[RELEASE] ScyllaDB 5.1 RC3
Tzach Livyatan
<tzach@scylladb.com>The Scylla team is pleased to announce ScyllaDB Open Source 5.1 RC3, the second Release Candidate for the Scylla Open Source 5.1 minor release.
We encourage you to run ScyllaDB 5.1 release candidates on your test environments; this will help ensure that an upgrade to ScyllaDB 5.1 General Availability will proceed smoothly with your workload. Use the release candidate with caution; RC3 is not production-ready yet. You can help stabilize Scylla Open Source 5.1 by reporting bugs here.
Only the last two minor releases of the ScyllaDB Open Source project are supported. Once Scylla Open Source 5.1 is officially released, ScyllaDB Open Source 5.1 and 5.0 will be supported, and ScyllaDB 4.6 will be retired.
For a complete description of ScyllaDB 5.1 see ScyllaDB 5.1 RC1.
Related Links
* ScyllaDB 5.1 RC1 , RC2
* Get ScyllaDB Open Source 5.1 (under “More Versions” for each distro)
Updates and bug fixes since 5.1 RC2
offline-installer: cqlsh is failing to run from a user which isn't root #11584
Stability: 2 Nodes failed after bloom_filter_fp_chance was modified to an unreasonable low number #11524
Alternator: DescribeTable calls should include ProvisionedThroughput #11222
Stability: heap-use-after-free hit in cross_shard_barrier_test (unlikely to happened outside this test) #11303
Stability: SStables in maintenance set will be forgotten if user stopped off-strategy compaction. This bug only affect when using repair base node operations (RBNO) #11543
Stability: Accidentally quadratic behavior when a large number of range tombstones is present in a partition has been fixed. #11211
Stability: sstables crawling reader can emit duplicated partition end. This issue affects scylla's scrub compaction (in all modes) as well as the validate operation of the scylla-sstable tool. #11421
CQL API: ScyllaDB doesn't send "per-partition rate limit exceeded exception" even if the driver supports it #11517. Per partition rate limit is a new feature introduced in ScyllaDB 5.1
Stability: shard_reader can leave partitions unclosed when buffer-copying loop yields #11561. Fixing a regression introduced in ScyllaDB 5.1.
Stability: Messaging service misconfigured tcp_nodelay for verbs. The result is gossiper messages are coming with increased latency. This may affect the nodes status visibility of each other and, which is worse, there's an increased chance of one node marking the alive peer as dead #11465