Kubernetes Agent v1.9.29

Includes Java Agent v2.2.68

September 30, 2026

Caution

Upgrade the master agent and node agent to the 1.9.29 image together. The default collection intervals, resource requests and limits, packet schema, and RBAC are unchanged.

• Feature Added support for collecting native sidecar containers

  • Recognizes a container in spec.initContainers with restartPolicy: Always as a long-running native sidecar and includes it in the container list and in CPU and memory collection.
  • Applies the same criteria to the node agent's collection cache and pod aggregation, and to the master agent's container metadata and statistics.
  • Reflects restart counts, readiness, OOM information, and volume mount metadata.
  • Resource request and limit totals are based on the declared values of long-running containers.
  Note

  Regular init containers and ephemeral containers are still excluded from continuous collection. This isn't a change to scheduler-effective resource calculation that accounts for the startup order of regular init containers, or to Envoy L7 and APM instrumentation.

• Changed Expanded the configurable range of the node agent collection interval

  • Lowered the minimum value of count_interval and collect_interval from 5000 ms to 1000 ms. The default remains 5000 ms for both options.
  • Set both options in 1000 ms steps within the 1000–30000 ms range. Setting both to 2000 requests a 2-second collection and transmission interval, and setting both to 1000 requests a 1-second interval.
  Caution

  1 second and 2 seconds are requested intervals. Because of source read latency and asynchronous processing, not every metric is guaranteed to complete with a new value on each interval, and the server storage and screen aggregation resolutions are separate. Before lowering the interval, check CPU, throttling, and transmission volume against the scale of your nodes and containers.

• Changed Optimized the collection load of node helper

  • Serializes cgroup v1 blkio file reads within the process to reduce kernel lock contention between containers.
  • Parses only the CPU totals from /proc/stat during cgroup performance collection, without converting the per-CPU and interrupt rows.
  • Removed duplicate network reads and extra JSON serialization from containerd stats.
  • Excludes clear non-targets first during process collection, and falls back to the existing path when a read fails or a name is ambiguous.
  • Doesn't reduce the number of metric reads or the collection scope, and doesn't reuse previous performance values.

Info

Agent download: kube_mon_1.9.29.tar