Hi everyone, TLDR: I propose we hold and re-roll 4.1+ releases under vote due
to a newly-identified critical data loss bug. On Saturday, a critical bug
surfaced dating back to 4.1.6 that requires an urgent release from the project:
https://issues.apache.org/jira/browse/CASSANDRA-21508 : Coordinator
load-shedding returns OverloadedException without setting streamId, misrouting
query responses In short, requests that trip native_transport_timeout result in
responses being sent to clients on the incorrect streamId. This causes query
responses to become incorrectly associated with different queries. Clients may
observe deserialization failures, nonsensical responses, or responses
containing data from other queries. There is no reasonable config mitigation to
guard against this issue. It requires urgent attention from the project, and
patches will be available for 4.1.x, 5.0.x, 6.0, and trunk shortly. Please see
the ticket for details. There is a second less-critical but still important
corruption bug identified in SAI. This issue is:
https://issues.apache.org/jira/browse/CASSANDRA-21516 : SAI Component Checksum
Validation Should be Segment-Aware This issue can result in SAI indexes
reporting as corrupt and unqueryable due to incorrect checksum validation.
However the indexes can be rebuilt so the impact is availability rather than
corruption / data loss. Patches should be committed for both of these issues
within the week. – Scott
