DRILL-8259: Supports advanced HBase persistence storage options

Description

Maximize performance with HBase as persistent storage.

Documentation

Example in drill-override.conf

sys.store.provider: {
  class: "org.apache.drill.exec.store.hbase.config.HBasePStoreProvider",
  hbase: {
    table : "drill_store",
    config: {
      "hbase.zookeeper.quorum": "zk_host3,zk_host2,zk_host1",
      "hbase.zookeeper.property.clientPort": "2181",
      "zookeeper.znode.parent": "/hbase-test"
    },
    table_config : {
      "durability": "ASYNC_WAL",
      "compaction_enabled": false,
      "split_enabled": false,
      "max_filesize": 10737418240,
      "memstore_flushsize": 536870912
    },
    column_config : {
      "versions": 1,
      "ttl": 2626560,
      "compression": "SNAPPY",
      "blockcache": true,
      "blocksize": 131072,
      "data_block_encoding": "FAST_DIFF",
      "in_memory": true,
      "dfs_replication": 3
    }
  }
}

Configuration requirements

Key	Type	Value Example	Reference
durability	String	ASYNC_WAL / SYNC_WAL / SKIP_WAL	Durability
compaction_enabled	Boolean	false	COMPACTION_ENABLED
split_enabled	Boolean	false	SPLIT_ENABLED
max_filesize	Number	10737418240	MAX_FILESIZE
memstore_flushsize	Number	536870912	MEMSTORE_FLUSHSIZE
versions	Number	1	MAX_VERSIONS
ttl	Number	2626560	TTL
compression	String	SNAPPY / LZ4	Compression$Algorithm
blockcache	Boolean	true	BLOCKCACHE
blocksize	Number	131072	BLOCKSIZE
data_block_encoding	String	FAST_DIFF / PREFIX	DataBlockEncoding
in_memory	Boolean	true	IN_MEMORY
dfs_replication	Number	3	DFS_REPLICATION

Testing

Added the TestHBaseTableProvider#testStoreTableAttributes()

@luocooong Thanks for submitting this. I was wondering, is there a reason why we are storing these variables in drill-override.conf instead of the configuration for the storage plugin? IMHO, it is better to put it in the config so that you don't have to restart Drill any time you make a config change

@cgivre Hi, Thank you for the questions. Actually, Drill PStore' variables are split from the storage configurations, because the goal is to define the initial variables before the Drill startup. And then, it has a different lifecycle from the storage configuration, so it is not recommended to be placed in the storage plugin.

Why compaction_enabled": false? I thought compaction is important for hbase to boost performance?

Why compaction_enabled": false? I thought compaction is important for hbase to boost performance?

As you know, HBase is a nightmare for operational services due to the complexity of the settings. The actual value in the above example is not a recommended value, no unique value is appropriate for every case, but is simply the type of value that this parameter has to fill, is "true/false", not "0/1".

And, would you mind helping me append this updated document to the drill-site?

Why compaction_enabled": false? I thought compaction is important for hbase to boost performance?

As you know, HBase is a nightmare for operational services due to the complexity of the settings. The actual value in the above example is not a recommended value, no unique value is appropriate for every case, but is simply the type of value that this parameter has to fill, is "true/false", not "0/1".

And, would you mind helping me append this updated document to the drill-site?

Of course :)

@cgivre @Z0ltrix I added two new options. If namespace is used, the namespace:table semantics are applied. And we can also use the table configuration only. The family is also an optional.

@Z0ltrix Would you mind doing a formal review on this PR? @luocooong asked me but I don't really have enough experience with HBase to comment intelligently on this. If you're already happy with this, all you have to do is leave a +1.

@Z0ltrix Would you mind doing a formal review on this PR? @luocooong asked me but I don't really have enough experience with HBase to comment intelligently on this. If you're already happy with this, all you have to do is leave a +1.

sorry for the late response, i would love to do the review :)