Error while loading sharded meshes

[Mencu]

Hello,

I have noticed an error in the web console that I can barely debug, for which I need a bit of help with.

Because we have a big data set, we have sharded (draco compressed) the data, in order to be fed to Neuroglancer. Some cell meshes still exceed a certain size and need to have a truncated requests/responses. I have noticed some of these meshes end up in an error in the console:

Their data is being fetched by Neuroglancer.
First request:

Second request:

But the mesh is not shown in the Viewer. I have tried to debug the chunk manager, but I had no luck. Also, the shard files seem to work, as I've loaded the problematic meshes with cloud-volume.

Do you have any idea where I could start debugging?

Best,
Andrei Mancu

[jbms]

proc_exit 1 indicates an unexpected error decoding the mesh, but I will have to investigate more what the issue is.

[Mencu]

Okay! I think I might have a lead then. Thanks Jeremy!

[nkemnitz]

@Mencu Running into the same issue with large draco mesh shards, did you ever figure this one out? CloudVolume/DracoPy is decoding the mesh just fine - 16,128,411 vertices, 31,662,220 faces, 0 vertex normals.

[nkemnitz]

Looks like it's a 32-bit memory limitation.

ALLOW_MEMORY_GROWTH=1 only works up to 2GB. Adding MAXIMUM_MEMORY=4GB already resolves the proc exit error for the 16M vertex mesh (see #470).

For a larger mesh (22.5M vertices), it now fails with RangeError: Start offset -1466876952 is outside the bounds of the buffer -
maybe hitting 4GB? If so, I believe we need to wait for https://github.com/WebAssembly/memory64?