Download the release on GitHub.
Features
Section titled “Features”Enable selective deletion of lookup table entries
Section titled “Enable selective deletion of lookup table entries”For lookup-table contexts, the new --erase option for context update
enables selective deletion of lookup table entries.
The context update operator now defaults the --key <field> option to the
first field in the input when no field is explicitly specified.
By @dominiklohmann in #4274.
Add percentile aggregation functions
Section titled “Add percentile aggregation functions”The p99, p95, p90, p75, and p50 aggregation functions calculate
commonly used percentiles of grouped values in the summarize operator.
By @dominiklohmann in #4273.
Support multiple publishers
Section titled “Support multiple publishers”The publish operator’s topics no longer have to be unique. Instead, any number
of pipelines may use the publish operator with the same topic. This enables
multi-producer, multi-consumer (MPMC) event routing, where streams of events
from different pipelines can now be merged back together in addition to being
split.
Inter-pipeline data transfer with the publish and subscribe operators is now
as fast as intra-pipeline data transfer between pipeline operators and utilizes
the same amount of memory.
Back pressure now propagates from subscribers back to publishers, i.e., if a
pipeline with a subscribe operator is too slow then all pipelines with
matching publish operators will be slowed down to a matching speed. This
limits the memory usage of publish operators and prevents data loss.
By @dominiklohmann in #4270.
Changes
Section titled “Changes”Add percentile aggregation functions
Section titled “Add percentile aggregation functions”The approximate_median aggregation function is now called median. We found
the longer name, despite being more accurate, to be rather unintuitive.
By @dominiklohmann in #4273.
Bug Fixes
Section titled “Bug Fixes”Handle loading of configured and non-configured contexts with the same name
Section titled “Handle loading of configured and non-configured contexts with the same name”Configured and non-configured contexts with the same name will not cause non-deterministic behavior upon loading anymore. The node will shut down instead.
Evaluate ip == subnet predicates
Section titled “Evaluate ip == subnet predicates”Predicates of the form ip == subnet and ip in [subnet1, subnet2, …] now
work as expected.
The lookup operator now correctly handles subnet keys when using the --retro
or --snapshot options.
By @dominiklohmann in #4268.