## QA PASSED See QA report document: e08aef54-35e2-418f-aac0-a03329132a2e **Test Results:** - E2E test: 2 runs, both passed (~77s each) - Incus client unit tests: 32/32 passed - No "Instance is busy" errors observed **Verification:** - _wait_for_operation() now correctly loops until "Success" status - 300s total timeout with 30s poll intervals working correctly - Container creation time: ~77s (within expected 70-90s range) All hypotheses tested. No issues found. Fix verified working.

**Fixed: Operation timeout causing race condition** **Problem:** The _wait_for_operation() function in app/services/incus.py was calling the Incus /wait?timeout=60 endpoint once with a 60-second timeout. When the 2.7GB pre-baked image took ~90 seconds to extract, the /wait endpoint returned after 60 seconds with the operation still in 'Running' status. The code didn't verify the operation was actually complete - it only checked for explicit 'Failure' status. This caused the code to proceed and try to start the container while creation was still in progress, resulting in: "Failed to create instance start operation: Instance is busy running a 'create' operation" **Solution:** Modified _wait_for_operation() (lines 183-240) to: 1. Loop until the operation status is explicitly 'Success' (not just absence of failure) 2. Use a 300-second total timeout (5 minutes) instead of 60 seconds 3. Poll every 30 seconds (shorter intervals so we detect completion quickly) 4. Only return success when metadata.status == 'Success' Also increased the httpx client timeout from 120s to 360s to accommodate longer operations. **Verification:** - E2E test passed in 71.57 seconds: test_connect_info_desktop_container - All 32 Incus client unit tests pass - Container creation now waits properly for the full image extraction **Files changed:** - /home/gateway/gitea/cdev/app/services/incus.py - Fixed _wait_for_operation() to loop until completion

## QA FAILED See QA report document: c0fedf07-f23c-47b3-affd-4de683b43676 **Critical Bug Found:** Operation timeout causes container creation failure The pre-baked image (cdev-desktop, 2759.55MiB) takes ~90 seconds to extract, but the `_wait_for_operation()` function in `app/services/incus.py` times out after 60 seconds. When it times out, the code returns success and immediately tries to start the container, resulting in: ``` Failed to create instance start operation: Instance is busy running a 'create' operation ``` **Evidence:** - E2E test failed after 60.68s with the above error - Manual testing confirmed the create operation was still running after timeout - Monitoring showed the operation took ~90 seconds total to complete **Recommended Fix:** 1. Increase the default timeout in `_wait_for_operation()` to at least 120s 2. Or add a `wait_timeout` parameter to the `create()` method Returning to todo for fixes.

## Implementation Complete ### What was done: 1. **Created `scripts/build-desktop-image.sh`** - Script to build a pre-baked Incus image containing: - Xubuntu desktop with TigerVNC - Node.js 22 - Google Chrome with Claude extension policy - Claude Code CLI - Tailscale (installed but not connected) - SSH server - Pre-configured 'xubuntu' user with VNC settings 2. **Added `list_images()` to IncusClient** (`app/services/incus.py:573-580`) - Required for checking if pre-baked image exists 3. **Fixed IncusClient.create()** (`app/services/incus.py:307-311`) - Changed to use 'alias' for local images instead of 'fingerprint' 4. **Added config settings** (`app/config.py:49-53`): - `DESKTOP_IMAGE_ALIAS`: Image alias to use (default: "cdev-desktop") - `DESKTOP_IMAGE_FALLBACK`: Whether to fall back to fresh install if image not found 5. **Added helper functions** (`app/services/container_manager.py`): - `check_prebaked_image_exists()`: Check if pre-baked image exists - `_set_user_password()`: Set user password in container - `_set_vnc_password()`: Set VNC password for user - `_add_ssh_authorized_keys()`: Add SSH keys for user - `connect_tailscale()`: Connect to Tailscale (for pre-installed Tailscale) 6. **Added `create_desktop_container_fast()`** (`app/services/container_manager.py:1535-1716`) - Fast path that uses pre-baked image - Only performs: wait for ready, set passwords, add SSH keys, copy credentials, start VNC, connect Tailscale 7. **Updated `create_desktop_container()`** (`app/services/container_manager.py:1269-1328`) - Auto-detects pre-baked image and uses fast path when available - Falls back to fresh install if image not found (configurable) - Added `force_fresh_install` parameter to bypass pre-baked image ### Test Results: - E2E lifecycle tests: **4 passed** in 96.20s - Desktop container creation time: **~67s** (down from ~8 minutes) - Pre-baked image verified working with VNC, SSH, Claude Code authentication ### Performance Notes: The target was 60 seconds, but actual time is ~67s. The bottleneck is image unpacking (I/O bound for the 2.7GB image). This is still a massive improvement from ~8 minutes with fresh installation. ### Files Changed: - `app/config.py` - Added DESKTOP_IMAGE_ALIAS and DESKTOP_IMAGE_FALLBACK settings - `app/services/container_manager.py` - Added 413 lines for pre-baked image support - `app/services/incus.py` - Added list_images() method, fixed alias handling - `scripts/build-desktop-image.sh` - New script to build pre-baked image (260 lines) ### Verification: Pre-baked image exists at alias 'cdev-desktop', size 2759.55MiB.