Posts

gcp artifact repositry getting started

gcp artifact repository helps tos store you container images. First we need to authenticate ourselves to it.  To do that, we simply, run the following commmands to enable and authenticate ourselves to the registry gcloud services enable artifactregistry.googleapis.com gcloud auth configure-docker australia-southeast1-docker.pkg.dev Pulling docker images Next we can pull our images using the following command  docker pull australia-southeast1-docker.pkg.dev/project-xxxxxxx/cloud-run-source-deploy/mytestapp:latest Pushing docker images - the weird moments When pushing, for some reason I am not able to do that with docker # Download crane  curl -sL "https://github.com/google/go-containerregistry/releases/latest/download/go-containerregistry_Linux_x86_64.tar.gz" | tar -xz crane  # Push both tags from the tarball  ./crane push ~/mytestapp.tar australia-southeast1-docker.pkg.dev/project-xxxxx/cloud-run-source-deploy/mytestapp:latest ./crane push ~/mytestapp.tar austr...

google adk eval set runnng from the command line

Image
Eval is important to ensure our model and apps works as expected - or atleast probablistically.  With google adk we can start running our eval set with some environment setup (powershell) $ env:GOOGLE_GENAI_USE_VERTEXAI="TRUE" $env:GOOGLE_CLOUD_PROJECT="project-xxx" $ env:GOOGLE_CLOUD_LOCATION="global" To get started we need 2 things - the dataases and evaluation criteria.  Getting the evalset.json - where we would like to test for a scenario - this is similiar to providing input and defining the expected output like how we would do it in a unit test.  {   "eval_set_id" : "hello_world" ,   "name" : "hello world smoke test" ,   "description" : "No provider named, so intake should default to gcp and mark it as assumed." ,   "eval_cases" : [     {       "eval_id" : "no_provider_defaults_to_gcp" ,       "conversation" : [         {           "invocation_id...

gcp bucket with terraform

Image
Given that this is my first time creating terraform for gcp, surprisingly found it to be similiar to terraform for azure. Here is the simple code to create a bucket and it runs way faster compare to Azure.  main.tf terraform {   required_version = ">= 1.5.0"   required_providers {     google = {       source   = "hashicorp/google"       version = "~> 6.0"     }   } } variable "project_id" {   description = "GCP project ID"   type         = string } variable "bucket_name" {   description = "Globally unique bucket name"   type         = string } variable "location" {   description = "Bucket location (australia-southeast1 = Sydney, australia-southeast2 = Melbourne)"   type         = string   default     = "australia-southeast1" } provider "google" {   project = var . p...

gke service mesh istio is slightly different than the ones you deployed it separately

Image
GKE service mesh uses istio but it is not the typical istio that you get out of the box.  No pod in istio-system For one, there's no pod running in the istio-system namespace (namespace exist but no pods) And when you do a  kubectl get controlplanerevision -n istio-system You will get this - asm-managed  Annotation is different   In order to ensure that sidecar are automatically injected in your pod, we do require this annotation to be added to our namespace and we have to remove the old label that contains "istio-injection" kubectl label namespace test istio.io/rev=asm-managed --overwrite Gateway deploy manually  There's no default load balancer or gateway that will be deployed when you enable service mesh. For that you have to do it manually.  To do that you will need to: kubectl create namespace istio-ingress kubectl label namespace istio-ingress istio.io/rev=asm-managed --overwrite And then deploy the instance using the following yaml  kubectl...

google adk enabling your agent self assessment capabilities

We are enabling agentic self assessment capabilities with loop agent available via google adk. In this setup we have intake agent, architect agent, reviewer agent and coder agent which works together to review each other work to ensure the final output meets the quality standard. Please note this is a simple example to illustrate how we can do this.  The code for the agent is shown here. """ agent.py A multi-agent "software delivery" pipeline built with Google ADK.     intake  ->  LoopAgent[ architect -> designer -> coder -> reviewer ]  ->  release_manager The task is deliberately trivial (App on AWS, Azure or GCP) so the interesting part is the *process*: the reviewer is built to REASSESS the architect's decisions from first principles, not merely proofread the code. Reassessment is made real by three things:   1. A shared, auditable decision log (log_decision tool). Every major decision      is PROPOSED, then judged...