Optimize compaction

[kamijin-fanta]

summary

• Add use_ttl new option.
• if use_ttl = false, disable unnecessary partition reading at compaction.

-- examle create table queries.
CREATE CUSTOM INDEX table_index ON ks1.table ()
USING 'com.stratio.cassandra.lucene.Index'
WITH OPTIONS = {
    'refresh_seconds': '15',
    'use_ttl': false,
    'partitioner': '{type: "vnode", vnodes_per_partition: 8}',
    'schema': '{
      fields: {
          filed: {type: "string"}
      }
    }'
};

describe

I was investigating the compaction speed when enabling this plug-in.

When SSTable increased, the speed of compaction fell below 1%. I anticipate that read latency will be higher. Then I checked if it really needed partition reading at compaction time.

72c4875#diff-298e208e90f66d25100b3cd7dbf19d5bR71

This code was added for support TTL . If do not use TTL, this overhead is unnecessary ...? I put in debugging code in IndexWriterWide#commit(). What is SinglePartitionReadCommand reading when transactiontype == COMPACTION? As a result, did not read anything. So I added a flag to disable this.

(I do not fully understand the life cycle of Cassandra Indexer. Please point out if there is a mistake.)