Release Notes

v4.5.1

New and Optimized Features

  • Operator rebased on Altinity clickhouse-operator 0.27.4. The operator and metrics exporter are now built from the upstream 0.27.4 release instead of 0.20.0, bringing the upstream fixes and features between those versions. The package name, the clickhouse.altinity.com/v1 API group and existing instances are unchanged.
  • Operator-managed Keeper quorums. The new ClickHouseKeeperInstallation resource (short name chk, API group clickhouse-keeper.altinity.com/v1) deploys and manages a ClickHouse Keeper quorum. The operator supplies the server image, the Keeper configuration, a Service per member, a client Service and a PodDisruptionBudget, and runs the pods under the Pod Security Admission restricted profile. Keeper runs from the ClickHouse server image the product already ships, so no image is added. See Deploy a Keeper quorum.
  • Refer to a quorum by name. A ClickHouseInstallation can name a ClickHouseKeeperInstallation under spec.configuration.zookeeper.keeper instead of listing member addresses under zookeeper.nodes. The operator waits for the quorum to be ready and writes the members' addresses into the ClickHouse® configuration. zookeeper.nodes remains supported for quorums you run yourself.
  • ClickHouse® engine 26.8.39.19. The default server image moves from 25.8.33.6 (25.8 LTS) to 26.8.39.19 (26.8 LTS). New instances and ClickHouseKeeperInstallation quorums use 26.8.39.19 by default; existing instances that pin a server image in their pod template keep that image until you update it, as described in Upgrade. On arm64, the server image packages the arm_v80compat (armv8.0-a+crc) build for compatibility with ARMv8.0+ processors including Huawei Kunpeng 920.
  • Runtime base and security updates. The clickhouse-server runtime base image moves from Ubuntu 20.04 (glibc 2.31) to AlmaLinux 10.3 Minimal (glibc 2.39), and the clickhouse-log sidecar image moves from busybox to package-managed alpine:3.23.3. The operator and metrics exporter are compiled with Go 1.26.11 on Alpine 3.23.3 with updated Go dependencies including golang.org/x/net 0.62.0 and OpenTelemetry 1.47.0.

Upgrade Notes

  • Upgrading the operator restarts ClickHouse pods once. The new operator renders host configuration and StatefulSets differently, so every instance is updated with one rolling restart, one host at a time, and keeps its data volumes. Instances with a single replica are unavailable while their pod restarts. See Upgrade.
  • Hand-built Keeper quorums are not affected. Instances that list a quorum's members under zookeeper.nodes keep working. Moving to a ClickHouseKeeperInstallation is optional and described in Deploy a Keeper quorum.
  • Rolling the operator back to v4.4.x is not supported once a ClickHouseKeeperInstallation exists.

v4.4.0

New and Optimized Features

  • New product name. The plugin is now listed as Alauda Data Services Analytical Database E1. The change is to display names only: the package name, the custom resource kinds and API group, and existing instances are unchanged.
  • ClickHouse® engine 25.8.33.6. The default server image moves from 25.8.16.34 to 25.8.33.6, a patch release on the same 25.8 LTS line. New instances use it by default; existing instances keep the image they specify until you change it, as described in Upgrade.
  • Rebuilt images. The operator and metrics exporter are compiled with an updated Go toolchain on an updated Alpine base, and the busybox helper image is pinned to a fixed tag instead of a floating one. The ClickHouse server image includes the GNU C Library security updates published for its base up to 2.31-0ubuntu9.18, on both amd64 and arm64.
  • Image provenance. The operator, metrics exporter, and bundle images carry org.opencontainers.image.source, org.opencontainers.image.revision, and org.opencontainers.image.ref.name labels identifying the source they were built from.
  • Memory-aware operator runtime. The operator and metrics exporter set the Go runtime's soft memory limit from their container memory limit, which reduces the risk of the operator being OOM-killed on large installations.
  • Licence notices. Files modified from the upstream project carry modification notices, and the licence text is included in the operator, metrics exporter, and server images.

Behaviour Changes

  • Instance status now reflects pod readiness. An instance is marked Completed only when every host's StatefulSet reports its desired replicas as ready, and it returns to InProgress if a host later loses readiness. When the pods become ready again, the instance moves back to Completed on its own; you do not need to change the instance to clear the status. After upgrading, an instance whose pods are unhealthy reports InProgress where earlier releases reported Completed. See Instance reports Completed while pods are unhealthy.
  • The operator requires a server image to be configured. The operator no longer falls back to a built-in server image reference when none is configured; it reports a configuration error at startup instead. Installations from the plugin package always configure the image, so they are not affected.

Upgrade Notes

  • Upgrading the operator restarts ClickHouse pods. The new operator changes the default server image and the shared logger configuration, so every instance that does not specify its own server image is updated with a rolling restart, one replica at a time. Instances with a single replica are unavailable while their pod restarts. Plan the operator upgrade for a maintenance window if you run single-replica instances.

Fixed Issues

  • ClickHouse could not start as a non-root user. The default logger configuration pointed the error log at /dev/null. The server derives its log directory from that path and tried to manage /dev, which fails without root privileges, so an instance whose pods run as a non-root user — including any instance in a namespace that enforces the Pod Security Admission restricted profile — never started. The error log is now written to /var/log/clickhouse-server/clickhouse-server.err.log.
  • Instances reported Completed while their pods were failing. The reconcile pass was marked complete once all objects had been submitted, without waiting for pods to become ready, and a status that had reached Completed was never revised when pods later failed. Both cases are fixed; see Behaviour Changes above.

Known Issues

  • Instances that set logVolumeClaimTemplate are rejected under the Pod Security Admission restricted profile unless you define the log container yourself. The operator adds a clickhouse-log container whose security context sets only a user ID. Define a container named clickhouse-log with a compliant security context in your pod template. See Pod rejected under Pod Security Admission.
  • Instances with a memory limit below 2 GiB can reject small queries. ClickHouse caps its own memory use at 90% of the container memory limit, and an idle server's resident memory can approach 1 GiB. With a 1 GiB limit, the server can exceed its cap while idle and fail even trivial queries with Code: 241 … (total) memory limit exceeded. Set resources.limits.memory of the clickhouse container to at least 2Gi. If you do not set a limit, the container inherits the default memory limit of the namespace's LimitRange, which can be below 2 GiB. This also affects earlier releases.
  • Deleting an instance while the operator upgrade updates it takes longer than usual. After you upgrade the operator to this release, instances that do not set a server image are updated to the new server image one host at a time. If you delete such an instance during this update, the deletion starts only after the update finishes on all hosts, which can take several minutes per host. The deletion then completes.

ClickHouse is a registered trademark of ClickHouse, Inc. https://clickhouse.com

Alauda is an independent vendor. This product is not affiliated with, endorsed by, or sponsored by ClickHouse, Inc. All trademarks are the property of their respective owners and are used here for identification purposes only.