Skip to content

Release notes

This page lists release notes for KX Sensors, in reverse chronological order.

When you upgrade, read the upgrade notes for every version between the one you're on and the one you're moving to. A release without an Upgrade notes section needs no action. For the upgrade procedure itself, see Before you begin.

3.3.0

Release date: 2026-09-23

Upgrade notes

  • The version of RT changes in v3.3. See RT upgrades.
  • Moving RT out of Docker is a manual part of the 3.2 to 3.3 upgrade: stop every node in the cluster, perform a hard reset of RT, and delete the RT Docker containers by hand. Because a hard reset discards RT log and state data, back up the contents of emsLogDir on each node first. You can remove the Docker Swarm overlay network and uninstall Docker at any point during or after the upgrade. See Move RT from Docker to the host.
  • The rtLogDir, rtSeqDir, and rtSesDir system parameters are removed. All RT log and state directories now derive from emsLogDir. Remove these entries from your systemParams.yaml when you upgrade. See RT log files.
  • rtoRetryFreq is renamed to rtoRetryInterval, with its default changing from 1000 to 2500, and rtoRetryLimit to rtoMaxRetries, with its default changing from 10 to 25. Rename these entries in your systemParams.yaml when you upgrade.
  • The rtoDockerLogMaxSize, rtoDockerLogMaxFile, and rtoDockerLogCompress system parameters are removed, since RT logs now follow normal KX Sensors application-log rotation. Remove these entries from your systemParams.yaml when you upgrade.
  • The network field in ems.yaml is deprecated and no longer used. No action is required: leave it in place and blank, since it's retained to avoid problems during dynamic upgrade.
  • RT now listens on host ports. The range from KXS_BASE_PORT + 800 to KXS_BASE_PORT + 1000 is reserved for RT, with 30 ports allocated per stream; make sure these are free and permitted by your firewall rules before you upgrade. The Docker Swarm ports that the RT overlay network required — 2377/TCP, 4789/UDP, and 7946/TCP — are no longer needed. See Network requirements.
  • If any EMS stream enables TLS, the KX_SSL_CA_CERT_FILE, KX_SSL_CERT_FILE, and KX_SSL_KEY_FILE variables must point to certificate files that exist on the host. RTO validates them at startup and exits without restarting if any are missing. See Configure TLS for ingestion from an external client through RT.
  • v3.3 includes required changes to the KX-supplied packages. Review your custom packages against the updated versions and merge the changes before you upgrade.
  • If a package in your deployment enables REST by setting isREST to true, make sure the REST port is free before you upgrade. By default the REST bridge listens on the SGW process port + 6; override it by setting KXS_REST_PORT in kxsenv.
  • Row-level retention is off by default, so no action is required on upgrade. Before you enable dbmRetention, check that each table's retentionDuration is shorter than the whole-partition retention that applies to it, otherwise the partition is deleted before the clause is ever applied. See Row-level retention.

KDB-X

Upgrades the underlying database from kdb+ 4.1 to KDB-X, which provides the module framework that future KX Sensors releases build on. Your existing license continues to work, but depending on the features you use, it may need to be reissued with additional feature flags.

REST

Reinstates REST support, which exposes your public APIs as JSON-over-HTTP endpoints without requiring changes to the APIs themselves. Endpoints are generated automatically from the SAPI registry and hosted by the SGW service class. REST is disabled by default; enable it by setting isREST to true in systemParams.yaml. See Call APIs over REST.

RT without Docker

RT no longer runs as a Docker container, so Docker and the Docker Swarm overlay network are no longer prerequisites for multi-node deployments. All RT processes now run directly on the host, and RTO remains responsible for starting, stopping, and monitoring them on each node.

RT is now a service class in its own right, so you inspect its processes with kxsctl status RT rather than with container tooling, and its application logs are written to the KX Sensors application log directory like those of any other process. The RT implementation ships as a q package instead of a container image, and RT listens on host ports rather than on container ports forwarded from the host.

Row-level retention

Adds row-level retention, which removes selected rows from aging HDB partitions instead of deleting whole partitions. Give a table a functional where clause and an age in days, and DBM prunes the rows the clause doesn't select once a partition reaches that age. You can apply different retention periods to groups of rows in the same table. For example, you can retain most readings for 100 days and a subset for 400 days.

Configure a policy with the new retentionDuration and retentionWC table properties in your schema files. A table can define several stages, each with its own age and clause. Row-level retention reuses the existing DBM tier-migration pipeline. A partition that needs both a tier move and a prune can have both applied in a single rewrite, and the same disk, memory, and EOD-yield safeguards apply. Enable it with the new dynamic system parameter dbmRetention.

kxsctl CLI enhancements

v3.3.0 includes several enhancements to kxsctl, the command-line interface for KX Sensors v3. For more information, see About kxsctl.

  • Three new commands. kxsctl refresh regenerates the systemd unit files from the current templates and manifest, kxsctl history shows the commands previously run through kxsctl on a node, and kxsctl manifest replaces kxsctl export and kxsctl import, adding the ability to edit a manifest interactively or programmatically.
  • kxsctl log replaces kxsctl show log. You can now view the logs of a process on any node, list the available logs for selection, and follow the current log across an hourly rollover. The -a/--age, -e/--no-less, and -n/--lines flags are removed.
  • kxsctl status is more concise and now colourised, highlights problem states, and includes disabled and inactive nodes rather than omitting them.
  • Enable, disable, start, stop, and restart behave consistently, and exit non-zero when a target never reaches the requested state. See Start and stop nodes individually and Start and stop service class instances.
  • Discovery errors now name the key and the underlying cause, and kxsctl show nodeState prints timestamps in a readable form with their time zone.
  • kxsctl-client reports TLS certificate errors and unreachable discovery more clearly, forwards kxsctl-client user arguments individually so that values containing spaces are preserved, and unpacks any .qpk archives found under a deps directory after a directory upload.
  • The -p and -n password shorthands are removed from kxsctl user and kxsctl-client user. Use --password and --new-password instead, and update any scripts that use the short forms.

Documentation site

KX Sensors product documentation is now published as a documentation site at code.kx.com/sensors, launching alongside this release. It replaces the Runbook that was previously distributed as a PDF.

The site is searchable and versioned, and it isn't tied to the product release cycle, so corrections and new content can be published as they're ready rather than waiting for the next release. Start at the Overview, or browse by section: Architecture, Develop, Configure and Deploy, and Operate.

3.2.11

Upgrade notes

There are no upgrade considerations when upgrading from v3.2.10.

Fixes and enhancements

  • Fixed message ID handling in external TPs to prevent messages from being incorrectly discarded as duplicates.
  • Updated SP request handling to send only one response when a request is deferred multiple times.
  • Added retries for requestSchemaUpgrade when a request times out.
  • Converted local timestamps to UTC in IdbStatus and HdbStatus.
  • Improved accounting of dynamic upgrade for in-memory delta partitions in partitionedMDB.q.
  • Updated Dynamic Upgrade to handle an edge case in a custom layer.
  • Prevented a domain error when missed EOD events occur after a dynamic upgrade that adds new MD tables.
  • Improved message handling after the buffer size is exceeded.
  • Removed direct reads of processVersion from disk after a dynamic upgrade.

3.2.10

Upgrade notes

There are no upgrade considerations when upgrading from v3.2.9.

Fixes and enhancements

  • Added an EOI offset to SM, avoiding disk contention when hosting multiple environments.
  • Fixed an RT sequencer issue that could cause it to skip ahead in the merged log.
  • Fixed updPubSub to prevent a regression of the loci table in DBW.
  • Added a default init column for migrated nxDbTbls during v2-to-v3 upgrades.
  • Changed the default of ctx.pub to false, correcting behavior in SPs.
  • Improved the getData API.
  • Updated the system commands used by RTO to support RHEL 10.

3.2.9

Upgrade notes

There are no upgrade considerations when upgrading from v3.2.8.

Fixes and enhancements

  • Fixed the response for isSequenced requests that require deferral.
  • Fixed seedNode, which could leave unexpected state if the etcd write needed to be retried.
  • Fixed dynamic upgrade so it renames and deletes # files for nested columns, and so the revision no longer jumps from 1 to 3 after an upgrade.
  • Fixed DBC reverse-channel replies selecting a peer with first, added a retry to the DBC handshake, and updated DBC to use comm.
  • Fixed archiveLogs running before TP processes had started.

3.2.8

This patch merges in fixes relating to the Utilities vertical. The fixes originated in the v2 stream of the product and are ported to v3 where applicable.

Upgrade notes

There are no upgrade considerations when upgrading from v3.2.7.

Fixes and enhancements

  • Added the ktbl helper function.
  • .dbw.hdbtrimdates now operates on a static link directory.
  • Improved etcd.sh in the installer.
  • Enhanced ChannelSMP.
  • Optimized DBCONV: apl operations now run with peach, and adding new tables is faster.
  • Revived the getData API and fixed it for non-partitioned tables.
  • Added a filterCalcs hook to processReadings in dc.q.
  • Fixed VEE issues, including a deferral bug, MSGID not remaining an int, and a type error when an SP received an addition to the deferral buffer after the deferral had resumed.
  • TP-specific fields can now be configured at the stream level.
  • Fixed wdur being incorrectly reset to 0 in the low-priority Gateway queue, and corrected gwMaxDurWt tuning that could leave the Gateway blocked waiting for requests to finish.
  • Disabled retention-based archival everywhere.

3.2.7

Upgrade notes

  • The OpenSSL 1.0 dependency is factored out of kxs-core. The installer no longer requires or supports the -o flag; see the updated Deployment Guide.
  • If upgrading from v3.2.5 or earlier, the version of RT changes, which requires a rolling upgrade. See RT upgrades.

Fixes and enhancements

  • Fixed a bug introduced in v3.2.6 that could cause messages from external clients to be incorrectly discarded as duplicates in single-node deployments. The origin is now stamped on the TP message payload.
  • Fixed an EOD notify timeout that could cause long-running queries to fail around EOD.
  • Updated kdb+ to 4.1-2026.07.06.
  • Removed the OpenSSL 1.0 dependency from the SI library.

3.2.6

Upgrade notes

  • This release includes an updated RT version, which requires a rolling upgrade. See RT upgrades.

Fixes and enhancements

  • kxsctl-client now supports calling any native kxsctl command remotely, rather than a subset. See About kxsctl.
  • Various fixes and small feature requests.

3.2.5

Upgrade notes

  • The SAPI ai field type has been reverted to Object, matching its v2 behavior. This may require minor code changes in SAPI clients.

  • After upgrading from a previous v3 version, restart the DBW and MON service classes. This step won't be required for future upgrades.

    kxsctl restart DBW
kxsctl restart MON

Fixes and enhancements

  • Added the MONIDB and MONHDB service classes, which surface persisted monitoring data via getMonStats.
  • Fixed a bug in .wm.resvenqwmt that returned incorrect watermarks.
  • Processes stopped gracefully with kxsctl stop no longer restart on server reboot.
  • Optimized DBCONV and EOD processing.
  • Fixed reloadable scripts so they're correctly reloaded during a dynamic upgrade.
  • Aligned qetcd log rollover with KX Sensors q processes.

3.2.4

Fixes and enhancements

  • Improved nxdiags scripts to support retrieving logs for specific dates.
  • Disabled retention-based archival for EMS logs by default.
  • Made .dbw.runEpilog timing configurable.
  • Added handling for missing files in .dbw.fillhdbtsranges during v2-to-v3 upgrades.

3.2.3

Upgrade notes

  • tpLogDir and rtLogDir are simplified into a single system parameter, emsLogDir. Update existing entries in systemParams.yaml when upgrading from a previous v3 version.
  • Since the default value of isCompressed is now false, review your existing v3 compression settings in your schema.
  • Installer flags have changed: --master-key supplies the master key (used to encrypt the user credential database) from the command line instead of install.profile, and the installer prompts for one and offers to generate it if neither is provided. --root-user supplies root user credentials from the command line instead of install.profile, and the installer also prompts for credentials so you can avoid storing them in plain text.

Fixes and enhancements

  • Added getData API functionality.
  • Fixed minor EMS issues and configuration changes.
  • Added support for re-rooting database files.
  • Used the DBW thread count when eoxSubtaskThreads is null.
  • Changed the default value of isCompressed to false, matching v2.
  • Optimized DBCONV for adding new tables.
  • Avoided a redundant upsert for MRU.
  • Updated the installer to prompt for missing authentication variables.

3.2.2

TP+ enhancements

  • Added TLS support for encrypting data in transit.
  • Improved robustness against data log corruption.
  • Validated integration with single-instance HA.
  • Added metrics generation for monitoring.
  • Added support for backing up data logs.

CLI enhancements

  • Updated kxsctl-client help output to include previously missing commands.
  • kxsctl, kxsctl-client, and kxsctl-agent now correctly display their versions in help output and via the version subcommand or --version flag.

3.2.1

Fixes and enhancements only, including improvements to MDQ weighting, API timeout handling, multi-table filtering, and EMS ping frequency, plus reliability improvements to VEE message tracking.

3.2.0

TP+

Introduces TP+, a new service class that replaces Reliable Transport (RT) as the default EMS for single-node deployments. It's topologically similar to TP in v2, with enhancements including subscription to arbitrary points in the log, asynchronous log replay on startup, message deduplication, and the ability to pause and resume subscriptions. TP+ is for single-node deployments only; use RT for multi-node, non-HA topologies.

TLS encryption of etcd traffic

etcd traffic between nodes and clients can now be encrypted with TLS, including client certificate validation.

Scheduled tasks

Reinstates the ability to schedule tasks for execution by a native q process, useful for log archival and cleanup, and for custom tasks that need awareness of KX Sensors application state.

Strong data consistency

Queries can now request strong data consistency (all preceding writes guaranteed to be reflected in the result) using a flag, at the cost of some additional latency. By default, KX Sensors provides eventual consistency.

Query-only nodes

You can now provision query-only nodes that serve queries without ingestion. Query-only nodes still receive data in real time, can use different HDB storage tiers, and can be targeted with custom query routing.

Multicast alternative

A new service class, MC_BRIDGE, provides an alternative to multicast for networks where multicast isn't supported, such as some cloud environments. Multicast is still recommended by default where available.

Persistence of monitoring data

Monitoring data is now subscribed to and written by DBW, so it can be retained for analysis in production systems.

Installer update for MTU

The installer now accepts DOCKER_NETWORK_MTU in install.profile to set the correct MTU when creating the Docker network.

3.1.2

Pacemaker integration

Adds support for integrating KX Sensors with Pacemaker for single-instance HA deployments.

Documentation improvements

Closed several documentation gaps and consolidated related documents.

Fixes and enhancements

Added hourly rollover of application logs, new MDQ weighting parameters, and support for DBC on a single date, plus fixes to diffDB, sub-request timeouts, and Gateway routing when nodes are marked as failed.

3.1.1

kdb+ 4.1

Upgrades the underlying kdb+ version from 4.0 to 4.1.

Authentication

Reintroduces username and password authentication, configurable independently per process (qetcd, kxsctl-agent, and q processes such as the Gateway). Installation requires a master encryption key to encrypt the credential database, stored in etcd, and credentials for an initial admin user. Manage additional users with kxsctl user commands.

Fixes and enhancements

EOD can now be configured to use the local time zone, and log file timestamps can use a configurable time zone.

3.1.0

Encryption, compression, TLS, and REPL — all available in KX Sensors v2 — are reintroduced in v3.

Encryption

Supports encryption of on-disk HDB and IDB data (RDB data isn't eligible), based on kdb+ encryption with a password-protected master key. You can selectively encrypt individual tables, columns, table categories, database tiers, and the symbol table.

Compression

You can select individual table columns for compression to improve query performance.

TLS

Connections from SAPI clients to q processes, and between q processes, support TLS encryption. TLS isn't yet supported between RT nodes.

SAPI discovery without Go

SAPI's discovery layer is reimplemented natively in C, removing the Go-based etcd client library used previously (which was incompatible with C# SAPI and .NET on Linux). SAPI clients now send discovery requests to the qetcd process on each node instead of connecting directly to etcd, using the base port + 3 for non-TLS connections and the base port + 4 for TLS connections.

REPL

The REPL service class replicates selected tables in CSV format via file copy, for consumption by an external system such as a SQL database or Hadoop.

Bridge mode

External SAPI clients can now publish directly to RT bridge, eliminating the need for local log files and child publishing processes on the client side. Bridge mode is recommended by KX but disabled by default, since it shifts responsibility for handling out-of-order delivery, message loss, and duplication to the client application.

Topic filtering

RT subscribers can now filter messages by topic via a new sub_server process, instead of each node persisting a full copy of the merged log. This reduces disk and CPU usage and improves HA stability at high ingestion rates, at the cost of some additional network I/O.

3.0.10

Minor fixes across Dynamic Upgrade, the CLI, the Gateway, and Discovery.

3.0.9

Minor fixes, an optimization for Reliable Transport in single-node systems, and improved system health checks.

3.0.8

CLI error handling and reporting improvements; installer improvements for air-gapped environments; various fixes and a new EMS version.

3.0.7

Official support for upgrading from v2 to v3, and the diffDB tool for detecting and remediating data mismatches across nodes.

3.0.6

Reinstated Most Recently Used (MRU) tables, integrated with the SP EMS and deferral framework.

3.0.5

SP EMS integration with VEE and MQ for reliable message recovery during failover, and high availability support for the MQ service class.

3.0.4

Minor core and installer fixes, plus support for converting individual v2 q files, not just full packages, to the v3 format.

3.0.3

Introduced the SP EMS framework, a consistent interface for streaming processors such as SDL; completed the dynamic upgrade implementation; added support for splitting large master tables; and improved IDB purviews for high-ingestion-volume systems.