")
# Start a gRPC server and use it as log sink.
server_uri = rr.serve_grpc()

# Connect the web viewer to the gRPC server and open it in the browser
rr.serve_web_viewer(connect_to=server_uri)

# Log some data to the gRPC server.
rr.log("data", rr.Boxes3D(half_sizes=[2.0, 2.0, 1.0]))

# Keep server running. If we cancel it too early, data may never arrive in the browser.
try:
    while True:
        time.sleep(1)
except KeyboardInterrupt:
    print("\nShutting down server...")

However no data shows up in the web viewer and it gives me the error "Data source rerun+http://127.0.0.1:9876/proxy has left unexpectedly: gRPC error, message: "js api error: TypeError: NetworkError when attempting to fetch resource.""

Hi Rerun team Firstly, thanks for building awesome software! I'm logging three camera feeds (640x480, RGB+depth) and 13 piper robot proprioception fields per step to rerun. I'm running at 20Hz, so I have a budget of 50ms for each step. The below logging code in rerun takes around 30ms, sometimes jumping above 50ms, and is reducing my FPS. Am I doing anything naive here, and is there a way to speed up the

rr.log

calls below? Thanks for your help! Tim

python

def _log_to_rerun(payload: tuple) -> None:

  rec, seq, rgb_items, depth_items, proprio_items = payload

  rr.set_time(
      timeline="episode_step",
      sequence=seq,
      recording=rec,
  )

  for path, frame in rgb_items:
    rr.log(
        path,
        rr.Image(frame),
        recording=rec,
    )

  for path, frame in depth_items:
    rr.log(
        path,
        rr.DepthImage(frame, meter=1000.0, depth_range=(0.0, 1500.0)),
        recording=rec,
    )

  for prop_key, proprio_frame in proprio_items:
    for dim in range(proprio_frame.shape[-1]):
      rr.log(
          f"{prop_key}/d{dim}",
          rr.Scalars(proprio_frame[..., dim]),
          recording=rec,
      )

Using a windows machine to remote connect to a jetson board, worked normal when openning the web-viewer page on browser once forwarded the port 9090, but could not get any response but the welcome page. Codes below:

import rerun as rr

rr.init("test_app", spawn=False)

# 1. Start the gRPC and web servers
grpc_uri = rr.serve_grpc()
rr.serve_web_viewer(connect_to=grpc_uri)

# 2. Log the data (this adds it to a buffer)
rr.log("my_points", rr.Points2D([[0, 0], [1, 1], [2, 2]]))

What did I miss here?

Hi all, I'm trying to stream from my Pi 5 to a web viewer that I can view on another machine. Things work okay, except the only way I can connect to the viewer and see data from my remote machine is if I use the following URL exactly:

http://<SERVER_IP>:9090/?url=rerun%2Bhttp%3A%2F%2F<SERVER_IP>%3A9876%2Fproxy

. The only way I found this URL is from the CLI when I used

--serve-web

. Using

http://<SERVER_IP>:9090/?url=rerun+http://<SERVER_IP>:9876/proxy

works in a browser on the pi using

localhost

but I can't seem to get it to work in a browser on my remote machine without using the URL above. Is there a more generic URL I can use? I've been looking through the docs with no success. Thanks in advance!

Hi All, i am using the visualiser with both Python and C++ SDK, using pixi. => rerun-sdk = ">=0.23.4,<0.24" The python logger works just fine to the visualisation, but the C++ code gives the following error in the visualiser: "dropping LogMsg due to failed decode: Failed to read chunk: Bad chunk schema: Missing row_id column". Maybe we are doing something wrong, the 0.24 release is not yet released for pixi so i can't just upgrade. C++ snipet:

// Create the RecordingStream for rerun.
recording_stream_ = std::make_shared<rerun::RecordingStream>("DataProcessor");

// Spawn a viewer.
recording_stream_->spawn().exit_on_failure();
recording_stream_->log("Input Image",

rerun::Image(rerun::borrow(cv_input_rgb.data, cv_input_rgb.total() * cv_input_rgb.elemSize()),
  rerun::WidthHeight(static_cast<uint32_t>(cv_input_rgb.cols),
  static_cast<uint32_t>(cv_input_rgb.rows)),
  rerun::ColorModel::RGB));

.

I'm running hard up against the gRPC broadcast receiver lagging behind until it errors out. Wondering if I'm holding it wrong or something of that nature

2025-07-31T08:16:52.893796772Z ERROR re_grpc_server: Error reading message from broadcast channel: channel lagged by 281974

Minimal repro: https://gist.github.com/hansihe/522103ad9c5d4e55e9bcfd99213f5b80 I can push a full crate to github too if that makes things easier to dig into, just let me know. I am attempting to use

re_renderer

directly within a custom

egui

application (using

wgpu

). With nothing queued in

ViewBuilder

I get nothing and see the background as expected. When I

queue_draw

a

GenericSkyboxDrawData

, I get a fully black viewport. Behavior of other renderers varies, but I never get any meaningful output. I captured a frame in RenderDoc to try to get more insight into what might be happening. The black screen seems to stem from the color value in the shader `nan`ing. Tracing it back in the debugger it seems to originate from the

re_renderer

frame uniform buffer

which is bound to descriptor set 0 being filled with nulls only. According to RenderDoc the buffer is only `clear`ed and barriered before it is used in the shaders for the

re_renderer

renderers. I imagine I am probably missing some important step in my code, does someone know what I might be doing wrong?

I am receiving a h264 video stream from a UDP socket and I would like to log it with the new VideoStream in C++. More specifically, I am receiving UDP packets (of length 1316 or a smaller multiple of 188) and send them to rerun with:

rec.log("video_stream",rerun::archetypes::VideoStream{}.with_sample(rerun::components::VideoSample{data}));

I had a first error similar to the one mentioned here: https://discord.com/channels/1062300748202921994/1397335975364919392 I applied the

uint32_t

fix and the next error was the ffmpeg version. Got the recommended version and the video stream doesn't get displayed:

Failed to decode video: FFmpeg error: [h264] [error] error while decoding MB 2 64, bytestream 16292

It gets the Dimensions of the video stream right (1920x1080) and the sent data chunks appear in the Video samples table but no video gets displayed. The video stream is created from that command:

ffmpeg -f rawvideo -pixel_format rgba -video_size {width}x{height} -framerate {frameRate} -i - -vf vflip -c:v libx264 -preset veryfast -tune zerolatency -f mpegts udp://localhost:1234?pkt_size=1316

and it is displayed correctly with

ffplay

.

Hello, I want to log data from two different sources (a python script and a c++ executable) in the same rerun recording. I would like the data to be saved to file as I log it with rerun (i.e. not all al the end) and also to be able to visualize it live in a single viewer. Is it possible with rerun? How would I implement this?

Hello, I have a list of joints from a robot encoder and I want to log them with rerun. Is it possible to log all of them in one call without looping on the list and call rr.log for each value?

I'm trying to render a surface patch generated from HiRISE DTM and orthoimages (e.g. https://www.uahirise.org/dtm/dtm.php?ID=PSP_002387_1985). I used the HiRISE DTM Importer blender plugin (https://hemelmechanica.nl/hirise-docs/quickstart.html) to generate the mesh from the DTM image and texture it use a matching grayscale orthoimage. Everything looks correct in Blender and also in https://gltf.report/ and https://gltf-viewer.donmccurdy.com/. However, in rerun, the model looks bright red as if its texture has the greyscale image in the R channel but G and B are just set to 0. Changing AlbedoFactor to rgb(255,0,0) has no effect on how the surface patch looks. But changing it to rgb(0,255,0) or rgb(0,0,255) or rgb(0,255,255) just makes the entire surface patch pitch black. I'm not sure what I'm doing wrong since when I import other GLB files in rerun (e.g. a glider model I found and a free rubber ducky model I found to use as a test), they look normal. https://cdn.discordapp.com/attachments/1398520508022132866/1398520508907126846/Screenshot_2025-07-25_at_9.01.37_PM.png?ex=6885a938&is=688457b8&hm=2de6e6ff4fe3488eae2d1a41a87b80ce707f6683f76b8bb3dcbe70be27ebf2df& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509552918639/Screenshot_2025-07-25_at_9.02.07_PM.png?ex=6885a938&is=688457b8&hm=ce2b4e462c1542db2b5c763923f96d77cffd7df42356d0a8e433063d318009b0& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509964095608/Screenshot_2025-07-25_at_9.03.30_PM.png?ex=6885a938&is=688457b8&hm=9c3609eac08dcf3c2f3a5568d8f32793ff84ccc0ec46fde8cbf400ef2b67d991& https://cdn.discordapp.com/attachments/1398520508022132866/1398520510509350963/Screenshot_2025-07-25_at_9.15.54_PM.png?ex=6885a938&is=688457b8&hm=7c68c1b8c144dbef323ffce38a4a1c8a000a248cdeb7f24035c439baaea3187a&

Is there a way I can programmatically set the viewpoint for a Spatial3DView using the PythonAPI so that I don't have to manually pan/zoom/rotate a pointcloud with the mouse everytime to see the view I need?

Hi, I'm trying to get the VideoStream support working on C++. The example on python works correctly, but when trying to port this over to C++ I keep getting:

re_viewer_context::component_ui_registry] Failed to deserialize component of type rerun.components.VideoCodec: Failed to deserialize "rerun.components.VideoCodec"

I've put the RUST_LOG up to debug on both the sender and the receiver but it gives me no other information other than the deserialization error. Any idea what I might be doing wrong? I've attached the code I'm working with. This also dumps the h264 data to a file which does reproduce correctly when using ffplay so I don't think this is an encoding error https://cdn.discordapp.com/attachments/1397335975364919392/1397335975645941882/on-the-fly.cpp?ex=688b3d49&is=6889ebc9&hm=f06d6a86384a371ebdcb0a5c7b02a7556716b8bace7b699e20bee65ea0ef7161&

I tested this by creating an application in C++ using the API: - Reads a blueprint (.rbl file) - Creates a recording stream - Attempts to spawn the viewer - Logs some data. When I compile and run this application, it normally spawns the viewer and functions as expected. But when I run this same application as a root user (using sudo), it gets stuck at reading the blueprint stage itself, and cannot proceed forward, unable to spawn the Viewer. If in my application, a root access is required to be used for other purposes, are there any environment variables I need to pass so that the GUI (Rerun viewer) is able to spawn correctly? Supporting evidence is attached. https://cdn.discordapp.com/attachments/1397186188128817222/1397188114430038036/image.png?ex=6880d054&is=687f7ed4&hm=504cba18bd88a0bb8610bcda35b82d9d1ab99b0acc0ef1e55a0d6ef779b8ee93& https://cdn.discordapp.com/attachments/1397186188128817222/1397188114937544834/image.png?ex=6880d055&is=687f7ed5&hm=9142e882a49cf4ddbe87f89c795155c5dad7fb6053aad5e9ba9c0af738984b6b&

When we make the pointcloud panel vertically small (wide and short panel) the points at either end of the viewer become abnormally big. This goes away when the viewer's aspect ratio is more 1:1 (square). Please let me know if this is a bug or configuration issue? I'm using python rerun-sdk 0.24.0 You can reproduce this even on the rerun nuscenes example: https://rerun.io/viewer?url=https://app.rerun.io/version/0.24.0/examples/nuscenes_dataset.rrd All you'll have to do is vertically squish the lidar panel and zoom in so that there are points in the left and right edges of the lidar viewer. https://cdn.discordapp.com/attachments/1397008726090580112/1397008727021850736/image.png?ex=68802943&is=687ed7c3&hm=15c74c5fb58d00906ea87ae9690e4803b87552e1e5195678cdc72bd501ba5266& https://cdn.discordapp.com/attachments/1397008726090580112/1397008728321949756/image.png?ex=68802943&is=687ed7c3&hm=ab2e9ce469b04d766f5e39418c86389a13667c7cc527f5bfa747cf56a3302fd5&

I'm trying to write my own

Renderer

using the

re_renderer

crate and the shader I want to use requires some `texture2d`s. From looking through the Rerun code I can see examples like:

ctx.gpu_resources.bind_groups.alloc(
                &ctx.device,
                &ctx.gpu_resources,
                &BindGroupDesc {
                    label,
                    entries: smallvec![
                        BindGroupEntry::DefaultTextureView(position_data_texture.handle),
                        BindGroupEntry::DefaultTextureView(color_texture.handle),
                        BindGroupEntry::DefaultTextureView(picking_instance_id_texture.handle),
                        draw_data_uniform_buffer_binding,
                    ],
                    layout: point_renderer.bind_group_layout_all_points,
                },
            )

however,

BindGroupEntry

is private. For the variant

BindGroupEntry::Buffer

there is

create_and_fill_uniform_buffer

, but I can't find anything that lets me pass my own texture handles to shaders. Am I missing something?

After upgrading my rerun dependency from 0.23 -> 0.24, my project does not compile anymore because

ashpd

fails to compile. Here's a part of the errors from

cargo build

. Does anybody know how to solve this? I use rust 1.88 on Linux.

error[E0432]: unresolved import `futures_util::AsyncReadExt`
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/ashpd-0.10.2/src/desktop/secret.rs:26:5
    |
26  | use futures_util::AsyncReadExt;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ no `AsyncReadExt` in the root
    |
note: found an item that was configured out
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:320:47
    |
320 |     AsyncBufRead, AsyncBufReadExt, AsyncRead, AsyncReadExt, AsyncSeek, AsyncSeekExt, AsyncWrite,
    |                                               ^^^^^^^^^^^^
note: the item is gated behind the `io` feature
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:316:7
    |
316 | #[cfg(feature = "io")]
    |       ^^^^^^^^^^^^^^

error[E0432]: unresolved import `futures_util::AsyncReadExt`
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/ashpd-0.10.2/src/helpers.rs:4:5
    |
4   | use futures_util::AsyncReadExt;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ no `AsyncReadExt` in the root
    |
note: found an item that was configured out
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:320:47
    |
320 |     AsyncBufRead, AsyncBufReadExt, AsyncRead, AsyncReadExt, AsyncSeek, AsyncSeekExt, AsyncWrite,
    |                                               ^^^^^^^^^^^^
...

NOTE: I saw [this](https://users.rust-lang.org/t/issue-with-futures-lite-dependency-resolution-in-ashpd-read-to-string-read-to-end-errors/125669/1) already, but it didn't apply to me

I have some questions about Revy and Rerun. I use Revy, which I want to make some changes to it but I'm new to rerun, so I want to ask some questions about the rerun side. Maybe also for people who know rerun, like for a long time, the questions sound a little bit dumb, so sorry for that in advance. My first question is, I want to leverage the reflection system and the apptyperegistry like Bevy Inspector Egui does and want to create a general-purpose like rerun Rust type component that presents like an enum or struct, which is a vec of fields. Is it possible, like in bevy, that I can register the same type but with a different component ID? In rerun, I think it is called component name or component descriptor, I'm not sure right now. I saw that rerun 24 came out, which introduces tag components. Would that be better for structuring the data? Because now we could have like pure data type components (bool, ints and floats) and have the Rust type or Bevy components be an archetype with the corresponding name, and use the data type components to be a part of the archetype. So, an archetype would represent the Bevy component. Would that be better for querying, structuring and projecting the data? Why I am suggesting the idea is that how the structure of revy is implemented now, you would have to write logger for every single type you have or it just creates a text component with the Ron serialized data in it, which I think is not good for projecting data and querying, but I can also be corrected on that. And creating a logger function for every type you have is a little bit tedious because I create a tool, only a tool, and I have already like 200 to 300 types. If you create a real game with Bevy, you would like maybe have more than hundreds of types, 1000 to 2,000 types, because every single thing is a data type in Bevy. So, writing a logger function for every single type is not a small barrier to entering Revy for logging the data.

Does the log function block until the data is stored to file? I'm writing a control loop and don't want to block the loop if the log function is taking a long time to run. If it doesn't do this already, I can use a crossbeam channel to send messages from my main loop to the logger thread.

For context, I'm running rerun inside of an egui app, similar to here: https://github.com/rerun-io/rerun/tree/e572612a9fa13c0b3168610f119bc8b7a4034f83/examples/rust/extend_viewer_ui. This gives me a little more access to rerun internals. I'm trying to add some quick buttons to my custom sidebar to essentially toggle the visibility of items in a 3D space on/off. From what I can tell, I'll need to log to the blueprint to the correct path to be able to do this. There seems to be a handful of API options to do the logging, - in `SystemCommand`: https://docs.rs/rerun/latest/rerun/external/re_viewer_context/enum.SystemCommand.html#variant.UpdateBlueprint - using a custom

RecordingStream

set to log to the blueprint directly - using

RecordingStream::send_blueprint

Unfortunately, all of these require the

StoreId

of the active blueprint and I can't seem to find a way to get it. The closest seems to be

StoreHub::active_blueprint_id

, but I don't see anyway to access that. Any suggestions here for how I can get that? Or another route I could take? Thanks!

Hey, I'm exploring using rerun to visualize geometry algorithms (and help debug edge cases), I have a rough implementation prototype for parry at https://github.com/Vrixyz/parry/pull/6. Is there a way to hide some values, or specify that a specific log should be displayed only on that frame ? Thx! I'm using rerun's

.set_time_seconds

and

.log

for attached demo. - I'd like to show input points only at first frame, then only facets, then only valid facets (but maybe another need would be to persist some data multiple frames) https://cdn.discordapp.com/attachments/1393153897325002814/1393153897517809706/parry_step_debugging.mp4?ex=6872232c&is=6870d1ac&hm=aa1692e9fa99b41628eef4de6cc50ed6cc2852c3d76467211b9f0a3643038e48&

I am trying to live plot the pose estimation of the camera. I have two issues. 1. I am not able to follow the camera ( track ) 2. If I try to reset the origin of the 3D window, it won’t let me with mouse. Only way I can see the out of view pose path is by zooming out. Any suggestions on how I can fix this ? https://cdn.discordapp.com/attachments/1392843579793215549/1392843580175028235/IMG_9541.jpg?ex=6871022b&is=686fb0ab&hm=2b673d2e401cd34ee443377e5a1c90e512f0da79808801d252a77d4e2dfe295a&

Hello, With the ability to log custom component, as mention here: https://rerun.io/docs/concepts/entity-component#adding-custom-data, it would be great to be able to change the color element use by the viewer directly inside the viewer. Basically, use any component of the entity type that make sense (Float, Int, Array, ...) Currently I didn't find a way to do this, I needed to precompute my color Component in my code. In my currently use case, I'm mostly interested in the

Point3D

archetype

Hi team, I am new to rerun and trying to make a live visualizer tool for robot teleoperation recording - with 2 cameras, tactile sensors, robot states, point cloud, etc. I subscribe to all the ROS2 topics and

rr.log

according to the respective type. It works fine on the direct host machine with the native GUI but has a lag when I use the web_viewer mode with

grpc

. My team prefers to run everything from inside dockers (combined with another ROS2 functionality) and I am having a hard time figuring it out. I don't want to use web_viewer since that brings in ~3 seconds lag on the visualization. Is there any way to run the native viewer through the docker?

Hi, I would like to log the motion of a parallel mechanism, similar to the example shown [at this link](https://github.com/Rhoban/placo-examples/blob/master/kinematics/planar_2dof_trajectory.py). My current robot is a Stewart platform, but this example illustrates the concepts well. While it is possible to rotate the two motors, the connection is not maintained. Additionally, directly moving the end effector is not feasible. Do you have any ideas on how to accomplish this? https://cdn.discordapp.com/attachments/1389638923768172694/1389638924951097546/image.png?ex=68660259&is=6864b0d9&hm=b6494c11dd36a0c7dfd26e264dce7718d01a99d2ede0986a011dfcfb27fd4cf4&