You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Segfault in omni.syntheticdata during the post-process graph tick (Isaac Sim 6.1.0) #905
Rarely, the app segfaults inside libomni.syntheticdata.plugin.so while SyntheticData._post_process_graph_tick evaluates its graph. It runs headless, with cameras and annotators. I've seen it three times:
at the first renders after the timeline starts (1 in 11 starts);
at a render after reloading a stage (1 in 70+ runs);
mid-run in Isaac Lab 3.0.0rc1, 32 envs with two cameras, on a normal step (1 in 4 runs).
From the minidumps: all three crash at the same instruction (plugin offset 0x137f0d, mov 0x10(%r13),%rcx), in a loop over 32-byte records. The array's pointer and count don't agree: twice the pointer was NULL with a count of 3, and once it was valid with a count of about 250 million. That suggests the tick reads a stale or freed object. The last command before each crash was CreateAttrCommand on state:_sdp_intergraph_downstream_node_handles_.
I can't trigger it on demand. 240 runs aimed at the first two cases didn't crash.
Two more crashes, both in Isaac Lab mid-run on a normal step, 8 and 18 minutes in. Same instruction (0x137f0d) and same last command as before. The array pointer and count were 0x100000000 / 1 in one and NULL / 3 in the other.
That's 3 crashes in 9 runs of this evaluation now. Two of the three had the GPU to themselves, so it happens without anything else running on the GPU.
Still on 6.1.0, Isaac Lab 3.0.0rc1, driver 595.99.02. Haven't tried 6.2.0 yet.
Hi, thanks for reporting your issue. Please try with Isaac Lab 3.0 GA with Isaac Sim 6.2 and if you still see the same, share the full log with your repro case so we can try to repro it on our end.
Activity
mfinean commentedon Oct 25, 2026
Two more crashes, both in Isaac Lab mid-run on a normal step, 8 and 18 minutes in. Same instruction (
0x137f0d) and same last command as before. The array pointer and count were 0x100000000 / 1 in one and NULL / 3 in the other.That's 3 crashes in 9 runs of this evaluation now. Two of the three had the GPU to themselves, so it happens without anything else running on the GPU.
Still on 6.1.0, Isaac Lab 3.0.0rc1, driver 595.99.02. Haven't tried 6.2.0 yet.
PeterL-NV commentedon Oct 26, 2026
Hi, thanks for reporting your issue. Please try with Isaac Lab 3.0 GA with Isaac Sim 6.2 and if you still see the same, share the full log with your repro case so we can try to repro it on our end.