Google Cloud Platform Technology Nuggets — May `16`–31, 2024 Edition

Romin Irani

Published in

Google Cloud - Community

8 min readJun 3, 2024

Welcome to the May 16–31, 2024 edition of Google Cloud Technology Nuggets.

Please feel free to give feedback on this issue and share the subscription form with your peers.

Infrastructure

A Google Cloud incident involving UniSuper in Australia, gained widespread attention last month. Both CEOs made a statement about the incident and there were lot of speculations on what happened, and customers were rightly concerned if this would occur to them to. Google Cloud has published a detailed blog post on what happened, how the recovery was done and remediation.

Are you running your workloads on Compute Engine and have understood various ways to save money on Compute Engine costs? It helps to take a look at a post that highlights 5 ways you can save money on Compute Engine. It includes looking at Auto Recommendations, Committed Use Discounts and more.

Google Cloud continues to invest in infrastructure across Africa. This includes Umoja, the first ever fiber optic route to directly connect Africa with Australia. Check out the post for more details.

Containers and Kubernetes

Running highly scalable services on GKE and finding it tough to allocate network space in a single VPC, and rapidly running out of IP addresses? Check out a design to completely reuse the IP space across your GKE clusters.

Identity and Security

GKE Security Posture Dashboard has been enhanced with software supply chain security insights for your Google Kubernetes Engine workloads. It includes insights into vulnerabilities and workload configuration checks. The dashboard also clearly points out which workloads are affected by a security concern and provides actionable guidance to address it. Check out the blog post for more details.

There are a couple of CISO bulletins for this month. The first CISO bulletin for May 2024 contains a summary of the discussion on a range of topics with the Mandiant CEO, at the RSA Conference fireside chat. The second one takes a look back at the year and offers a glimpse into cybersecurity trends for the year.

Your CI/CD pipelines in respective tools may not be the best places to store your secrets. It is essential to use a dedicated secret manager for the same. Check out this Mandiant analysis of how attackers got a hold of secrets inside of CI/CD pipelines in Bitbucket.

Networking

Google Cloud global front-end released last year had got new updated that were announced at Cloud Next ‘24. As a refresh, the global front end solution consists of the Cloud External Global Application Load Balancer, Cloud CDN and Cloud Armor. The blog post goes into the details of these new features and those include Service Extension Callouts, which allows for more programmability at the Load Balancer layer, custom error responses and more.

What is FQDN (Fully Qualified Domain Name)? How does it fit in within the Cloud Next Generation Firewall (NGFW)? How does it help in ease of management with Firewall Rules configuration? Check out this blog post to know more.

Image Reference: https://en.wikipedia.org/wiki/Fully_qualified_domain_name#/media/File:DNS_schema.svg

Machine Learning

To tune or not to tune LLMs is the question? What are the different methods available to us today to tune, ground the LLM responses to our data. Check out this post and the useful decision tree below.

Are you a gaming developer focused on building out your games and facing a crunch when it comes to designing game assets. Generative AI to the rescue again. Check out how gaming developers can utilize a range of models available on Model Garden on Vertex AI and explore the potential for scaling game-asset creation.

Are you using Google’s foundation models as is and not happy with it being able to return results that are probably just a Google Search away, especially with recent events, etc. You can now ground your model results with Google Search and the process is dead simple as this blog post demonstrates.

Ray, a powerful distributed Python framework, with Google Cloud’s Vertex AI is now generally available (GA). Check out this blog post that highlights the why Ray on Vertex AI, Getting Started with Ray on Vertex AI and ton of other resources.

CloudSQL for PostgreSQL is pushing the envelope when it comes to making it easier for developers to build out Generative AI applications without the complexities. Two things are making this possible. First was the The pgvector extension adds support for vector types along with various Approximate Nearest Neighbour (ANN) index types such as IVFFLAT and HNSW. And now with google_ml_integration extension , that removes the need for external pipelines to integrate your database with LLMs for embedding generation during indexing and search, all with the familiarity and transactional guarantees of SQL. Check out the blog post that highlights the case with an example.

Databases

Cloud SQL has introduced extended support for MySQL and PostgreSQL end-of-life versions. This means that even if you are using a version of the databases that have reached end-of-life, you can opt-in for extended support (a paid service) that will give you the flexibility to migrate at your pace while still giving you security and bug fixes, SLAs and more. Check out the blog post for versions supported and note that pricing is scheduled to be announced in August 2024 as per the post.

Data Analytics

Looking to launch campaigns quickly, efficiently, and with a level of personalization that was previously impossible, leading to increased engagement, conversions, and customer satisfaction? That’s what the blog post states and provides you with a deep dive on how you can use multimodal large language models (LLMs) in BigQuery to create and launch a marketing campaign. Check out the blog post for more details.

We have been reading about how RAGs (Retrieval Augmented Generation) can help to overcome some of the limitations of LLMs. How about an article that gives not just another overview of LLMs, how RAGs help but to create a RAG solution that combines with vector search and BigQuery, thereby tapping into domain-specific knowledge, real-time information and more. Check out the solution.

BigQuery has announced the public preview of numeric search indexes, which enables optimized lookups on INT64 and TIMESTAMP data types. As the blog post states, “ the EQUAL(=) and IN operations on these data types can utilize search indexes to reduce byte scans for improved performance. So now your lookups for account IDs or transactions IDs or log timestamps can get faster and cheaper.” Check out the blog post for more details and a demonstration of the gains on real data, showcasing index creation and queries on a 100TB log table.

Dataflow continues to see significant enhancements to ensure that it remains a solid platform for your streaming applications that need to now make that data available to AI systems for analysis and action. New features have been added to Dataflow ML to allow for support of common machine learning use cases. This includes Dataflow’s new right fitting allows users to mix-and-match compute types to only use GPUs when necessary, reducing cost. At Next ’24, there was a preview announced of continuous queries in BigQuery, that allows for users to now directly create stream processing jobs to create real-time change streams based on the latest data coming into BigQuery. Check out the blog post for more details.

Did you know that BigQuery includes native CDC(Change Data Capture) support? Check out a blog post that highlights BigQuery’s new CDC capability in Dataflow along with the new Dataflow at-least-once streaming mode to drastically simplify your CDC pipeline and reduce costs.

Developers and Practitioners

I am assuming that you are now firmly in Artifact Registry land when it comes to storing, managing, and securing your build artifacts. What if you have multiple projects across an organization and are looking to share the Artifact Registry artifacts outside of the project? Check out this relevant solution which might be a good thing to implement across your organization.

DevOps and SRE

If you have been using SLOs feature within Google Cloud Monitoring, you would have used to seeing a list of services that are auto-discovered and provided to you to monitor on SLO Metrics. The services auto-discovered were from GKE, App Engine and Cloud Run. This has changed, you will need to add these services manually from now on. Check out the post for more details.

Learn Google Cloud

Platform Engineering has been garnering a lot of mindshare of late. We all have our own definition of Platform Engineering and have assumed a few things about Platform Engineering that are likely to be busted in this blog post. The post does a great job to explain Platform engineering by focusing on a few myths but make no mistake, the post explains key points via Google Cloud Platform products and that really helps to map these concepts to actual products. Highly recommended reading.