After open sourcing, it was noted that "dumponexit" generated recording file size do not exceed ~10 mb. 10 mb is the size of the default memory buffer system.

Turns out that some logic were lost in code refactoring:

On a full global buffer, the notification is never sent to the recorder thread for it to flush to disk. This means the in-memory buffer system fills up until eventually no free space is left. At this point threads will start to discard data and write "data loss" events.

Need to add back the missing notification.