Title: Antigravity freezes for several minutes on every launch, immediately after model loading completes (Ubuntu 24.04, reproducible with blank workspace + extensions disabled)
Environment:
- OS: Ubuntu 24.04.4 LTS (kernel with
apparmor_restrict_unprivileged_userns=1) - Antigravity version: 1.23.2-1776332190 (installed via official apt repo)
- Hardware: Intel Haswell iGPU, 4 cores, 12GB RAM
- Installed via:
sudo apt install antigravity(official Google apt repo)
Description:
Antigravity freezes/becomes unresponsive (“The window is not responding” dialog) on nearly every launch, for anywhere between 30 seconds and 5+ minutes, before sometimes recovering on its own. This is fully reproducible even with a blank workspace (--new-window, no folder opened) and all extensions disabled (--disable-extensions), ruling out project size, file watchers, and third-party extensions as the cause.
Steps already ruled out (confirmed via logs, not the cause):
- GPU driver — fixed
libva/iHD_drv_video.somismatch by forcingLIBVA_DRIVER_NAME=i965for our Haswell iGPU. No change to freeze. - AppArmor sandbox — found
apparmor="DENIED" ... capability=21 sys_adminblockinguserns_createon every launch (Ubuntu 24.04’sunprivileged_usernsrestriction). Added a permissive AppArmor profile for the binary; denials stopped, freeze persisted. - inotify limits — found
ENOSPC: System limit for number of file watchers reached. Raisedfs.inotify.max_user_watchesto 524288 andmax_user_instancesto 512. No change. - Workspace indexing — added a
.antigravityignoreexcludingvendor/,node_modules/, logs, etc. No change. - Extensions — reproduces identically with
--disable-extensions. - Workspace — reproduces identically with
--new-window(no project open at all).
Log evidence (main.log), consistent across every launch:
[Auth] Auth state changed to: signedIn
[CloudCode] POST v1internal:fetchAvailableModels traceId: ...
— immediately followed by total silence for 4+ minutes, then:
[error] CodeWindow: detected unresponsive
[error] [uncaught exception in main]: UnresponsiveSampleError ...
with repeated stack samples inside the renderer showing tight loops in minified functions involving Array.sort, Array.shift, Array.forEach, Set.forEach, and shouldComponentUpdate, called from what appears to be setSelectedModel/setCascadeConfig codepaths (based on function names visible in earlier, less-minified traces from the same build).
Pattern: The freeze consistently begins right after fetchAvailableModels/model-list load completes and the UI attempts to apply the selected model/cascade config, suggesting a render-thread loop triggered during post-auth UI state updates rather than anything workspace- or extension-related.
Request: Would appreciate guidance on whether this is a known issue with a fix in progress, and if there’s a way to disable/reset the automatic model-selection state on startup as a workaround.