This short report brings to light some interesting commits to scylladb.git master from the last week. Commits in the 88bb75b9229..9cda315bbab range are covered.
There were 82 non-merge commits from 22 authors in that period. Some notable commits:
Tablet-aware restore from backup overrides the tablet count hints to ensure we don’t need to split tablets. It now restores the original hints so the load balancer is free to create more tablets.
Repair failed to start on large tablets due to the load balancer generating too-large Raft commands. The load balancer now optimizes the commands so they don’t overflow the limit.
When the commitlog approaches its size limits, it flushes memtables that intersect its tail. Previously, it flushed memtables from all tablets if one intersected its tail. Now, it flushes only those memtables that actually intersect.
A problem with concurrent reads from different positions in Google Cloud Storage objects was fixed.
Logstor is a new, experimental storage engine for key/value workloads. It is now able to cache values (previously only keys were held in memory).
Tablet-aware restore can now be aborted
.
It is now possible to create vector indexes with partition keys that are different from the base table’s partition keys.
The topology change coordinator now prevents some mistakes related to the removenode and replace operations.
The tablet restore task now has progress tracking.
The CQL3 type system now assigns types to literals (e.g. “1”) even when their exact type cannot be inferred from context. This allows for more natural expressions.
See you in the next issue of last week in scylladb.git master!