In #14 we introduce a B-tree block index mapping from start_timestamp to filename. There's currently nothing enforcing unique start_timestamps besides an expectation of well-behaved data. We can consider creating a sort of "timestamp checkpoint", where we refuse to keep any data with timestamps less than or equal to our latest start_timestamp.
If we want to avoid losing late data, we can consider keeping the block around for awhile in-between resetting the active block and flushing the old block to disk. Something like Timon.
In #14 we introduce a B-tree block index mapping from start_timestamp to filename. There's currently nothing enforcing unique start_timestamps besides an expectation of well-behaved data. We can consider creating a sort of "timestamp checkpoint", where we refuse to keep any data with timestamps less than or equal to our latest start_timestamp.
If we want to avoid losing late data, we can consider keeping the block around for awhile in-between resetting the active block and flushing the old block to disk. Something like Timon.