- Convert queue_id to int in assign_job_to_worker and cancel_job
- Fixes type mismatch issues between str and int comparisons
- Ensures active_jobs lookup works correctly

- Update global cluster_master instance for cross-module access
- Allows queue manager to access active jobs for cancellation
- Enables proper job_id logging in cancellation messages

- Assignment: 'Assigned job 1 (job_xxx) to worker'
- Cancellation: 'Job 1 (job_xxx) cancelled'
- Waiting: 'Job 1 waiting for available workers' (no cluster id yet)
- Clear correlation between queue and cluster job IDs

- Show actual cluster job_id in cancellation log
- More informative logging with job_xxx format
- Consistent with assignment logging

- cancel_job logs job cancellation to console
- Provides feedback when jobs are cancelled
- Clean, single log message per cancellation

- Silent job cancellation operations
- No logging for cancel commands sent to workers
- Clean operation without console noise

- Management loop silently completes cancelled jobs
- Eliminates duplicate completion messages
- Cleaner logging with only cancel_job messages

- Prevents active_jobs from persisting across server restarts
- Eliminates leftover job tracking from previous runs
- Ensures clean state for each server start

- Only complete and log cancelled jobs that are still active
- Prevents duplicate completion messages for jobs already handled
- Cleaner logging for job cancellation

- Don't remove jobs based on age since jobs can run for hours
- Leftover jobs from previous runs will be handled by restart cleanup
- Prevents interrupting long-running legitimate jobs

- Remove active jobs older than 10 minutes to prevent accumulation
- Cleans up leftover jobs from previous runs or crashes
- Prevents duplicate job tracking issues

- Management loop only completes cancelled jobs, doesn't send duplicate cancel messages
- Cancel_job sends the cancel command, management loop cleans up
- Prevents duplicate cancellation logs

- Check if job is already assigned before assigning again
- Prevents multiple active jobs for the same queue entry
- Fixes duplicate cancellation attempts

- Pass queue_id to workers for proper cancellation detection
- Workers now check correct job id for cancellation status
- Workers receive effective stop commands via database polling

- Workers receive stop processing commands when jobs are cancelled
- Ensures workers halt processing immediately on cancel
- Maintains proper cleanup of resources

- Workers are freed immediately when jobs are cancelled
- Clean up active jobs in cluster master when cancelling processing jobs
- Remove unnecessary cleanup from restart (handled by cancel)

- Implements job cancellation by notifying workers
- Sends cancel messages to local backend or remote clients
- Cleans up cancelled job resources

- Pass queue_id to _assign_local_job method
- Fix NameError when assigning local jobs

- Store queue_id in active_jobs tracking
- Properly detect cancelled jobs by checking queue status
- Clean up worker resources when jobs are cancelled
- Workers become available for new jobs after cancellation

- Cluster master sends cancel_job messages to backend/client when jobs are cancelled
- Add _handle_cancel_job to process cancel confirmations from clients
- Workers can be notified faster to stop processing and free resources

- Restarted jobs set to 'queued' status
- Cluster master looks for 'queued' jobs, sets to 'processing', then assigns
- Proper job lifecycle: queued -> processing -> assigned -> completed/failed

- Restarted jobs now set to 'processing' with empty job_id
- Cluster master will pick up restarted jobs for assignment

- QueueManager now only handles job submission and management
- Job processing is handled exclusively by cluster master
- Eliminates duplicate queue processing between web and cluster processes

- Show when worker connects to backend
- Show when worker registers
- Help debug why jobs aren't being received

- Queue manager marks jobs as processing, cluster master assigns them
- Changed query back to look for processing jobs without job_id

- Always use cluster master for job assignment, even for local jobs
- Remove separate local processing path
- Local processes treated same as remote, except for auto weight adjustment

- Allow local jobs to start even if worker socket check fails
- Backend handles worker availability, queue manager should always allow local processing

- Add restart button in job history for cancelled jobs
- Add /job/<id>/restart route in web interface
- Add restart_job method in QueueManager to reset cancelled jobs to queued

- Worker now prints when receiving jobs and sending results
- Cluster master uses TCP polling for consistency with clients

- Use get_queue_by_job_id to check job status
- More reliable than TCP polling for local jobs

- Initialize self.pending_jobs dict for job monitoring tasks
- Fixes AttributeError when assigning local jobs

- Use 'analyze_request' instead of 'analysis_request'
- Match the expected message type in worker processes

- Change response.get('msg_type') to response.msg_type
- Change response.get('data') to response.data
- Message objects don't have get method, use attributes instead

- If --weight is not specified, master weight changes to 0 when clients connect
- If --weight is specified, master participates in job selection with that weight

- Cluster master now participates in job selection even when clients are connected
- Local workers compete with external workers based on weight and VRAM

- Jobs are inserted as 'queued', not 'processing'
- Cluster master now finds and assigns queued jobs

- Shows total VRAM detected on local GPUs when registering processes

- Local workers now require sufficient VRAM like other workers
- Since the server has 24GB VRAM and jobs need 16GB, the check passes normally

- Local jobs now monitor for completion and handle results
- Prevents jobs from hanging without result retrieval

- Changed local client ID to 'local' and marked as local to prevent cleanup
- Local clients are not cleaned up after 60 seconds
- Prevents 'Client local disconnected' messages