Effective CQL - Mastering Apache Cassandra - page 114

larger the number of rows indexes store in memory, which means a larger
memory footprint and faster row retrieval.
Note
Cassandra 2.1.0 and onwards, introduced two intervals:
min_index_interval and max_index_interval . This is slightly
smarter than the previous version. Based on the frequency of access of an SST-
able, it determines the index interval to use. So, for a rarely accessed SSTable, the
index interval will be equal to max_index_interval , and for an in-demand
SSTable, the index interval will be min_index_interval . For the SSTables
that are neither in high demand nor in low demand, the index interval will be set
to somewhere between the min and max values. The index interval also depends
on the availability of free memory in the index summary memory pool.
• memtable_flush_period : Memtables are serialized to SSTable after a time
period defined by this attribute. This represents the number of milliseconds after
which a Memtable is forced to be flushed to the disk. The default is zero.
• replicate_on_write : This is the only feature in Cassandra 2.0.x. When
replicate_on_write is set to true , the data is written to all the replicas ir-
respective of the consistency level setting. It is true for counter tables by default.
• speculative_retry : Cassandra makes only as many requests as consistency
level dictates. If a node dies during a read, Cassandra waits till timeout and then
responds with a timeout error message. With speculative_retry , Cas-
sandra avoids failure by sensing the unusual delay that a replica takes, and retries
on a different replica. Here are the available settings:
◦ ALWAYS : Cassandra reads from all the replicas and returns as soon as the
consistency level is met.
◦ Xpercentile : Retry triggers if the read operation takes more time
than what the fastest X percent of reads took in the past. For example, if
it is set to 99percentile , the retry will trigger if a read operation
takes more time than the fastest 99 percent of the queries took.
◦ Xms : Retry triggers if the read did not return after X milliseconds of
sending the request. So, if you wanted a speculative retry to happen if a
replica did not respond in under 42 ms, set speculative_retry to
42 ms .
◦ NONE : No retries.
Here's an example: