Cast between JSON and VARCHAR

Casts of both directions between JSON and VARCHAR have been removed. If you have such casts in your scripts or views, they will fail with a message when you move to release 0.116. To get the semantics of the current casts, use:

• JSON_PARSE(x) instead of CAST(x as JSON)
• JSON_FORMAT(x) instead of CAST(x as VARCHAR)

In a future release, we intend to reintroduce casts between JSON and VARCHAR along with other casts involving JSON. The semantics of the new JSON and VARCHAR cast will be consistent with the other casts being introduced. But it will be different from the semantics in 0.115 and before. When that comes, cast between JSON and VARCHAR in old scripts and views will produce unexpected result.

Cluster Memory Manager Improvements

The cluster memory manager now has a low memory killer. If the cluster runs low on memory, the killer will kill queries to improve throughput. It can be enabled with the query.low-memory-killer.enabled config flag, and the delay between when the cluster runs low on memory and when the killer will be invoked can be configured with the query.low-memory-killer.delay option.

General Changes

• Add multimap_agg() function.
• Add checksum() function.
• Add max() and min() that takes a second argument and produces n largest or n smallest values.
• Add query_max_run_time session property and query.max-run-time config. Queries are failed after the specified duration.
• Removed experimental.cluster-memory-manager-enabled config. The cluster memory manager is now always enabled.
• Removed task.max-memory config.
• optimizer.optimize-hash-generation and distributed-joins-enabled are both enabled by default now.
• Add optimization for IF on a constant condition.