Pull Hive-specific ConnectorSplit properties out of the base class

[hdikeman]

Description

There are two properties in the ConnectorSplit which are only used for Hive:

• cacheable: turns on/off retention in the data cache for Hive. Only accessed via HiveConnectorSplit objects
• splitWeight: used for scheduling, but only set to nonzero for Hive. Accessed via base ConnectorSplit objects

Some connectors (Tpch, Tpcds) set one or both of these, but never use them. Preferably we would remove these from the base class and place them in the Hive split implementation instead

cc @mbasmanova

[mbasmanova]

CC: @Yuhta @ericyuliu

[Yuhta]

splitWeight is not Hive specific, it's a scheduling functionality.
cacheable is for interaction with caching layer, not specific to Hive either.

[hdikeman]

splitWeight is not Hive specific, it's a scheduling functionality.
cacheable is for interaction with caching layer, not specific to Hive either.

@Yuhta it appears that both are only used by HiveConnectorSplit and some derivatives (Iceberg, Cudf)

• cacheable is only used internally, in HiveConnectorUtil, here:

  velox/velox/connectors/hive/HiveConnectorUtil.cpp

  Line 631 in 72ffdce

  readerOptions.setNoCacheRetention(!hiveSplit->cacheable);

• splitWeight is only ever modified from the default for HiveConnector and related splits, the rest use the default weight and do not allow setting a nonzero weight

This came up from a discussion regarding ConnectorSplit serdes, since it's not clear what to do with these base class fields when the [X]ConnectorSplit constructor does not allow setting them

[xiaoxmeng]

splitWeight is not Hive specific, it's a scheduling functionality. cacheable is for interaction with caching layer, not specific to Hive either.

splitWeight is used by task scheduling and recorded in task stats: runningTableScanSplitWeights but not sure if this is for stats report only? If it is, then we can move to hive split as well. cc @spershin