Cassandra Schema¶

Data Points Column Family¶

The row key is made up of 3 parts all concatenated together.

1. Metric name (UTF-8) null terminated.
2. Row time stamp (64 bits). The time stamp for the row to begin at.
3. Datastore type (UTF-8). Length preceded (1 byte) utf-8 string with null at the end. Yes kinda redundant but it works nice.
4. Concatenated string of tags (tag1=val1:tag2=val2…) (UTF-8)

The column name is different depending on of it from the legacy row_key_index or not.

Legacy Column Name: The column name is 32 bits of data. The first 31 bits is the unsigned time offset from the row key (in milliseconds). The last bit is unused.

Current Column Name: 32 bit value (no shifting and unsigned) that represents the number of time units from the row key timestamp. The time unit is either seconds or milliseconds depending on how the cluster was configured when first created.

The value of the column varies depending on the type of value. The exact format is defined by the writeValueToBuffer method on the DataPoint.

The length of the row defaults to exactly three weeks of data or 1,814,400,000 columns. The width of the row can be changed in the cluster configuration when the cluster is created. It cannot be changed after the schema has been created.

Row Key Index Column Family¶

Legacy This row is primarily used when querying the data. The row key is the name of the metric. The names of the columns are the row keys from the data_points column family. The columns have no values.

Row Key Time Index Column Family¶

The purpose of this table is to index the time frame a metric shows up for. It also provides a future mapping if we ever want to get away from the data points table. This is the first table hit when doing a query.

Row Keys Column Family¶

This is the second table hit when doing a query. The values in this table have all the combinations of tags that show up for the specified metric. Results from this table are filtered if any tags are specified in the query. The results are then used to lookup the actual data in the data points table.

Tag Indexed Row Keys Column Family¶

When a metric is configured with an additional index this table is used for queries under certain circumstances. When you configure a metric to be indexed on a tag, say the host tag, data will be written to this table as well as the row keys table. If you do a lookup on the metric and specify a host tag in the query the query will use this table to speed up the query results. if no host tag is specified then the query will use the row keys table like usual.

String Index Column Family¶

Just an index to lookup what metric names, tag names and tag values are in the system. There are three rows one for each of the above mentioned.

Service Index Column Family¶

All of the information stored using the metadata api is stored in this table.

Spec Column Family¶

This is an internal table currently only used for storing row width and time granularity. The values in this table are not to be changed by external means.