add stall mechanism for a read-after-write

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.

[EricPyZhou]

1. modify get_watches() to only query watches from watches Dict.
2. add partial removal in add_watch_notification() to get watches from watches Dict, and then add the WATCH_NOTIFICATION to eventQueue.
3. to stall the direct request(get, exists ...), we call get_watches() to verify if there is still watch waited to be delivered. if there is, add the direct request back to eventQueue. if no, future.set_result.

The modification is based on the following reason:

1. now add_watch() and add_watch_notification() are the only two function that add and remove watches from watches Dict respectively. it is easier to manage the Dict.
2. w.generate_messages() in the direct result handling is removed because w itself does not have watch event type. Watch_event is not and can not be specified because it is add_watch_notification()'s responsibility to assign the correct watch event, which is not possible in the original get_watches(). Therefore, we rely on add_watch_notification() to interact with watches Dict, and modify get_watches() to just retrieve watches info.
3. we stall the request by adding it back to the queue so that it may be appended behind the WATCH_NOTIFICATION. If WATCH_NOTIFICATION is before CLOUD_DIRECT_RESULT in the queue, CLOUD_DIRECT_RESULT will be resolved next time because watch notification is delivered; If WATCH_NOTIFICATION is after CLOUD_DIRECT_RESULT in the queue (or just not in the queue), CLOUD_DIRECT_RESULT will be in the queue waiting for it until it arrives (assuming the watch cloud function performs normally).
   Now I do think waiting in the queue forever is a bit strict. maybe we can think of ways to kinda remedy it.