goTempM is a full stack Golang microservices sample application built on top of the Micro platform. Note that the original goTemp, which is built directly on the go-Micro framework, is still available and can be found in the goTemp repo. It is worth noting that Micro itself uses go-Micro as its underlying framework.
In it current incarnation (this is wip), this mono-repo uses the following stack as backend:
Golang
as its main implementation technologyMicro
as the microservices platform running our servicesgRPC
for inter-service real time communicationNATS
for Pub/Sub event driven communicationmulticast DNS
for service registration and discoveryPostgreSql
for transactional data storageTimescaleDB
time series DB used for historical audit data storageArangoDB
is a multi-model database used for master data storageRedis
is used to cache data and reduce number of data requests to other servicesVault
for credentials management when running in Kubernetes
In terms of the web front end, the stack is as follows:
Javascript
as its main implementation technologySvelte
is used as the compilation engine (via rollup)Sapper
is the javascript frameworkSveltestrap
provides the css framework and is based on bootstrapFont Awesome
to display icons in the application
As far as observability, the application uses:
Prometheus
scrapes metrics from the microservices, databases and brokerGrafana
provides graphic visualization of application metrics
Finally, for orchestration, the stack is as follows:
Docker
for creating application imagesDocker-compose
to run the applicationMinikube
to run the application in Kubernetes
Below is a diagram that displays the overall setup of the application:
In a nutshell. the application functionality is as follows in the backend:
- The frontend connects to the different services through the Micro API gateway
- For each service the frontend provides:
- Search page
- Detail page
- Additionally, the frontend provides:
- Landing page
- Login page
- Register page
- Each service performs the basic CRUD operations to their underlying databases
- All services authenticate via the user service
- All completed CUD operations are forwarded to the NATS broker which in turn forwards the message to the auditing service. This service saves the data into TimescaleDB.
- Each service has a client which can be used to test all basic CRUD functionality
Note: At this time, starting multiple services in a row (which the startup scripts do) sometimes causes an error to be thrown. The details of the issue are described in issue 8. As such, portions of the startup scripts may have to be re-run manually to bring the application up fully.
Before running the application the first time:
- Clone the repository
- cd into
gotempM/web/sapper
- Run the following command to generate the javascript dependencies.
npm install
- Docker must be installed and running
To start the application:
- Ensure that Docker is installed and running. Then, execute the following command from a terminal in the goTempM root folder:
make microup
Depending on whether you have run the application before, docker may have to download all the dependent images (Micro, PostgreSql, TimescaleDB, Nodejs, etc). This may take a while depending on your internet connection speed. Once everything has been downloaded and started, you should see a message in the terminal indicating that the application is listening at localhost:3000. At that point, yo can open your browser and navigate to:
http://localhost:3000
Additionally, observability tooling can be accessed at the addresses below
Prometheus: http://localhost:9090
Grafana: http://localhost:3001
To stop the application:
make microdown
- Ensure Micro is installed on the host
- Docker must be installed and running
- Ensure that Minikube is installed and running
- Similarly, ensure that Helm is installed on the host.
- Install the Micro Helm Chart
The application front end connects with the API gateway using via a K8s ingress resource. As such, the ingress addon must be enabled in Minikube. To enabled it, run:
minikube addons enable ingress
Check the ingress is working using the command below. The command's results should include an entry for the ingress.
kubectl get pods -n kube-system
Out of the box, the Kubernetes manifest will pull existing Bolbeck goTemp images for the front end and some of the DBs from Docker Hub.
You are welcome to change the Kubernetes manifests in the ./cicd/K8s
folder to pull your own images.
To build your own image of the front end and push it to docker hub run the command below for each of the services:
make hubpushcontext SERVICE=web FOLDER=web
where serviceName is the name of the service for which the image should be built folderName is the folder that contains the docker file used to build the service image
Note that Micro will automagically create and deploy the containers for the different services
Enable Port forwarding:
kubectl port-forward svc/proxy -n micro 8081:443
Note that the application itself does not need this port forwarding, this is only needed so that we can send commands to Micro and start the app.
Start application by running:
make microk8sup
Once the application is deployed, check the address and host assigned to the ingress:
kubectl get ingress -n micro
Note that it takes a couple of minutes for K8s to assign the IP to the ingress. As such wait for that happens before moving ahead.
If this is the first time running the app in Minikube: Grab the address & the host from the
result of the command above, and add it to your /etc/hosts
file:
<ipAddress> gotempm.tst
Finally, access app:
minikube service web -n micro
The microservices can be integrated with Vault when running in K8s to manage their credentials.
To enable this integration, please first refer to the README in the ./vault
directory to setup Vault and all the microservices secrets.
Once that is configured, and the application is running, just execute :
make vkubpatchdeploy
Once that completes, the microservices' credentials to the different dependencies (DBs, brokers,etc ...) can be managed in Vault.
Observability tools access while the application is running in K8s (with or without Vault):
- Prometheus:
minikube service prometheus
- Grafana:
minikube service grafana
to stop the application, execute:
make microK8sdown
Note: The port forwarding to Micro should also be stopped. Also, if the Vault Integration is enabled, and the VAULT UI is enabled, then the associated port-forwarding should be stopped as well.
Start the micro server:
micro server
Wait for the server to be up and ensure that you can access localhost:8080 before continuing to next step. Start the application:
make microlocalup
Access the application in your browser at:
http://localhost:3000
To stop the application, run:
make microlocaldown
The project is organized in a way that each folder represents either a service, a database or a shared library package. Currently, we have the following:
arangodb
: Volumes mounted to the ArangoDB container as well as data initialization scriptsaudit
: Audit service to collect and store historical audit informationcicd
: Holds files related to CI/CD and orchestrationcustomer
: Customer master data servicediagramforDocs
: Diagrams used in the readme documentsglobalCache
: Enables Micro to use Redis as a cache store which can then be used in our servicesglobalErrors
: Generic errors shared packageglobalMonitoring
: Generic monitoring utilities shared packageglobalProtos
: Generic protobuf message definitions shared across packagesglobalUtils
: Generic utilities shared packagegrafana
: Grafana configuration and custom dashboard definitionsmicro
: Hosts the custom Dockerfile that is used to run Micro.nats
: NATS dockerfile and configurationpostgres
: Volumes mounted to the PostgreSQL DB container as well as data initialization scriptsproduct
: Product master data servicepromotion
: Promotion service to track product discounts (this was the first service built)prometheus
: Prometheus configuration and exportersredis
: Volumes mounted on the redis container as well as config files (if any)timescaleDB
: Volumes mounted to the Timescale DB container as well as data initialization scriptsuser
: User and authentication serviceVault
: Scripts & policies needed to run the app in K8s with Vaultweb
: application web frontend
Additionally, we have the following files in the root directory as well:
.dockerignore
: Files to be ignored when building service images.gitignore
: Files to be ignored by gitdocker-compose
: File controls the building of the different services and their dependenciesgo.mod and go.sum
: Go modules controlmain.go
: Not used for services yetMakefile
: shortcuts to common actionsReadme.md
: Well... this file...
We use Micro
as the main GO microservices platform. Using Micro simplifies many of the tasks associated with building
microservices including (but not limited to):
- Service discovery
- gRPC for inter service communication
- Built in async messaging (in our case used to set up pub/sub messages to NATS )
- Built-in data storage interface (in our case used to interact with Redis)
- API gateway:
- Request routing
- Load balancing
- Automatic conversion of frontend JSON payloads to backend gRPC messages
Each one of the services has a similar structure:
client
: Contains a client service that calls the server service to perform multiple operationsproto
: Proto buffer messages and services definitions. Empty if service does not handle real time inter-service communication.server
: Service that performs a number of actions like interacting with the DBDockerfile
: Build the image for the server serviceDockerfileCLI
: Build the image of the client servicedocker-compose.env
: Environment variable required to run the service when running the service with docker-composedocker-compose-cli.env
: Environment variable required to run the client when running the client with docker-compose
The services must be started using Micro run. Since we are using the Dockerized version of Micro, we can start a service as follows:
If Micro is not already running:
make microserveup
This will start Micro as well as all the DBs used by goTempM. Get into the Micro container, if not in it already:
docker exec -it microservercont bash
Start the service:
make micro<serviceName>
where can be usersrv, auditsrv, customersrv, productsrv or promotionsrv.
Note: all the commands below must be run within the Micro container
To run some data through a service once it is started, we can run the service client:
micro run --name <serviceClientName> <serviceFolder>/client
where can be usercli, auditcli, customercli, productcli or promotioncli. For example, we could start the user client as follows:
micro run --name usercli user/client
This will bring up run the client service which will attempt to create,update and delete a user. The results will be printed to the log. To see the logs, run:
micro logs -f <serviceOrClientName>
The server user service will update the DB as necessary and send the updated information to the broker (NATS) so that the audit service may eventually store it in the time series DB. The audit service can be started using:
make microauditsrv
The project initializes each of the DBs and seeds them with tables and data. Data changes made at run time are automatically persisted using mounted volumes when running via docker-compose. See the folders for each DB for details as well as the docker-compose file.
Our web front end is built with Svelte and Sapper which have some interesting benefits:
- Server-side initial rendering of our pages
- File based routing
- Smaller code base than other Javascript frameworks. Does more with less.
- Svelte translates the code to vanilla javascript. Thus, smaller application footprint than most frameworks
- Emphasis on component re-usability
The web application lives in the ./web
folder. Since Sapper
and Svelte
generate multiple files and folders, we will just discuss the relevant folders below:
sapper
: The main folder containing the web appsrc
: This is where the bulk of the application residescomponents
: Contains re-usableSvelte
componentsglobalUtils
: Shared javascript utilitiesroutes
: application routes to the different pagesclient.js
: This is a required file. It is used to startSapper
.server.js
: Used to configure the app with items like middleware and compressiontemplate.html
: Main page that contains our application. We added Bootstrap and Font Awesome CDN references in this page.
static
: Holds static items
Dockerfile
: Used to build the docker image for the web app
All of our main routes are pretty standard in terms of organization. We will use the customer route (./web/sapper/src/routes/customer
) as an example:
index.svelte
: Main customers search page that serves at localhost:3000/customer_searchGridSlot
: Component holds the template for the search grid to be display in hte search page (index.svelte)new.svelte
: Page to be displayed when user want to create a new customer. Displayed at localhost:3000/customer/new .[slug].svelte
: Page to view and modify existing customers. Displayed at localhost:3000/customer/[customerid]_detail.svelte
: Holds the gui and bulk of the logic for adding or editing customers. It is called by new.svelte and [slug].svelte .
There are three routes that do not share the structure above as they have very little functionality and thus are server by a single index.svelte component: root, register and login.
The application configuration in K8s can be seen in the diagram below. Note that the diagram shows just one of the different microservices and its associated database. The configuration for all other microservices, beyond the shared ingress and API Gateway, is similar to the one depicted in the diagram. Note that Micro builds and spins out the service pods.
Notes:
- When running the application with Vault, the microservices secrets will be superseded by the secrets stored in Vault.
- The microservices, ArangoDB and NATS have Prometheus metric scrape endpoints built-in. On the other hand, Redis, PostgresDB and TimescaleDB use adapter containers to expose the data to Prometheus.
- To keep the diagram simple, the K8s services for the database, Vault, Prometheus and Grafana are not displayed in the diagram.
- Since Micro automatically manages the K8s objects for the microservices, exposing the metrics ports to Prometheus is done by patching the Micro created K8s services via service patches. The patches can be found under
cicd/K8s/microservicesPatch
The K8s files live in the ./cicd/K8s
folder, and is organized as follows:
dbsAndBroker
: Contains the manifests for all the databases and for the broker.Ingress
: Manifest to create the ingress resource that allows the frontend, and the back end to communicatemicroservicesPatch
: contains the manifests to open the ports required to scrape metrics from the microservicesmonitoring
: holds the manifests to deploy monitoring resources (Prometheus, Grafana)Vault
: Manifests to create the service accounts and patches to integrate the application with Vaultweb
: Manifest for the web front end.
Note that within each of the folders, most related manifests are organized using a prefix. For example, all the front end related services start with the 'web' prefix.
Additional information can be found in the individual folders either in a readme.md
or a doc.go
file.
Additionally, the Makefile contains many command samples that can be used for development.