# Release Note

## v1.1
### Features
- Create service that makes pods can communicate with each other
- ```kubectl create -f templates/service.yaml```

## v1.0.12 - 2020/03/28
### Features
- Modify the threshold of the light curtain
- the h141st's light curtain threshold is modify from 90 to 70.
- Modify the stdout_logfile_maxbytes, stderr_logfile_maxbytes of the supervisord config
- all the camera is change from 1MB to 50MB

## v1.0.11 - 2020/03/27
### Features
- Modify the ROI for new camera view
- the camera angle of the h141st is modified

## v1.0.10 - 2020/03/02
### Camera
- Add a new camera: `h42nd` (2020/03/03)
### Features
- Log inference time in watchdog.
- Replace tmux to supervisord.
- The supervisor webUI is avalible at port 9001 in a container of the pod, but we have to forward the pod's port to the host's port first.

        # kubectl port-forward --address $host -n $namespace $pod_name $host_port:9001
        kubectl port-forward --address 0.0.0.0 -n smfac safetydetection1-0123456789-abcde 9001:9001

- And the web can be accessed by `$host:$host_port`
- The model-updated python script/template/yaml are also modified.
### Bugs
- If the reading frame function is timeout, set the time to ```$timeinterval``` by milliseconds not seconds.
- If the flag of the reading frame/inferencing model function is not freeze but the processed frame is 0, set the ```avg_$function_time``` to ```-1``` not the default ```$timeinterval```. Because the service is not freeze by the function, it is caused by someone else.

## v1.0.9 - 2020/02/14
### Features
- Add more cpp uniitests.
- Log RTSP response time in watchdog.
- Put reading frame function to another thread and add a timeout mechanism.
- If the thread is freeze when the RTSP server is not stable, the main thread will create another reading thread to receive the frame. And the old reading thread will terminate if it is back online.

## v1.0.8
### Features
- Add HTTP request timeout.
- Log http response time in watchdog.
- Update failure times mechanism
- When safeDetector fails, then add the counter 1, and do not count the request failure.

## v1.0.7
### Features
- Add self-monitored mechenism to monitor worker
- If the worker cannot connect to the RTSP stream, the master will restart the worker. After try out over maximum retry times, then stop the worker and disconnect the heartbeat and let the Azure alter.

## v1.0.6 (not backward compatible)
### Features
- Advise dangerous event wording

## v1.0.5
### Bugs
- Fix memory leak.
- It is caused by use cv::Mat data in muti-thread and not clear when deque.pop_front().

## v1.0.4
### Features
- Update two cases of dangerous events.

## v1.0.3
### Features
- Modify failure counting mechanism.
- Remove counting of socket connection failure.

## v1.0.2
### Bugs
- Update model id.

## v1.0.1
### Features
- Reduce updating skill times.
- Update skill every second, not per frame.