Currently, I save the memory in remote / non-persistent storage everytime the queue is killed.
It is reloaded in-flight everytime the queue is non-empty.
When Chrome crashes, the state will be reloaded and may be out of sync. See issue #2 .
we may want to define proper "sessions". How can we define "the user closed all its windows, it is restarting" ?
If we go long-term statistics, sessions are required to compress data.
We don't clean the storage. It has a limited size and many garbage collectors.
Sessions should be defined for long term statistics but this is not a valid issue.
Currently, I save the memory in remote / non-persistent storage everytime the queue is killed. It is reloaded in-flight everytime the queue is non-empty.