surfaces are rendered in parallel that should not be visible

[mlohry]

this image is a render of a field with a camera positioned above a wing, where the surface topology is defined by tris in 3D space. geometric entities on the lower side that shouldn't be visible are seen in the render -- the flap farings, and the aft part of the engine nacelle. this was done on 28 ranks:

running the same flow on 1 rank i see the expected output:

[mlohry]

In the above bad image, I was giving conduit/ascent only the connectivity for the surface elements, but it still had all the solution and coordinate information for the volume vertexes. I removed all that and left only those on the surface, purely to make the file exports smaller, but that seems to have inadvertently fixed the render issue so now it looks fine in parallel.

[mlohry]

Digging deeper I think there might be an issue here related to multiple topologies. I have the following input:

-
  action: "add_scenes"
  scenes:
    s2:
      plots:
        p1:
          type: "pseudocolor"
          field: "p_surface"
      renders:
        r1:
          image_width: 1600
          image_height: 2400
          image_prefix: "./pressure_planform_view_surface_only_%06d"
          camera:
            fov: 30.0
            look_at: [1440.0,686.0,182.0]
            position: [1440.0,686.0,8274.0]
            up: [0.0, 1.0, 0.0]
            zoom: [2.7]

I'm providing conduit with two topologies and corresponding solutions named like data["topologies/volume/coordset"] = "volume_coords" and data["topologies/surface/coordset"] = "surface_coords" along with their own connectivities. Field data is specified like data["fields/p/topology"] = "volume" and data["fields/p_surface/topology"] = "surface".

When rendering p_surface (and only in parallel) I'm getting the above bad image with artifacts of the back side showing through. Then when I simply removed all the volume data leaving only the one surface topology, I get the correct render. So somehow the p_surface render is "seeing" the volume topology when my understanding from discussion in #1296 that these should effectively be separate. Maybe some place not checking for topology name and just taking the first available by default?

[cyrush]

Thanks for the updated info, I will see if I can reproduce and get to the bottom of this.

[cyrush]

There issue here is the original spatial extents of the overall dataset are impacting the camera.

Specifically the near and far clipping plane.
The volume is very large compared to the other dataset, leading to a near clipping plane that is beyond the dataset.
That is what causes the mesh to render in a way that looks inside out.

This can be avoided by providing the near_plane and far_plane camera options, or by providing dataset_bounds

Here is an example set of actions:

-
  action: "add_scenes"
  scenes:
    s2:
      plots:
        p1:
          type: "pseudocolor"
          field: "p_surface"
      renders:
        r1:
          image_width: 1600
          image_height: 2400
          image_prefix: "./tout_pressure_view_surface_and_volume_%06d"
          camera:
            fov: 30.0
            look_at: [1440.0,686.0,182.0]
            position: [1440.0,686.0,8274.0]
            up: [0.0, 1.0, 0.0]
            zoom: [2.7]
            # GOOD SETTINGS!
            # near_plane: 274.769
            # far_plane: 27476.9
            # BAD SETTINGS
            # near_plane: 8274
            # far_plane: 827400
            # fix with dataset bounds
            dataset_bounds: [92.5, 2562, 0, 1159.93, 17.7024, 343.14]

While exploring this I noticed that the near_plane and far_plane camera params were not being output as part of Ascent::info(). #1299 resolves that issue.