Database Reference
In-Depth Information
• Use the standard OpenTSDB graph, such as the one shown here, for
monitoring, which is the primary OpenTSDB use case:
However, here we are more interested in the OpenTSDB design principles, since
this shows us the experience of others rather than our limited experience. So, the
designers give us those pointers (with the explanation).
The overall design of TSDB
Here are the lessons that we can glean from the overall design of TSDB. All
OpenTSDB data points are stored in a single, massive table, named tsdb by default.
This is to take advantage of HBase's ordering and region distribution. All the values
are stored in the t column family.
The lesson that we learned here is to use a simple schema, as you will have enough
complexity anyway. So, go with the recommended HBase design.
The row key
Row keys are byte arrays that comprise of the metric UID, a base timestamp, and the
UID for tag key-value pairs, for example, <metric_uid><timestamp><tagk1><tagv
1>[...<tagkN><tagvN>] . By default, UIDs are encoded on three bytes.
 
Search WWH ::




Custom Search