Google Cloud release notes

The following release notes cover the most recent changes over the last 30 days. For a comprehensive list, see the individual product release note pages .

You can see the latest product updates for all of Google Cloud on the Google Cloud release notes page.

To get the latest product updates delivered to you, add the URL of this page to your feed reader, or add the feed URL directly: https://cloud.google.com/feeds/gcp-release-notes.xml

September 17, 2020

Anthos

Anthos 1.4.3 is now available.

Updated components:

Anthos GKE on-prem

Anthos GKE on-prem 1.4.3-gke.3 is now available. To upgrade, see Upgrading GKE on-prem. GKE on-prem 1.4.3-gke.3 clusters run on Kubernetes 1.16.11-gke.11.

Fixes:

  • Fixed CVE-2020-14386 described in Security Bulletin.

  • Preflight check for hostname validation was too strict. We updated the hostname validation following the RFC 1123 DNS subdomain definition.

  • There was an issue in the 1.4.0 and 1.4.2 releases where the node problem detector didn't start when the node restarted. This is fixed in this version.

BigQuery BigQuery Data Transfer Service Cloud Healthcare API

You can now configure a ParserVersion in an HL7v2 store, which allows you to set the version of the unschematized parser. A new parser version, V2, has been added, which allows parsedData to include every given non-empty message field. This version parses the MSH segment starting with the MSH-1 field and is in accordance with the HL7v2 standard.

Cloud Logging

In the Logs Viewer (Preview), you can now pin log entries within the Query results and Histogram panes. To learn more, go to Pinning logs.

In the Logs Viewer (Preview), you can now view a log entry in its resource context. To learn more, go to Viewing a pinned log entry in its resource context.

You can now copy a link to a log entry in the Logs Viewer (Preview). To learn more, go to Copy a link to a log entry.

Compute Engine

You can now migrate a VM instance from one network to another. This feature is available in Beta.

Identity and Access Management

The issue with undeleting service accounts has been resolved. You can now undelete most service accounts that meet the criteria for undeletion.

September 16, 2020

Compute Engine

Troubleshoot VMs by capturing a screenshot from the VM. This is Generally Available.

September 15, 2020

Cloud Load Balancing

Added total latency to external HTTP(S) load balancer Cloud Logging entries. Total latency measures from when the external HTTP(S) load balancer receives the first bytes of the incoming request headers until the external HTTP(S) load balancer finishes proxying the backend's response to the client. This feature is now available in General Availability.

Cloud SQL for MySQL

Cloud SQL now offers serverless export. With serverless export, Cloud SQL performs the export from a temporary instance. Offloading the export operation allows databases on the primary instance to continue to serve queries and perform other operations at the usual performance rate.

Cloud SQL for PostgreSQL

Cloud SQL now offers serverless export. With serverless export, Cloud SQL performs the export from a temporary instance. Offloading the export operation allows databases on the primary instance to continue to serve queries and perform other operations at the usual performance rate.

The following PostgreSQL minor versions have been upgraded:

  • PostgreSQL 9.6.16 is upgraded to 9.6.18.
  • PostgreSQL 10.11 is upgraded to 10.13.
  • PostgreSQL 11.6 is upgraded to 11.8.
  • PostgreSQL 12.1 is upgraded to 12.3.
Compute Engine

SSD persistent disks attached to certain VMs with at least 64 vCPUs can now reach 100,000 write IOPS. To learn more about the requirements to reach these limits, see Block storage performance.

September 14, 2020

Cloud CDN

Cache Modes, TTL overrides and custom response headers are now supported on backend buckets and backend services, and are available in beta.

Cache modes allow Cloud CDN to automatically cache static content types, including web assets like CSS, JavaScript and fonts, as well as image and video content.

TTL overrides support fine-tuning how long Cloud CDN caches your responses, and custom response headers introduce a new {cdn_cache_status} variable that is populated with the cache status response.

External HTTP(S) Load Balancing now supports setting custom response headers on backend buckets and services. This feature is available in beta.

Custom response headers make it easier to set common web security headers and/or override response headers from your application at the load balancer.

Cloud Load Balancing

The External HTTP(S) Load Balancer now supports setting custom response headers on backend buckets and services. This feature is available in Beta.

Custom response headers make it easier to set common web security headers and override response headers from your application at the load balancer.

Cloud Logging

Cloud Logging now offers the ability to view a history of your ran queries through the Recent queries tab. To learn more, go to the Recent queries section on the Building queries page.

Compute Engine

Compute-optimized (C2) machine types are now available in Sydney, Australia australia-southeast1-a. See VM instance pricing for details.

Datastore

The gcloud datastore index create and gcloud datastore index cleanup commands now require gcloud version 279.0.0 or greater. To update the gcloud CLI, use the gcloud components update command.

Migrate for Compute Engine

There is no longer a requirement that the subnet of the deployment cluster is under the same network as the Cloud Extension.

Video Intelligence API

Bug fix for shot change detection API: Tuned internal model parameters to reduce false positives under certain scenarios.

September 11, 2020

BigQuery

You can now use the BQ.JOBS.CANCEL system procedure to cancel a job. For more information, see Canceling jobs.

Cloud TPU

Compute Engine TPU Metrics and Logs In-Context

New Monitoring tab for TPUs provides key TPU Metrics and access to logs at a glance. You can see a variety of key TPU metrics including MXU utilization, CPU, memory, sent/received traffic, and more. In addition, it provides instant access to TPU logs which give insight into important events.

Compute Engine

You can build highly available deployments of stateful workloads on VM instances using stateful managed instance groups (stateful MIGs). A stateful MIG preserves the unique state of each instance (instance name, attached persistent disks, and/or metadata) on machine restart, recreation, autohealing, or update. Stateful MIGs are Generally available.

Dataproc

Added the PrivateIpv6GoogleAccess API field to allow configuring IPv6 access to Dataproc cluster.

New sub-minor versions of Dataproc images: 1.3.68-debian10, 1.3.68-ubuntu18, 1.4.39-debian10, 1.4.39-ubuntu18, 1.5.14-debian10, 1.5.14-ubuntu18, 2.0.0-RC10-debian10, and 2.0.0-RC10-ubuntu18.

1.3-1.5 Images:

HIVE-18323: Vectorization: add the support of timestamp in VectorizedPrimitiveColumnReader for parquet

1.5 and 2.0 preview images:

Upgraded the jupyter-core and jupyter-client packages in the 1.5 and 2.0 images to be compatible with the installed notebook package version.

2.0 preview image:

Fixed a regression that could cause clusters to fail to start if user-supplied keystore/truststore are provided when enabling Kerberos.

Firestore

September 09, 2020

Cloud CDN Cloud Load Balancing Cloud Monitoring

The API for creating and managing alerting policies is now Generally Available. For information on using this API, see Managing alerting policies by API.

Cloud Spanner

Cloud Spanner introduces a new introspection tool that provides insights into queries that are currently running in your database. Use Oldest Active Queries to analyze what queries are running and how they are impacting database performance characteristics.

Identity and Access Management

You cannot undelete most service accounts at this time. Our engineering team is working to resolve this issue.

September 08, 2020

Cloud Data Loss Prevention

STREET_ADDRESS infoType detector is now available in all regions.

Cloud Healthcare API

Two new permissions, healthcare.locations.get and healthcare.locations.list, have been added to the IAM permissions. These permissions are checked whenever the new GetLocation and ListLocations methods are called.

Cloud Storage

New conditions available for Object Lifecycle Management:

  • Noncurrent time conditions allow you to define lifecycle actions based on when an object became noncurrent.
  • Custom time conditions allow you to define lifecycle actions based on a custom time that you define for objects.

Custom time metadata is now available for objects.

  • Custom time metadata allows you to associate an arbitrary date and time with objects.
Security Command Center

Security Command Center Premium is now in general availability (Container Threat Detection remains in beta). Read these notes to learn about updates, usability improvements, and new features.

Improved Summary Dashboard

  • A new set of interactive charts and tables provide a high-level overview of all threats and vulnerabilities.
  • An updated time selector lets you choose preset and customizable time ranges for reviewing findings and creating reports.
  • New page headers provide users with more page-specific context.

Learn more about Using the Security Command Center dashboard.

Onboarding and configuration upgrades

  • A streamlined interface lets you manage organization-wide service enablement settings.
  • A dedicated settings page for integrated services has been added to the configuration interface.

Learn more about Setting up Security Command Center.

Security Health Analytics now supports real-time detections, with some exceptions. Read more about Security Health Analytics detectors and findings.

Managed Web Security Scans are now available to all Security Command Center Premium users. Learn more about managed scans in our Overview of Web Security Scanner.

gcloud integration with new, simplified Beta APIs (Alpha)

  • The gcloud command line interface can now access configuration functionality through new Beta APIs. The Beta APIs provide stable, programmatic interaction equivalent in functionality to the Security Command Center interface. Learn to use gcloud to manage Security Command Center settings.

Documentation

September 04, 2020

Data Catalog

Data Catalog is now available in Jakarta (asia-southeast2).

Dataproc

Switched 1.3 and 1.3-debian image version aliases to point to 1.3 Debian 10 images.

When Enhanced Flexibility Mode is enabled, increased app master, task, and Spark stage retries to 10 to improve resiliency of applications to downscaling and preemption of preemptible VMs.

Support more than 8 local SSDs on VMs. Compute Engine supports 16 and 24 SSDs for larger machine types.

Changed secondary workers default boot disk size to 1000 GB in clusters created with 2.0 preview images.

Improved node memory utilization in clusters created with 2.0 preview images.

September 03, 2020

Config Connector

BigtableInstance: numNodes on resources is now optional. You can then programmatically scale your Bigtable instances. You cannot add the numNodes field after creating a BigtableInstance.

For production instances where the numNodes will be managed by Config Connector, this field is required with a minimum of 1. For a development instance or for an existing instance where the numNodes is managed outside of Config Connector, this field must be left unset.

Traffic Director

Traffic Director can now be set up for GKE Pods with automatic Envoy injection.

September 02, 2020

Cloud CDN

Reduced cache fill pricing from Cloud Storage, Compute Engine, and external origins by up to 80% starting September 1st.

Google Cloud has also removed cache-to-cache fill and cache invalidation charges.

Virtual Private Cloud

Firewall Rules Logging metadata controls is now available in General Availability.

September 01, 2020

Assured Workloads for Government

Assured Workloads for Government is now generally available for the FedRAMP Moderate compliance regime.

Cloud Logging

Cloud Logging updated IAM custom role permissions. Users with custom roles should verify their permissions are correct. For a list of the permissions and roles, go to the Permissions and Roles section on the Access control page.

Dialogflow

Beta launch of Dialogflow CX. Dialogflow CX provides a new way of designing agents, taking a state machine approach to agent design. This gives you clear and explicit control over a conversation, a better end-user experience, and a better development workflow.

As part of introducing Dialogflow CX, some changes have been made to editions:

  • Dialogflow Enterprise Edition is no longer used as an umbrella term for paid editions.
  • Dialogflow Standard Edition is now called Dialogflow Trial Edition.
  • Dialogflow Plus Edition and Dialogflow Essentials Edition have been merged to one Dialogflow Essentials (ES) Edition.
  • Dialogflow Customer Experience (CX) Edition is a new edition with a simpler pricing and quota model.

The service previously named Dialogflow is now called Dialogflow ES, and the term Dialogflow is now an umbrella term used to describe both the Dialogflow ES and Dialogflow CX services.

Filestore

Filestore Troubleshooting page published. Now you can troubleshoot common Filestore issues.

VPC Service Controls

Beta stage support for the following integration:

August 31, 2020

BigQuery

Updated version of Magnitude Simba JDBC driver includes bug fixes and enhancements such as automatically turning on the BigQuery Storage API for anonymous table reads (no charge for temporary table reads).

Cloud Load Balancing

Google Cloud internal HTTP(S) load balancers have native support for the WebSocket protocol when you use HTTP or HTTPS as the protocol to the backend. The load balancer does not need any configuration to proxy WebSocket connections.

Cloud SQL for SQL Server

Cloud SQL for SQL Server supports cloning using the Cloud Console, the gcloud command, and the Cloud SQL Admin API. When you clone an instance, you create an independent copy of the source instance.

Google Cloud Armor

Google Cloud Armor Managed Protection Plus Tier is in Beta. Managed Protection Plus Tier offers a monthly subscription that includes all of the features of Standard Tier, and bundles Google Cloud Armor WAF policy, rules, HTTP request usage, and named IP lists.

Pre-configured rules for local file inclusion (LFI), remote file inclusion (RFI), and remote code execution (RCE) are now in General Availability.

Network Intelligence Center

In the details pane for a connection, under View flows in BigQuery, Network Topology now provides a table that lists the subnets relevant to the currently selected connection and whether those subnets have VPC Flow Logs enabled. For more information, see Using generated queries to retrieve detailed logs.

August 30, 2020

Cloud Composer

August 28, 2020

AI Platform Prediction

Runtime version 2.2 is now available. You can use runtime version 2.2 to serve online predictions with TensorFlow 2.2.0, scikit-learn 0.23.1, or XGBoost 1.1.1. Runtime version 2.2 does not currently support batch prediction.

See the full list of updated dependencies in runtime version 2.2.

AI Platform Training

Runtime version 2.2 is now available. You can use runtime version 2.2 to train with TensorFlow 2.2.0, scikit-learn 0.23.1, or XGBoost 1.1.1. See the full list of updated dependencies in runtime version 2.2.

Cloud Data Loss Prevention

Risk analysis job creation is now available in the Cloud DLP UI in Cloud Console.

Added additional infoType detectors:

  • STORAGE_SIGNED_URL
  • STORAGE_SIGNED_POLICY_DOCUMENT
Cloud Monitoring

Cloud Trace exemplars can now be viewing in Cloud Monitoring. For more information about Trace exemplars, see Cloud Trace exemplars. For more information about viewing exemplars, see Exploring charted data.

Cloud Trace

Cloud Trace exemplars can now be viewing in Cloud Monitoring. For more information about Trace exemplars, see Cloud Trace exemplars. For more information about viewing exemplars, see Exploring charted data.

Dataproc

Launched Dataproc Workflow Timeout feature, which allows users to set a timeout on their graph of jobs and automatically cancel their workflow after a specified period.

Dataproc Metastore integration, which allows users to create a cluster using a Dataproc Metastore service as an external metastore, is now available for Alpha release testing.

Identity and Access Management

New features are available for Credential Access Boundaries, currently in beta:

  • You can now manage permissions for Cloud Storage objects, in addition to buckets.
  • You can now use IAM Conditions to control which permissions are available in a short-lived OAuth 2.0 access token. For an example, see Limit permissions for specific objects.
  • You can now use Credential Access Boundaries with a Cloud Storage bucket that does not use uniform bucket-level access.

For Credential Access Boundaries, currently in beta, you must migrate to a new API endpoint, sts.googleapis.com. To learn how to use the new API endpoint, see Exchanging the OAuth 2.0 access token.

Pub/Sub

Pub/Sub subscription detachment is now generally available.

August 27, 2020

AI Platform Optimizer

AI Platform Optimizer's name has changed to AI Platform Vizier.

Anthos Config Management

Anthos Config Management now includes Config Connector v1.15.1.

Anthos Policy Controller has been updated to include a more recent build of OPA Gatekeeper (hash: 1de87b6).

This release includes several logging and performance improvements.

An issue with git submodule support is preventing syncing of configuration stored in submodule repositories. If this affects you, please contact support so we can suggest ways to handle your required use cases while we correct this.

Anthos GKE on AWS

GKE on AWS 1.4.2-gke.1 is released. This release includes Kubernetes version 1.16.13-gke.1401.

This release includes bug fixes and security improvements. We recommend you update your clusters to this version.

To upgrade your clusters, perform the following steps:

  1. Upgrade your management service to aws-1.4.2-gke.1.
  2. Upgrade your user cluster's AWSCluster and AWSNodePools to 1.16.13-gke.1401.
Anthos Service Mesh

1.6.8-asm.9 is now available

Adds beta support for joining multiple clusters from different projects into a single Anthos Service Mesh on Google Kubernetes Engine.

Adds Citadel CA support for gcp profiles.

Fixes an issue for enabling trust domain validation at the transport socket level.

BigQuery ML

Time series model support is now Generally Available (GA). This release includes a new training option: AUTO_ARIMA_MAX_ORDER.

For more information about time series model support, see the following documentation:

Config Connector

Support referencing org-level IAM custom roles for IAMPolicy/IAMPolicyMember

Increase support for cross-project references

August 26, 2020

AI Platform Deep Learning VM Image

M55 release

  • Restricts Jupyter memory usage to fix 5* issues
  • Updates TensorFlow 2.3 dependencies
  • Uses CUDA 11.0 in TensorFlow deep learning images
  • Adds support for the us-east4 region
BigQuery

INFORMATION_SCHEMA views for BigQuery reservations are now Generally Available (GA).

Cloud Functions

Cloud Functions Node 10 runtime now builds container images in the user's project, providing direct access to build logs and removing the preset build-time quota.

Cloud Healthcare API

The v1beta1 API has begun updating to a revised version. The update will occur over the next several weeks. View Updates to the v1beta1 API for a list of changes and how to update your client code.

Resource Manager

The Organization policy constraint for Cloud NAT has launched into beta.

Traffic Director

Traffic Director supports advanced traffic management features with proxyless gRPC applications in General Availability. This release includes path- and header-based routing, as well as weight-based traffic splitting.

August 25, 2020

App Engine standard environment Go

The Go 1.14 runtime for the App Engine standard environment is now generally available.

BigQuery

INFORMATION_SCHEMA views for jobs metadata by timeslice are now available.

Cloud SQL for MySQL

Cloud SQL for MySQL 8.0 is now generally available. If you have not used MySQL 8.0, familiarize yourself with the What is new in MySQL 8.0 and Changes in MySQL 8.0.

For information about the implementation of MySQL 8.0 on Cloud SQL, see the following topics:

To get started using MySQL 8.0 on Cloud SQL, see Creating instances.

Identity and Access Management

Uploading public keys for service accounts is now generally available.

Speech-to-Text

Speech-to-Text has launched the new On-Prem API. Speech-to-Text On-Prem enables easy integration of Google speech recognition technologies into your on-premises solution.

Workflows

Beta launch of Workflows.

August 24, 2020

BigQuery

BigQuery support for using service account credentials with scheduled queries is Generally Available (GA).

Cloud Data Fusion

Highlights

Cloud Data Fusion 6.1.4 provides performance and scalability improvements that increase developer productivity and optimize pipeline runtime performance. The release includes scaled-up previews that support up to 50 concurrent runs, capabilities to handle large and complex schemas in Pipeline Studio, an enhanced log viewer, and other critical improvements and fixes.

This release is in parallel with the CDAP 6.1.4 release.

You can now create autoscaling Dataproc clusters.

You can now use the schema support feature in the UI to edit the precision and scale fields.

Cloud Data Fusion now has improved memory performance in pipelines by utilizing a disk-only auto-caching strategy.

Cloud Data Fusion previews now support up to 50 concurrent runs.

Cloud Data Fusion now supports large and deeply nested schemas (>5K fields with 20+ levels of nesting).

Fixed a bug where the setting for the number of executors in streaming pipelines was ignored.

Fixed a race condition where runtime monitoring failed when programs launched concurrently.

Fixed the preview page in table mode so that it shows multiple inputs and outputs with tabs.

Fixed the stability of state transitions for starting pipelines in AppFabric when AppFabric restarts.

Fixed a bug where a metric incorrectly counted the number of the records written in the Google Cloud Storage sink.

Cloud Logging

You can now view Cloud Trace information from within the Logs Viewer. To learn more, go to the Trace data section on the Using Logs Viewer (Preview) page.

Cloud Run for Anthos

Cloud Run for Anthos on Google Cloud version 0.16.0-gke.1 is now available for following GKE minor version: - 1.17 - 1.16

When Cloud Run is enabled on GKE new or existing cluster, version installed will be mapped to GKE master minor version. For more details see this.

Compute Engine

You can now protect your Compute Engine resources using VPC Service Controls. This feature is available in Beta.

Compute Engine committed use discounts are Generally Available for SUSE Linux Enterprise Server for SAP images. Learn more about discounted image pricing at Committed use discounts.

Document AI

Form Parser model updates

The Form Parser model has been updated. The model update includes the following features:

  • Improved OCR quality for English detection.
  • Improved key-value pair, checkbox, and table parsing detection quality, particularly for rotated images and handwritten text.
  • Decreased latency for complex tables.
Managed Service for Microsoft Active Directory

You can now create Managed Microsoft Active Directory instances using Terraform.

Security Command Center

Audit logs are now available in Security Command Center as part of Cloud Audit Logs. Learn more about Security Command Center audit logging.

Text-to-Speech

Text-to-Speech now offers four new English (US) voices, available as both WaveNet and Standard models. See the supported voices and languages page for more details.

Text-to-Speech now offers four new Chinese (Hong Kong) voices, available as Standard models. See the supported voices and languages page for more details.

August 21, 2020

Cloud Composer
  • New versions of Cloud Composer images: composer-1.11.3-airflow-1.10.3, composer-1.11.3-airflow-1.10.6, and composer-1.11.3-airflow-1.10.9. The default is composer-1.11.3-airflow-1.10.6. Upgrade your Cloud SDK to use features in this release.
  • The max size for serialized DAGs has been increased to 128 MB.
  • PyPi dependency conflicts will now cause environment updates to fail by default. This can now be changed by setting the environmental variable IGNORE_PYPI_DEPENDENCY_CONFLICTS=True . This is the default behavior For older environments, but you can set IGNORE_PYPI_DEPENDENCY_CONFLICTS=False to force conflicts to fail.
  • The Airflow configuration [core]load_examples can no longer be modified.
  • Domain Restricted Sharing setup with buckets in the tenant project is now applied during environment creation using the Beta API only when it is required.
  • Airflow 1.10.3, 1.10.6 and 1.10.9: Disabled editing DAG properties that will be overwritten when the DAG is reloaded.
  • Airflow 1.10.9:
    • Fixed an issue that caused BigQuery operators from backported providers to fail when DAG serialization was turned on.
    • The Airflow web server UI will now correctly reload when the plugin folder is modified.
    • Rendered template_fields can now be displayed without accessing DAG files.
  • Airflow 1.10.6: Fixed intermittent issues template rendering when DAG serialization is turned on.
Dataproc

Announcing the Beta release of Dataproc Enhanced Flexibility Mode (EFM), which manages shuffle data to minimize job progress delays caused by the removal of nodes from a running cluster.

New sub-minor versions of Dataproc images: 1.3.67-debian10, 1.3.67-ubuntu18, 1.4.38-debian10, 1.4.38-ubuntu18, 1.5.13-debian10, 1.5.13-ubuntu18, 2.0.0-RC9-debian10, and 2.0.0-RC9-ubuntu18.

Image 1.3 and 1.4: upgraded the Cloud Storage connector to version to 1.9.18.

Changed 1.4 and 1.4-debian image version aliases to point to 1.4-debian10. The version name 1.4-debian9 will continue to be available, but it won't get updates in future releases.

August 20, 2020

Anthos Anthos GKE on-prem

Anthos GKE on-prem 1.4.2-gke.3 is now available. To upgrade, see Upgrading GKE on-prem. GKE on-prem 1.4.2-gke.3 clusters run on Kubernetes 1.16.11-gke.11.

GPU support (beta solution in collaboration with Nvidia)

In partnership with Nvidia, users can now manually attach a GPU to a worker node VM to run GPU workloads. This requires using the open source Nvidia GPU operator.

Note: Manually attached GPUs do not persist through node lifecycle events. You must manually re-attach them. This is a beta solution and can be used for evaluation and proof of concept.

The Ubuntu image is upgraded to include the newest packages.

gkectl delete loadbalancer is updated to support the new version of configuration files for admin and user clusters.

Fixes:

  • Resolved a few incorrect Kubelet Metrics' names collected by Prometheus.
  • Updated restarting machines process during admin cluster upgrade to make the upgrade process more resilient to transient connection issues.
  • Resolved a preflight check OS image validation error when using a non-default vSphere folder for cluster creation; the OS image template is expected to be in that folder.
  • Resolved a gkectl upgrade loadbalancer issue to avoid validating the upgraded SeesawGroup. This fix lets the existing SeesawGroup config be updated without negatively affecting the upgrade process.
  • Resolved an issue where ClientConfig CRD is deleted when the upgrade to the latest version is run multiple times.
  • Resolved a gkectl update credentials vsphere issue where the vsphere-metrics-exporter was using the old credentials even after updating the credentials.
  • Resolved an issue where the VIP preflight check reported a user cluster add-on load balancer IP false positive.
  • Fixed gkeadm updating config after upgrading on Windows, specifically for the gkeOnPremVersion and bundlePath fields.
  • Automatically mount the data disk after rebooting on admin workstations created using gkeadm 1.4.0 and later.
  • Reverted thin disk provisioning change for boot disks in 1.4.0 and 1.4.1 on all normal (excludes test VMs) cluster nodes.
  • Removed vCenter Server access check from user cluster nodes.
App Engine standard environment Java

Updated Java SDK to version 1.9.82.

Cloud Spanner

A new multi-region instance configuration is now available in North America - nam11 (Iowa/South Carolina).

Cloud TPU

PyTorch/XLA 1.6 Release (GA)

Highlights

Cloud TPUs now support the PyTorch 1.6 release, via PyTorch/XLA integration. With this release we mark our general availability (GA) with the models such as ResNet, FairSeq Transformer and RoBERTa, and HuggingFace GLUE task models that have been rigorously tested and optimized.

In addition, with our PyTorch/XLA 1.6 release, you no longer need to run the env-setup.py script on Colab/Kaggle as those are now compatible with native torch wheels. You can still continue to use that script if you would like to run with our latest unstable releases.

New Features

  • XLA RNG state checkpointing/loading (https://github.com/pytorch/xla/pull/2096)
  • Device Memory XRT API (https://github.com/pytorch/xla/pull/2295)
  • [Kaggle/Colab] Small host VM memory environment utility (https://github.com/pytorch/xla/pull/2025)
  • [Advanced User] XLA Builder Support (https://github.com/pytorch/xla/pull/2125)
  • New op supported on PyTorch/XLA
    • Hardsigmoid (https://github.com/pytorch/xla/pull/1940)
    • true_divide (https://github.com/pytorch/xla/pull/1782)
    • max_unpool2d (https://github.com/pytorch/xla/pull/2188)
    • max_unpool3d (https://github.com/pytorch/xla/pull/2188)
    • Replication_pad1d (https://github.com/pytorch/xla/pull/2188)
    • Replication_pad2d (https://github.com/pytorch/xla/pull/2188)
  • Dynamic shape support on XLA:CPU and XLA:GPU (experimental)

Bug Fixes

  • RNG Fix (proper dropout)
  • Manual all-reduce in backward pass (https://github.com/pytorch/xla/pull/2325)
Compute Engine

The Organization Policy for restricting protocol forwarding creation has launched into general availability.

Document AI

Invoice Parsing updates

  • Document AI now supports normalized values for certain entities returned from Invoice Parsing requests.
  • We have improved confidence scores for entities returned from Invoice Parsing requests.
Istio on Google Kubernetes Engine

Istio 1.4.10-gke.5

Fixes an issue with protocol detection connection timeouts.