GCP Release Notes: July 28, 2025

Apigee X

Announcement

On July 28, 2025, we released an updated version of Apigee (1-15-0-apigee-7).

Feature

Server-sent events and EventFlows are Generally Available (GA)

Apigee supports continuous response streaming from server-sent event (SSE) endpoints to clients in real time. The Apigee SSE feature is useful for handling large language model (LLM) APIs that operate most effectively by streaming their responses back to the client. SSE streaming reduces latency, and clients can receive response data as soon as it is generated by an LLM. This feature supports the use of AI agents that operate in real time environments, such as customer service bots or workflow orchestrators. For more information, see Streaming server-sent events.

Streaming from SSE endpoints is available in Apigee and in Apigee hybrid v1.15.0 and newer.

Fixed

BigQuery

Libraries

A weekly digest of client library updates from across the Cloud SDK.

Feature

You can now associate data policies directly on columns. This feature enables direct database administration for controlling access and applying masking and transformation rules at the column level. This feature is in Preview.

Bigtable

Libraries

A weekly digest of client library updates from across the Cloud SDK.

Cloud Composer

Fixed

Fixed an issue that caused unexpected restarts of Airflow component workloads
in the environment’s cluster.

Fixed

(Cloud Composer 3) The DAGS_FOLDER reserved environment variable now
correctly points to the local directory where DAG files are stored.

Changed

New Airflow builds
are available in Cloud Composer 3:

• composer-3-airflow-2.10.5-build.10 (default)
• composer-3-airflow-2.9.3-build.30

Changed

New images
are available in Cloud Composer 2:

• composer-2.13.8-airflow-2.10.5 (default)
• composer-2.13.8-airflow-2.9.3

Deprecated

Cloud Composer version 2.8.6 has reached its
end of support period.

Cloud Load Balancing

Feature

Global external Application Load Balancers now support the JA4 fingerprint. The JA4 fingerprint can be added to a custom request header using the tls_ja4_fingerprint variable.

This capability is now in General Availability.

Gemini Code Assist

Feature

Inline diff (Preview)

VS Code Gemini Code Assist 2.42.0

Inline diff
is available for VS Code Gemini Code Assist 2.42.0 in
Preview. With the inline diff
view, you can visually highlight code changes directly in your editor—green for
additions, red for deletions. This feature boosts productivity by offering
immediate visibility on modifications, streamlining reviews, and enabling
precise control over integrating new code, all without leaving your current
view.

If preferred, you can
change the diff view settings to have a side-by-side diff window
instead of the inline diff.

Google Cloud Architecture Center

Feature

(New guide) Well-Architected Framework: Financial services industry (FSI) perspective: Principles and recommendations that are specific to FSI, aligned to each pillar of the Architecture Framework.

Google Kubernetes Engine

Feature

In GKE version 1.33.2-gke.1335000 and later, the
GKE Gateway controller
supports
Gateway API v1.3 CRDs.

Feature

In GKE version 1.33.1-gke.1788000 and later, you can target specific reservation
sub-blocks in a reservation block by using the
reservationSubBlock field in compute classes.

Feature

In GKE version 1.32.2-gke.1359000 and later, you can now configure
collection scheduling
for single-host and multi-host TPU node pools by using
compute classes.
Collection scheduling lets you set a Service Level Objective (SLO) for your TPU
workloads.

Feature

In GKE version 1.33.2-gke.1335000 and later, the
GKE Gateway controller
supports
Gateway API v1.3 CRDs.

Feature

In GKE version 1.33.1-gke.1788000 and later, you can target specific reservation
sub-blocks in a reservation block by using the
reservationSubBlock field in compute classes.

Feature

In GKE version 1.32.2-gke.1359000 and later, you can now configure
collection scheduling
for single-host and multi-host TPU node pools by using
compute classes.
Collection scheduling lets you set a Service Level Objective (SLO) for your TPU
workloads.

Announcement

Control plane datastore maintenance

Starting in May, 2025, Google is performing maintenance on the internal control
plane datastore for all GKE clusters to improve scalability and
reliability. We expect to complete these improvements across GKE
by October, 2025.

This maintenance is happening gradually across all GKE clusters,
and will occur in your clusters only during configured maintenance windows. The
maintenance process is expected to take approximately 15 minutes to complete
during your cluster’s maintenance window.

Expected impact

During the internal control plane datastore maintenance, the
Kubernetes API server will be unavailable for 15 minutes, regardless of
whether you use a regional cluster or a zonal cluster. During this 15-minute
period, you won’t be able to interact with the Kubernetes API server for your
cluster.

Consider the following potential disruptions to your normal workflows during the
maintenance window for your cluster:

• Kubernetes API unavailability: you can’t use the kubectl tool or any
  other Kubernetes API client to issue commands to the control plane,
  regardless of whether the cluster is regional or zonal. Attempts to deploy,
  modify, or query resources by using the Kubernetes API will fail during this
  period.
• Halted deployments: automated deployment pipelines (CI/CD) that interact
  with the Kubernetes API will fail to complete tasks such as deploying or
  updating applications in the cluster.
• Google Cloud console limitations: operations for the cluster in the
  Google Cloud console that communicate with the Kubernetes API might fail
  during the maintenance period.
• Delayed control plane automation: features that are managed by the
  control plane, such as the cluster autoscaler, Horizontal or Vertical
  Pod Autoscaling adjustments, or some node auto-repair operations might be
  paused until the API server is online.

The following resources have no expected impact during the maintenance period:

• Running applications: any running applications and services on your
  nodes should continue to function without interruptions.
• Node pool operations: existing nodes should remain connected and
  operational.
• Network traffic: traffic in the data plane, such as traffic to and from
  your running workloads, shouldn’t be affected.

What you need to do

No action is required from you for the maintenance to occur. To plan for this
maintenance, we recommend that you do the following:

• Review maintenance windows: review your cluster’s
  maintenance window and exclusions settings
  and schedule maintenance windows during periods that minimize disruptions to
  your normal workflows.
• Plan for Kubernetes API unavailability: if you run critical operations
  in your cluster that require access to the Kubernetes API, avoid scheduling
  these operations during maintenance windows.

  Fixed

  A fix is available for an issue in which the Compute Engine Persistent Disk CSI
  driver failed with an invalid cpuString error on GKE nodes that used custom
  machine types. This issue prevented successful attachment and mounting of
  Persistent Disk volumes on affected nodes. The fix is available in the following
  GKE versions:

  • 1.31.10-gke.1021000 and later
  • 1.32.4-gke.1698000 and later
  • 1.33.1-gke.1386000 and later
    

    SAP on Google Cloud

    Announcement

    New SAP NetWeaver certification: C4D bare metal machine types

    For use with SAP NetWeaver, SAP has certified the following Compute Engine bare metal machine types: c4d-standard-384-metal and c4d-highmem-384-metal.

    For more information, see the following:

    • Certifications for SAP applications on Google Cloud
    • C4D machine series

    Security Command Center

    Changed

    Model Armor filter updates

    • The prompt injection and jailbreak detection filter now supports 10,000 tokens.
    • For the Sensitive Data Protection filter, SKIP_DETECTION is returned if the prompt or response exceeds the token limit.
    • For all other filters, MATCH_FOUND is returned if malicious content is found, and SKIP_DETECTION is returned if no malicious content is found.

Source: Google Cloud Platform