This short report brings to light some interesting commits to scylladb.git master from the last week. Commits in the 2e8e9f874f8..8066859d359 range are covered.
There were 131 non-merge commits from 24 authors in that period. Some notable commits:
The tablet load balancer now refreshes load statistics if it fails to rearrange them post split/merge, avoiding a load-balancing stall until the next refresh period.
The table size estimation logic, used by vnode-to-tablets migration to select the right tablet count, was fixed.
The partitioned_sstable_set data structure is used to hold sstables for run-oriented compaction strategies, (Incremental Compaction Strategy and Leveled Compaction Strategy). It now uses a blocked array rather than an interval map, for reduced memory consumption in the worst case.
A commitlog bug involving lost tracking of a commitlog segment under the conditions of a mutation that spans multiple segments and timeouts was fixed.
A table’s tablet hints now survive the loss of the node driving restore.
The Google Cloud Storage driver now correctly interprets a 308 Resume Incomplete response with no range header.
The tablet load balancer caps the concurrency of tablet streaming operation per shard. This is now also enforced for rebuilds due to replication factor changes.
The DESC modifier for a clustering key no longer leaks into CQL tuple comparisons.
Compaction sub-options are now validated for materialized views and indexes.
The automatic vnodes-to-tablets migration system was marked as generally available.
Indexed SELECT statements now re-validate the index key in the base table rows to avoid inconsistent responses due to replication lag.
The nodetool stop COMPACTION command now stops major compaction as well.
See you in the next issue of last week in scylladb.git master!