Gremlin

gremlin

Gremlin is a leading software company focusing on in the market. It also has a tool similar to which belongs to Netflix, but is more customized to test the system with random loads or scheduled shutdowns. In the article below we will be testing a simple Kubernetes cluster running on EKS with Chaos Test. chaos-test Chaos Monkey Why is Chaos Test Important? Chaos Engineering is used to improve system resilience. Gremlin’s helps to find weaknesses in the system before problems occur. “Failure as a Service” Overview To successfully experience Chaos Engineering with Gremlin, firstly we have two requirements: a running EKS cluster and two applications deployed to EKS. This tutorial helps to produce the requirements and create a scenario to 'simulate an attack with Gremlin'. Step 1 - Prepare Cloud9 IDE Step 2 - Create an EKS cluster using eksctl Step 3 -  Deploy Kubernetes Dashboard Step 4 - Install Gremlin using Helm Step 5 - Deploy a Microservice Demo Application Step 6 - Run a Shutdown Container Attack using Gremlin Prerequsities 1. An AWS account 2. A Gremlin account which can be registered from here Step 1 - Prepare Cloud9 IDE Firstly, let's start to create the Cloud9 environment. Login to your AWS account and navigate Cloud9 to service page. Click on and type the name with anything, in this example we have chosen . Keep all default settings as they are stated since it is needed only to reach EKS resources. Get Started chaous gremlin Wait for a while for the new console to build. Then close all terminals to open a new one. To start creating the cluster, firstly check whether AWS CLI is installed or not with the command below: Administrator:~/environment $ aws --version
aws-cli/1.17.5 Python/2.7.16 Linux/4.14.158-101.185.amzn1.x86_64 botocore/1.14.5
Administrator:~/environment $ Step 2 - Creating an EKS cluster using eksctl 1. We will use to create our EKS clusters. eksctl curl --silent --location | tar xz -C /tmp


sudo mv -v /tmp/eksctl /usr/ /bin "https://github.com/weaveworks/eksctl/releases/download/latest_release/eksctl_ _amd64.tar.gz" $(uname -s) local 2. Making executable eksctl Administrator:~/environment $ sudo chmod +x /usr/ /bin/eksctl
Administrator:~/environment $ local 3. Confirming whether the command works: eksctl Administrator:~ $ eksctl version
[ℹ]  version.Info{BuiltAt: , GitCommit: , GitTag: }
Administrator:~ $ "" "" "0.13.0" 4. Below we are creating the cluster named with three EC2 nodes. Just a word of warning - EKS can cost a lot so please do not forget to delete your resources after you have done with your failure test. gremlin-eksctl eksctl create cluster --name=gremlin-eksctll --nodes=3 --managed --alb-ingress-access --region= ${AWS_REGION} It might take around 15-30 minutes to get ready which you can cluster on the EKS service page. - Fee of EKS is 0,20$/h and fee of EC2 with a m5.large instance type that EKS runs on is 0.096$/h. Estimation of total cost per day will be around 11-12$. Quit Tip 5. Checking the cluster whether it is working and getting its status. As expected, there is only one cluster created. Administrator:~/environment $ eksctl get clusters
NAME            REGION
gremlin-eksctl  us-east-1
Administrator:~/environment $ 6. Updating the file by giving the cluster name and region via AWS CLI tool. kubeconfig Administrator:~/environment $ sudo aws eks --region us-east-1  update-kubeconfig --name gremlin-eksctl
Updated context arn:aws:eks:us-east-1:312867154612:cluster/gremlin-eksctl /root/.kube/config
Administrator:~/environment $ in 7. On checking we see 3 nodes in the cluster. Two of them are worker-nodes while the third is cluster management master-node. Administrator:~/environment $ kubectl get nodes
NAME                            STATUS   ROLES    AGE   VERSION
ip-192-168-14-59.ec2.internal   Ready       59m   v1.14.7-eks-1861c5
ip-192-168-33-12.ec2.internal   Ready       58m   v1.14.7-eks-1861c5
ip-192-168-49-55.ec2.internal   Ready       58m   v1.14.7-eks-1861c5
Administrator:~/environment $ Step 3 - Deploying Kubernetes Dashboard Next we would deploy the Kubernetes dashboard to Kubernetes cluster by using and . These two tools will help our sample application to be shown in the dashboard. We will start with deploying our Kubernetes dashboard as the first step. Heapster InfluxDB Heapster is a performance monitoring and metrics collection system compatible with Kubernetes (versions 1.0.6 and above). It allows for the collection of not only performance metrics about your workloads, pods, and containers, but also events and other signals generated by your cluster. The great thing about Heapster is that it is fully open source as part of the Kubernetes project, and supports a multitude of backends for persisting the data, including but , , , and . Heapster not limited to Influxdb Elasticsearch Graphite InfluxDB InfluxDB is a time series database designed to handle high volume of writing and query loads. 1. Deploying the Kubernetes Dashboard to your EKS cluster: Administrator:~/environment $ kubectl apply -f https://raw.githubusercontent.com/kubernetes/dashboard/v1.10.1/src/deploy/recommended/kubernetes-dashboard.yaml
secret/kubernetes-dashboard-certs created
serviceaccount/kubernetes-dashboard created
role.rbac.authorization.k8s.io/kubernetes-dashboard-minimal created
rolebinding.rbac.authorization.k8s.io/kubernetes-dashboard-minimal created
deployment.apps/kubernetes-dashboard created
service/kubernetes-dashboard created
Administrator:~/environment $ 2. Firstly, access to cluster is given since it is deployed to our private cluster while we need to access it via a proxy. This can be done through as the preferred proxy tool. In your workspace, run the following command: Kube-Proxy Administrator:~/environment $ kubectl proxy --port=8080 --address= -- -filter= &
[1] 336
Administrator:~/environment $ W0125 10:28:19.746961     336 proxy.go:140] Request filter disabled, your proxy is vulnerable to XSRF attacks, please be cautious
Starting to serve on [::]:8080

Administrator:~/environment $ '0.0.0.0' disable true 3. Deploying : Heapster Administrator:~/environment $  kubectl apply -f https://raw.githubusercontent.com/kubernetes/heapster/master/deploy/kube-config/influxdb/heapster.yaml
serviceaccount/heapster created
deployment.extensions/heapster created
service/heapster created
Administrator:~/environment $ 4. Deploying : InfluxDB Administrator:~/environment $  kubectl apply -f https://raw.githubusercontent.com/kubernetes/heapster/master/deploy/kube-config/influxdb/influxdb.yaml
deployment.extensions/monitoring-influxdb created
service/monitoring-influxdb created
Administrator:~/environment $ 5. Creating cluster role binding for the Dashboard. Heapster Administrator:~/environment $ kubectl apply -f https://raw.githubusercontent.com/kubernetes/heapster/master/deploy/kube-config/rbac/heapster-rbac.yaml
clusterrolebinding.rbac.authorization.k8s.io/heapster created
Administrator:~/environment $ 6. The next step is to create an service account. It will let you connect to the Kubernetes Dashboard with permissions. eks-admin admin To authenticate and use the Kubernetes Dashboard: Administrator:~/environment $ kubectl apply -f https://raw.githubusercontent.com/tammybutow/eks-aws/master/eks-admin-service-account.yaml
serviceaccount/eks-admin created
clusterrolebinding.rbac.authorization.k8s.io/eks-admin created
Administrator:~/environment $ 7. To access the Kubernetes Dashboard: a- In your Cloud9 environment, click to open the Dashboard URL. Tools > Preview > Preview Running Application b- Append the following to the end of the URL: /api/v1/namespaces/kube-system/services/https:kubernetes-dashboard:/proxy/ 8. Open a new Terminal tab on Cloud9 and enter the following: aws eks get-token --cluster-name gremlin-eksctl | jq -r '.status.token' 9. Select and then copy the output of the command above and paste it to the text field as shown below: Token Step 4 - Installing Gremlin using Helm 1. Download your Gremlin certificates: Start by signing-in to your Gremlin account. If you don't have one, create an account . here Navigate to and click on your . Click the button to download and save certificates to your local drive. Please note that the downloaded contains both a and a matching . Team Settings Team Download certificate.zip public-key certificate private key Unzip the and save it to your Gremlin folder on your desktop. Rename your certificate as and key files as . certificate.zip gremlin.cert gremlin.key 2. Creating Gremlin Namespace: Administrator:~/environment $ kubectl create namespace gremlin
namespace/gremlin created
Administrator:~/environment $ Create a Kubernetes Secret for your certificate and private key, copy and to Cloud9. A quick tip is to create these by the Vim Editor instead of copying from your local computer. gremlin.cert gremlin.key Administrator:~/environment $ kubectl create secret generic gremlin-team-cert \
> \--namespace=gremlin  \
> \--from-file=/home/ec2-user/environment/gremlin.cert \
> \--from-file=/home/ec2-user/environment/gremlin.key
secret/gremlin-team-cert created
Administrator:~/environment $ Check the files on Dashboard whether they are deployed. 3. Installation with Helm The simplest way of installing the Gremlin client on your Kubernetes cluster is to use Helm. Once Helm is installed and configured, the next steps will be to add the Gremlin and to install the client. repo A. Installing Helm source code and making it executable: Administrator:~/environment $ curl https://raw.githubusercontent.com/kubernetes/helm/master/scripts/get > get_helm.sh
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100  7164  100  7164    0     0  45630      0 --:--:-- --:--:-- --:--:-- 45341
Administrator:~/environment $ 
Administrator:~/environment $ chmod +x get_helm.sh
Administrator:~/environment $ 
Administrator:~/environment $ ./get_helm.sh
Downloading https://get.helm.sh/helm-v2.16.1-linux-amd64.tar.gz
Preparing to install helm and tiller into /usr/ /bin
helm installed into /usr/ /bin/helm
tiller installed into /usr/ /bin/tiller
Run to configure helm.
Administrator:~/environment $ local local local 'helm init' B. Configuring Helm to access with RBAC Helm relies on a service called which requires special permission on the Kubernetes cluster, for which we will need to build a for using . Next step is to then apply this RBAC to the cluster. Tiller Service Account Tiller C. Creating a new service account: Administrator:~/environment $ cat < ~/environment/rbac.yaml
> ---
> apiVersion: v1
> kind: ServiceAccount
> metadata:
>   name: tiller
>   namespace: kube-system
> ---
> apiVersion: rbac.authorization.k8s.io/v1beta1
> kind: ClusterRoleBinding
> metadata:
>   name: tiller
> roleRef:
>   apiGroup: rbac.authorization.k8s.io
>   kind: ClusterRole
>   name: cluster-admin
> subjects:
>   - kind: ServiceAccount
>     name: tiller
>     namespace: kube-system
> EoF
Administrator:~/environment $ D. Applying configurations: Administrator:~/environment $ kubectl apply -f ~/environment/rbac.yaml
serviceaccount/tiller created
clusterrolebinding.rbac.authorization.k8s.io/tiller created
Administrator:~/environment $ E. Installing for : Tiller Helm Tiller A companion server component, , that runs on your Kubernetes cluster, listens for commands from , and handles the configuration and deployment of software releases on the cluster. tiller helm Administrator:~/environment $ helm init --service-account tiller
Creating /home/ec2-user/.helm 
Creating /home/ec2-user/.helm/repository 
Creating /home/ec2-user/.helm/repository/cache 
Creating /home/ec2-user/.helm/repository/ Creating /home/ec2-user/.helm/plugins 
Creating /home/ec2-user/.helm/starters 
Creating /home/ec2-user/.helm/cache/archive 
Creating /home/ec2-user/.helm/repository/repositories.yaml 
Adding stable repo with URL: https://kubernetes-charts.storage.googleapis.com 
Adding repo with URL: http://127.0.0.1:8879/charts has been configured at /home/ec2-user/.helm.

Tiller (the Helm server-side component) has been installed into your Kubernetes Cluster.

Please note: by default, Tiller is deployed with an insecure policy.
To prevent this, run `helm init` with the --tiller-tls-verify flag.
For more information on securing your installation see: https://docs.helm.sh/using_helm/ Administrator:~/environment $ local local $HELM_HOME 'allow unauthenticated users' #securing-your-helm-installation This will install into the cluster and will give access to in your cluster. Please note the security policy alert as shown above, which you can feel free to ignore or follow as per your policy settings. tiller managed resources F. Activating bash-completion for Helm: Administrator:~/environment $ helm completion bash >> ~/.bash_completion
Administrator:~/environment $ . /etc/profile.d/bash_completion.sh
Administrator:~/environment $ . ~/.bash_completion
Administrator:~/environment $ G. To run the Helm install, you will need your Gremlin Team ID. It can be found in the Gremlin app on the page, where you downloaded your certificates earlier. Click on your in the list. The ID you’re looking for can be found under as . Team Settings Team Configuration Team ID H. Export your Team ID as an variable: environment Administrator:~/environment $ GREMLIN_TEAM_ID=13ba2df0-5b0b-572c-a479-8136af79da66
Administrator:~/environment $ export I. Next, export your cluster ID, by giving a name for your Kubernetes cluster. Administrator:~/environment $ GREMLIN_CLUSTER_ID=boraozkanchaos
Administrator:~/environment $ export J. Now add the Gremlin Helm repo, and install Gremlin: Administrator:~/environment $ helm repo add gremlin https://helm.gremlin.com has been added to your repositories
Administrator:~/environment $ "gremlin" Administrator:~/environment $ helm install gremlin/gremlin \
> \--namespace gremlin \
> \--name gremlin \
> \-- gremlin.teamID= \
> \-- gremlin.clusterID= NAME:   gremlin
LAST DEPLOYED: Sat Jan 25 12:37:45 2020
NAMESPACE: gremlin
STATUS: DEPLOYED

RESOURCES:
==> v1/ClusterRole
NAME             AGE
gremlin-watcher  0s

==> v1/ClusterRoleBinding
NAME  AGE
chao  0s

==> v1/DaemonSet
NAME     AGE
gremlin  0s

==> v1/Deployment
NAME  AGE
chao  0s

==> v1/Pod(related)
NAME                   AGE
chao-698b9fbfb4-5thjp  0s
gremlin-425wv          0s
gremlin-krdlt          0s
gremlin-l252z          0s

==> v1/ServiceAccount
NAME  AGE
chao  0s


Administrator:~/environment $ set $GREMLIN_TEAM_ID set $GREMLIN_CLUSTER_ID Step 5 - Deploying a Microservice Demo Application The demo environment we are going to deploy on to our EKS cluster is the Hipster Shop: Cloud-Native Microservices Demo Application 1. Clone repo of app source code: Administrator:~/environment $ git https://github.com/GoogleCloudPlatform/microservices-demo.git
Cloning into ...
remote: Enumerating objects: 30, .
remote: Counting objects: 100% (30/30), .
remote: Compressing objects: 100% (21/21), .
remote: Total 2987 (delta 16), reused 16 (delta 7), pack-reused 2957
Receiving objects: 100% (2987/2987), 5.08 MiB | 38.24 MiB/s, .
Resolving deltas: 100% (2039/2039), .
Administrator:~/environment $ clone 'microservices-demo' done done done done done 2. Change directory to the one just created: Administrator:~/environment $ microservices-demo/
Administrator:~/environment/microservices-demo (master) $ cd 3. Deploying the application: Administrator:~/environment/microservices-demo (master) $ kubectl apply -f ./release/kubernetes-manifests.yaml
deployment.apps/emailservice created
service/emailservice created
deployment.apps/checkoutservice created
service/checkoutservice created
deployment.apps/recommendationservice created
service/recommendationservice created
deployment.apps/frontend created
service/frontend created
service/frontend-external created
deployment.apps/paymentservice created
service/paymentservice created
deployment.apps/productcatalogservice created
service/productcatalogservice created
deployment.apps/cartservice created
service/cartservice created
deployment.apps/loadgenerator created
deployment.apps/currencyservice created
service/currencyservice created
deployment.apps/shippingservice created
service/shippingservice created
deployment.apps/redis-cart created
service/redis-cart created
deployment.apps/adservice created
service/adservice created
Administrator:~/environment/microservices-demo (master) $ 4. Wait until pods are in a ready state. Administrator:~/environment/microservices-demo (master) $ kubectl get pods
NAME                                     READY   STATUS    RESTARTS   AGE
adservice-84449b8756-qj4sp               1/1     Running   0          4m57s
cartservice-6cbc9b899c-ww5rf             1/1     Running   0          4m58s
checkoutservice-56b48b77c8-fszpx         1/1     Running   0          4m58s
currencyservice-b9fcb4c98-hx9kq          1/1     Running   0          4m58s
emailservice-797cdcc76d-d257v            1/1     Running   0          4m58s
frontend-785c44fd98-zvgj4                1/1     Running   0          4m58s
loadgenerator-665c4ddb74-xwm8m           1/1     Running   3          4m58s
paymentservice-84d7bf956-fdgxr           1/1     Running   0          4m58s
productcatalogservice-5664f59f54-b5mk5   1/1     Running   0          4m58s
recommendationservice-7f9855d7c6-b2zv6   1/1     Running   0          4m58s
redis-cart-6448dcbdcc-97d55              1/1     Running   0          4m58s
shippingservice-6b6f49747d-svb5f         1/1     Running   0          4m58s
Administrator:~/environment/microservices-demo (master) $ 5. Getting the frontend IP address: Administrator:~/environment/microservices-demo (master) $ kubectl get svc frontend-external -o wide
NAME                TYPE           CLUSTER-IP       EXTERNAL-IP                                                              PORT(S)        AGE     SELECTOR
frontend-external   LoadBalancer   10.100.226.157   a0c9bed9a3f7111eab7c912c03cd100e-265884000.us-east-1.elb.amazonaws.com   80:30461/TCP   6m31s   app=frontend
Administrator:~/environment/microservices-demo (master) $ 6. Visit the URL on your browser: Step 6 - Running a Shutdown Container Attack using Gremlin We are going to create our first Chaos Engineering experiment where we would validate the EKS reliability. Our hypothesis is, “After shutting down my container, we will not suffer from downtime and EKS will give us a new one.” cart service Going back to the Gremlin UI, select from the menu on the left and select . We’re going to target a Kubernetes resource, so click on on the upper right. Attacks New Attack Kubernetes Choose : cartservice Choose and : State Shutdown Attacking this pod with our Gremlin UI: We will be shutting down the containers. As a test, we attacked twice and the cartservice pods restarted itself. Which signifies that it is working as expected. Note that it re-generates itself even when you attack to shut down your pods. cartservice When we attacked our containers, the cluster resisted to failure and restarted itself, which symbolizes that our system is now resistant to failure. We have seen what happens when a failure occurs, in this example the failure is shutting down the pods. As a result, we understand that our cluster already has auto-scaling feature. As a reminder; do not forget to delete your and ide. cluster Cloud9 eksctl delete cluster --name=gremlin-eksctl Conclusion Congrats! You’ve installed an AWS EKS cluster, deployed the Kubernetes Dashboard, deployed a microservice demo application, installed the Gremlin agent as a daemon-set, and ran your first Chaos Engineering attack to validate Kubernetes reliability! This article was originally published on https://appfleet.com/blog/chaous-test-to-your-kubernetes-with-gremlin/ .

The Secret To Succesully Simulate An Attack With Gremlin To Test Kubernetes

In September 2019, Graph Query Language is accepted as a new database query language in a vote by the International SQL Standards Committee. The unification of GQL takes time. In this post, we’ve selected some mainstream graph query languages and compared the CRUD usage in these languages respectively. The Graph Query Languages to Be Compared Gremlin is a graph traversal language developed by Apache TinkerPop and has been adopted by a lot of graph database solutions. It can be either or . Gremlin declarative imperative Gremlin is Groovy-based, but has many language variants that allow developers to in many modern programming languages such as Java, JavaScript, Python, Scala, Clojure and Groovy. write Gremlin queries natively : Janus Graph, InfiniteGraph, Cosmos DB, DataStax Enterprise(5.0+) and Amazon Neptune. Supported graph databases Cypher is a declarative graph query language that allows expressive and efficient data querying in a property graph. Cypher The language was designed with the power and capability of SQL. The keywords of the Cypher language are not case sensitive, but attributes, labels, relationship types and variables are case sensitive. : Neo4j, AgensGraph and RedisGraph Supported graph databases nGQL introduces its own query language, , which is a declarative, textual query language like SQL, but designed for graphs. Nebula Graph nGQL The keywords of the nGQL language are case sensitive and it support statement composition so that there’s no need for statement embedding. : Nebula Graph Support graph databases Terms Comparison Before comparing the three graph query languages, let’s take a look at their terms and concepts first. The table below explains how these languages define nodes and edges: Syntax Comparison - CRUD After understanding the common terms in Gremlin, Cypher, and nGQL, let’s take a look at the general syntax of these graph query languages. This section will walk you through the basic CRUD syntax for Gremlin, Cypher and nGQL respectively. Graph Refer to the following example on how to create a graph space. We omitted Cypher since you don’t need to create a graph space before you can add any data to the graph databases. g = TinkerGraph. ().traversal() CREATE gods # Create a graph that Gremlin can traverse open # Create a graph space in nGQL SPACE Vertex We all know that graph consists of nodes and edges. A node in Cypher is called a vertex in Gremlin and nGQL. And an edge is the connection among two nodes. Refer to the following example on inserting new vertex in these query languages respectively. # vertex Gremlin
g.addV(vertexLabel).property()

# vertex Cypher (:nodeLabel {property})

# vertex nGQL VERTEX tagName (propNameList) vid:(tagKey propValue) Insert in Insert in CREATE Insert in INSERT VALUES Vertex Type The nodes/vertices can have types. They are called labels in Gremlin and Cypher, and tags in nGQL. A vertex type can have multiple properties. For example, vertex type have two properties, i.e. and . Person name age Create Vertex Type Refer to the following example on vertex type creation. We omitted Cypher since you don’t need a label before inserting data. g.addV(vertexLabel).property() CREATE tagName(PropNameList) # Create vertex Gremlin type in # Create vertex nGQL type in Note that Gremlin and nGQL both support . This keyword automatically detects if the corresponding vertex type exists. If it does not exist, a new one is created. Otherwise, no vertex type is created. IF NOT EXISTS Show Vertex Types When the vertex types are created, you can show them with the following queries. They will list all the labels/tags rather than certain labels/tags. # vertex Gremlin
g.V().label().dedup();

# vertex Cypher MATCH (n) labels(n)
# vertex Cypher db.labels();

# vertex nGQL TAGS Show types in Show types in method 1 RETURN DISTINCT Show types in method 2 CALL Show types in SHOW CRUD on Vertices This section introduces the basic CRUD operations on vertices using the three query languages. Insert Vertices # vertex certain Gremlin
g.addV(String vertexLabel).property()

# vertex certain Cypher (node:label) 

# vertex certain nGQL VERTEX <tag_name> (prop_name_list) <vid>:(prop_value_list) Insert of type in Insert of type in CREATE Insert of type in INSERT VALUES Get Vertices # vertices Gremlin
g.V(<vid>)

# vertices Cypher
MATCH (n) condition properties(n)

# vertices nGQL PROP <tag_name> <vid> Fetch in Fetch in WHERE RETURN Fetch in FETCH ON Delete Vertices g.V( ).drop() MATCH ( ) 
DETACH DELETE vertex nGQL
DELETE VERTEX # Delete vertex in Gremlin <vid> # Delete a vertex in Cypher node :label node # Delete in <vid> Update a Vertex’s Property This section shows you how to update properties for vertex. # vertex Gremlin .V(<vid>).property()

# vertex Cypher . = V

# vertex nGQL VERTEX <vid> <update_columns> Update in g Update in SET n prop Update in UPDATE SET Both Cypher and nGQL use keyword to set the vertex type, except that the keyword is added in nGQL to identify the operation. The operation of Gremlin is similar to the above-mentioned fetching vertex, except that it adds operations for changing property. SET UPDATE Edges This section introduces the basic CRUD operations on edges. Edge Types Like vertices, edges can also have types. g.edgeLabel() CREATE EDGE edgeTypeName(propNameList) # Create an edge Gremlin type in # Create an edge nGQL type in CRUD on Edges Insert Edges of Certain Types Inserting an edge is similar to inserting vertices. Cypher uses and and nGQL uses to represent edges respectively. Gremlin uses the keyword to indicate edge direction. -[]-> -> to() Edges are by default directed in the three languages. The chart on the left below is a directed edge while the one on the right is an undirected edge. # edge certain Gremlin
g.addE(String edgeLabel). (v1). (v2).property()

# edge certain Cypher (<node1- >:<label1- >)-
  [(<relationship- >:<relationship-label- >)]
  ->(<node2- >:<label2- >)

# edge certain nGQL EDGE <edge_name> ( <prop_name_list> ) <src_vid> -> <dst_vid>: ( <prop_value_list> ) Insert of type in from to Insert of type in CREATE name name name name name name Insert of type in INSERT VALUES Delete Edges # edge Gremlin
g.E(<eid>). ()

# edge Cypher
MATCH (<node1- >:<label1- >)-[r:relationship-label- ]->() r

# edge nGQL EDGE <edge_type> <src_vid> -> <dst_vid> Delete in drop Delete in name name name DELETE Delete in DELETE Fetch Edges # edges Gremlin
g.E(<eid>)

# edges Cypher
MATCH (n)-[r:label]->() condition properties(r)

# edges nGQL PROP <edge_name> <src_vid> -> <dst_vid> Fetch in Fetch in WHERE RETURN Fetch in FETCH ON Other Operations In addition to the common CRUD on vertices and edges, we will also show you some combined queries in the three graph query languages. Traverse edges # Traverse edges specified vertices Gremlin
g.V(<vid>).outE(<edge>)

# Traverse edges specified vertices Cypher
Match (n)->[r:label]->[]
WHERE id(n) = vid
RETURN r

# Traverse edges specified vertices nGQL
GO FROM <vid> OVER <edge> with in with in with in Traverse edges reversely In reverse traverse, Gremlin uses to indicate reversion, Cypher uses . nGQL uses keyword . in <- REVERSELY # Traverse edges reversely specified vertices Gremlin
g.V(<vid>). (<edge>)

# Traverse edges reversely specified vertices Cypher
MATCH (n)<-[r:label]-()

# Traverse edges reversely specified vertices nGQL
GO FROM <vid>  OVER <edge> REVERSELY with in with with Traverse edges bidirectionally If the edge direction is (either direction is acceptable),  Gremlin uses , Cypher use , and nGQL use keyword . irrelevant bothE() -[]- BIDIRECT g.V( ).bothE( ) MATCH (n)-[r: ]-() GO FROM OVER BIDIRECT # Traverse edges reversely with specified vertices Gremlin <vid> <edge> # Traverse edges reversely with specified vertices Cypher label # Traverse edges reversely with specified vertices nGQL <vid> <edge> Query N hops along specified edge Gremlin and nGQL use and respectively to represent N hops. Cypher uses . times STEP relationship*N # hops along specified edge Gremlin .V(<vid>). ( (<edge>)).times( )

# hops along specified edge Cypher
MATCH ( )-[r: * ]->()
WHERE condition # hops along specified edge nGQL
GO STEPS FROM <vid> OVER <edge> Query N in g repeat out N Query N in n label N RETURN r Query N in N Find paths between two vertices # Find paths two vertices Gremlin
g.V(<vid>).repeat( ()). (<vid>).path()

# Find paths two vertices Cypher
MATCH p =(a)-[.*]->(b) condition p

# Find paths two vertices nGQL
FIND <vid> <vid> * between in out until between in WHERE RETURN between in ALL PATH FROM TO OVER Example queries This section introduces some demonstration queries. Demo model: The Graphs of Gods The examples in this section make extensive use of the toy graph distributed with called as diagrammed below。 Janus Graph The Graphs of Gods This example describes the relationships between the beings and places of the Roman pantheon. Inserting data ==>v[1] ==>v[2] ==>v[31] ==>v[32] ==>e[13][2-father->1] # Inserting vertices # # nGQL nebula> INSERT VERTEX character(name, age, ) VALUES ( ):( , 10000, ), ( ):( , 5000, ); type hash "saturn" "saturn" "titan" hash "jupiter" "jupiter" "god" # # Gremlin gremlin> saturn = g.addV( ).property(T.id, 1).property( , ).property( , 10000).property( , ).next(); "character" 'name' 'saturn' 'age' 'type' 'titan' gremlin> jupiter = g.addV( ).property(T.id, 2).property( , ).property( , 5000).property( , ).next(); "character" 'name' 'jupiter' 'age' 'type' 'god' gremlin> prometheus = g.addV( ).property(T.id, 31).property( , ).property( , 1000).property( , ).next(); "character" 'name' 'prometheus' 'age' 'type' 'god' gremlin> jesus = g.addV( ).property(T.id, 32).property( , ).property( , 5000).property( , ).next(); "character" 'name' 'jesus' 'age' 'type' 'god' # # Cypher cypher> CREATE (src:character {name: , age: 10000, : }) "saturn" type "titan" cypher> CREATE (dst:character {name: , age: 5000, : }) "jupiter" type "god" # Inserting edges # # nGQL nebula> INSERT EDGE father() VALUES ( )-> ( ):(); hash "jupiter" hash "saturn" # # Gremlin gremlin> g.addE( ).from(jupiter).to(saturn).property(T.id, 13); "father" # # Cypher cypher> CREATE (src)-[rel:father]->(dst) Deleting # nGQL nebula> DELETE VERTEX ( ); hash "prometheus" # Gremlin gremlin> g.V(prometheus).drop(); # Cypher cypher> MATCH (n:character {name: }) DETACH DELETE n "prometheus" Updating ==>v[32] # nGQL nebula> UPDATE VERTEX ( ) SET character.type = ; hash "jesus" 'titan' # Gremlin gremlin> g.V(jesus).property( , 6000); 'age' # Cypher cypher> MATCH (n:character {name: }) SET n.type = ; "jesus" 'titan' Fetching/Reading # nGQL # Gremlin
gremlin> g.V(saturn).valueMap();
==>[name:[saturn],type:[titan],age:[10000]]
# Cypher
cypher> MATCH (n:character {name:"saturn"}) RETURN properties(n) nebula> FETCH PROP ON character hash("saturn");
=================================================== | character.name | character.age | character.type |
=================================================== | saturn         | 10000         | titan          |
--------------------------------------------------- ╒════════════════════════════════════════════╕ │"properties(n)"                             │ ╞════════════════════════════════════════════╡ │{"name":"saturn","type":"titan","age":10000}│ └────────────────────────────────────────────┘ Finding the name of Hercules’s Father # nGQL
nebula>  LOOKUP .name == | \
      -> GO $-.VertexID father YIELD $$ .character.name;
=====================
| $$ .character.name |
=====================
| jupiter           | # Gremlin
gremlin> g.V().hasLabel( ).has( , ). ( ). ( );
==>jupiter
# Cypher
cypher> MATCH (src: { :"hercules"})-[:father]->(dst: ) dst.name
      ╒══════════╕
      │"dst.name"│
      ╞══════════╡
      │"jupiter" │
      └──────────┘ ON character WHERE character 'hercules' FROM OVER --------------------- 'character' 'name' 'hercules' out 'father' values 'name' character name character RETURN Finding the name of Hercules’s Grandfather # nGQL
nebula> LOOKUP .name == | \
     -> GO STEPS $-.VertexID father YIELD $$ .character.name;
=====================
| $$ .character.name |
=====================
| saturn            | # Gremlin
gremlin> g.V().hasLabel( ).has( , ). ( ). ( ). ( );
==>saturn
# Cypher
cypher> MATCH (src: { :"hercules"})-[:father* ]->(dst: ) dst.name
      ╒══════════╕
      │"dst.name"│
      ╞══════════╡
      │"saturn"  │
      └──────────┘ ON character WHERE character 'hercules' 2 FROM OVER --------------------- 'character' 'name' 'hercules' out 'father' out 'father' values 'name' character name 2 character RETURN Find the characters with age > 100 # nGQL # Gremlin
gremlin> g.V().hasLabel( ).has( ,gt(100)).values( );
==>saturn
==>jupiter
==>neptune
==>pluto
# Cypher
cypher> MATCH (src:character) WHERE src.age > 100 RETURN src.name nebula> LOOKUP ON character WHERE character.age > 100 YIELD character.name, character.age;
========================================================= | VertexID             | character.name | character.age |
========================================================= | 6761447489613431910  | pluto          | 4000          |
--------------------------------------------------------- | -5860788569139907963 | neptune        | 4500          |
--------------------------------------------------------- | 4863977009196259577  | jupiter        | 5000          |
--------------------------------------------------------- | -4316810810681305233 | saturn         | 10000         |
--------------------------------------------------------- 'character' 'age' 'name' ╒═══════════╕ │"src.name" │ ╞═══════════╡ │  "saturn" │ ├───────────┤ │ "jupiter" │ ├───────────┤ │ "neptune" │ │───────────│ │  "pluto"  │ └───────────┘ Find who are Pluto’s cohabitants, excluding Pluto himself # nGQL
nebula>  GO hash("pluto") lives YIELD lives._dst place | GO $-.place lives REVERSELY \
$$ .character.name != "pluto" YIELD $$ .character.name cohabitants;
===============
| cohabitants |
===============
| cerberus    | # Gremlin
gremlin> g.V(pluto). ( ). ( ). ( (neq(pluto))). ( );
==>cerberus
# Cypher
cypher> MATCH (src: { :"pluto"})-[:lives]->()<-[:lives]-(dst: ) dst.name
      ╒══════════╕
      │"dst.name"│
      ╞══════════╡
      │"cerberus"│
      └──────────┘ FROM OVER AS FROM OVER WHERE AS --------------- out 'lives' in 'lives' where is values 'name' character name character RETURN Pluto’s Brothers # which brother lives which place?
## nGQL
nebula> GO hash("pluto") brother YIELD brother._dst god | \
GO $-.god lives YIELD $^.character.name Brother, $$ . .name Habitations;
=========================
| Brother | Habitations |
=========================
| jupiter | sky         | | neptune | sea         | ## Gremlin
gremlin> g.V(pluto). ( ). ( ). ( ). ( ). ( , ). ( );
==>[god:jupiter, place:sky]
==>[god:neptune, place:sea]
## Cypher
cypher> MATCH (src: { :"pluto"})-[:brother]->(bro: )-[:lives]->(dst) bro.name, dst.name
      ╒═════════════════════════╕
      │"bro.name"    │"dst.name"│
      ╞═════════════════════════╡
      │ "jupiter"    │  "sky"   │
      ├─────────────────────────┤
      │ "neptune"    │ "sea"    │
      └─────────────────────────┘ in FROM OVER AS FROM OVER AS location AS ------------------------- ------------------------- out 'brother' as 'god' out 'lives' as 'place' select 'god' 'place' by 'name' Character name Character RETURN In addition to the basic operations in the three graph query languages, we will work on another piece of comparison of advanced operations in these languages. Stay tuned! Previously published at https://nebula-graph.io/posts/graph-query-language-comparison-cypher-gremlin-ngql/

Gremlin vs Cypher vs nGQL: Graph Query Language Comparison 

The Power of Market Disruption: How to Detect Fraud With Graph Data

They say the devil's in the details, but in the case of , it's more like a mischievous gremlin with a penchant for blockchain security. I plunged headfirst into this tech rabbit hole to witness Mitja Gorochevsky, the wild mind behind and this radical consensus algorithm, unleash his latest creation on a crowd of unsuspecting tech enthusiasts. Acki Nacki GOSH The Scene: A Digital Carnival of Quirk Picture a virtual stage where coding wizards and crypto evangelists collide—that's the kind of energy that crackled through the air as Mitja took the reins. The vibe was part tech conference, part cyberpunk circus, with Acki Nacki taking center stage as a personified defender of integrity. blockchain But even gremlins have their off days. Internet gremlins, that is. The demo hit a few snags thanks to connectivity hiccups, but that didn't dampen the spirit. It just added a layer of unpredictable chaos to the mix, as if Acki Nacki was flexing its trickster muscles. Enter Acki Nacki: The Multi-Headed Hydra of Blockchain Now, hold on to your crypto wallets because Acki Nacki is no ordinary consensus algorithm. Imagine a multi-core, multi-threaded execution environment where anyone can drop a contract and run operations in parallel with lightning-fast, under-1-second finality. And all this on the decentralised supercomputer known as GOSH. "The odds of a successful malicious attack on GOSH (with Acki Nacki) is not just less likely than a successful attack on Bitcoin," Mitja boldly declared, "it's less likely than an asteroid hitting Earth and wiping us all out tomorrow!" Hackathon Hoedown and the GOSH Gold Rush But the real party starts with the "Underworld" hackathon. A siren call to tech rebels, inviting them to bend the boundaries of what's possible. Here are the challenges they laid down: Wallet apps that make your crypto quiver with glee. Visualisation tools to make blockchains look like trippy art installations. Security tooling so tight that even the NSA would be impressed. SDKs, TVMs, ZKPs... basically, if it sounds like alphabet soup, it's fair game. Why go through all this trouble? Because the only way to score GOSH tokens is either through this hackathon or the "Die Last" validator games. It's a tech wild west where your coding skills are your six-shooter. Is anyone paying attention, though, Ben? Well, I'm glad you asked. For me, the big-news-plot-twist of the evening, away from all the tech bluster, is that the United Nations, the global epitome of diplomatic collaboration, has chosen GOSH to host its decentralized autonomous organization (DAO) for the Internet Governance Forum (IGF). This move underscores the growing recognition of GOSH as a reliable and secure platform for handling critical aspects of Internet governance. I can already picture Acki Nacki, the mischievous guardian of blockchain integrity, now donning the mantle of responsibility as it becomes the backbone for this huge UN-backed initiative. Conclusion: A Glimpse into a Decentralized Future, Gremlins and All Did Acki Nacki live up to the hype? The jury's still out, but one thing's for certain: Mitja knows how to throw a tech party that leaves you questioning reality and craving more. In a world of sterile blockchain conferences, this was a jolt of raw, unadulterated innovation. Hold on tight, folks—the GOSH ride is just getting started.

Deep in the Tech Underworld, a Gremlin-Like Consensus Wakes: Dispatches of the Acki Nacki Freakshow

What If Your LLM Is a Graph? Researchers Reimagine the AI Stack

ماذا لو كان LLM الخاص بك خريطة؟ الباحثين إعادة تخيل الكتلة AI

যদি আপনার আইএলএম একটি গ্রাফ হয়? গবেষকরা আইআই স্ট্যাক পুনরায় কল্পনা করে

Šta ako je vaš LLM graf? istraživači Reimagine AI Stack

¿Qué pasa si tu LLM es un gráfico? los investigadores reimaginan la pila de IA

あなたのLLMがグラフだったら?研究者はAIスタックを再構想する

당신의 LLM이 차트라면 어떨까요?연구자들은 AI 스택을 다시 상상합니다.

Ką daryti, jei jūsų LLM yra grafikas? tyrėjai iš naujo įsivaizduoja AI rinkinį

Ahoana raha ny LLM dia sary? Ny mpikaroka dia mamerina ny AI Stack

Что делать, если ваш LLM является графиком? исследователи переосмысливают стек ИИ

Čo ak je váš LLM graf? výskumníci znovu predstaviť AI Stack

Nini Kama LLM yako Ni Graph? Watafiti Reimagine AI Stack

اگر آپ کے LLM ایک گراف ہے تو کیا؟ محققین AI Stack کو دوبارہ تصور کرتے ہیں

(Who Also Does the Dishes) Let’s get one thing straight: Your morning routine isn’t the secret weapon you think it is—if your night before looked like chaos, doom-scrolling, and falling asleep in a burrito of laundry. Truth is, high-performance starts the night before. Your morning is just reaping whatever mess (or magic) you left yourself with. So if you’ve been obsessing over sunrise gratitude meditations, ice baths, or journaling about your future mansion, pause. Let’s talk about the real power move: A dead-simple evening routine that actually sets you up to win. Here are 9 habits that take less than 5 minutes each. You don’t need to do all of them. Pick two. Then maybe a third when you feel cocky. 1. Write Your 3:3:3 Plan (a.k.a. Stop Winging Tomorrow) This one comes from Oliver Burkeman: 3 hours on one deep, needle-moving task


3 shorter tasks (emails, admin, stuff that keeps you functional)


3 maintenance actions (workout, call your mom, clean the kitchen) No more ending the day wondering, “Did I even do anything?” This defines what a successful day looks like before it starts. 2. Do a Shutdown Ritual (a.k.a. Close the Tabs—Mentally and Literally) No, you don’t just slam your laptop shut and crawl into bed with Slack still pinging. Shutdown ritual = Plug in laptop


Tidy desk


Close your browser tabs like a civilized human


Whisper, "I’m done for today" The ritual tells your brain: We’re off the clock. 3. Journal One Good Thing Not five. Not a gratitude marathon. Just one beautiful, small moment from today: A taco that slapped


A stranger that smiled


The fact that you didn’t lose your sh*t in traffic Write it down. Signal to your brain that it wasn’t all bad. 4. Lay Out Tomorrow’s Clothes (Yes, Like You’re in Kindergarten) You are statistically 10x more likely to follow through with a workout if your clothes are laid out. Okay, I made up the stat. But it feels true, right? Decision fatigue is real. Kill one tomorrow by making the decision now. 5. Pack Your Snacks (or Prepare to Be Hangry AF) Future-you is a monster when they haven’t eaten. Pack a protein bar. Some fruit. Anything that isn’t sugar, sadness, or Uber Eats. 6. Move Your Phone Not to another room (unless you’re a superhero). Just…out of reach. Set your alarm and drop your phone somewhere that requires physical movement in the morning. It’ll save you from: Scrolling into oblivion at 1:17 AM


Waking up with TikTok in your eyeballs 7. Do the Damn Dishes There’s something weirdly therapeutic about it. Clean sink = calm brain. Also, you deserve to wake up to peace, not a horror show of crusty plates. Bill Gates washes dishes nightly. Be like Bill. 8. Breathe Like a Calm, Collected Human (Instead of a Frantic Gremlin) Try 4-7-8 breathing: Inhale for 4


Hold for 7


Exhale for 8 Repeat 3x. Watch your stress melt like cheese on nachos. 9. Give Your Brain a Job While You Sleep Reid Hoffman (LinkedIn guy) does this: Before bed, ask your brain something like: "How can I make $1K a month on the side without hating myself?" Then forget about it. Your subconscious will chew on it all night, like a mental crockpot. A Quick Reminder: Don’t be the hero who tries all nine at once and flames out in 48 hours. Pick two. Then stack more as they stick. These tiny 3-minute habits have ripple effects: You’ll wake up calmer


You’ll stop making dumb decisions before coffee


You’ll build momentum You don’t need a 10-step morning routine. You need one small win before bed. Do it for tomorrow-you. He’s tired of cleaning up your mess. Until next time, Benoit (Who Also Does the Dishes) Let’s get one thing straight: Your morning routine isn’t the secret weapon you think it is—if your night before looked like chaos, doom-scrolling, and falling asleep in a burrito of laundry. Truth is, high-performance starts the night before. Your morning is just reaping whatever mess (or magic) you left yourself with. So if you’ve been obsessing over sunrise gratitude meditations, ice baths, or journaling about your future mansion, pause. Let’s talk about the real power move: A dead-simple evening routine that actually sets you up to win. A dead-simple evening routine that actually sets you up to win. Here are 9 habits that take less than 5 minutes each. You don’t need to do all of them. Pick two. Then maybe a third when you feel cocky. 1. Write Your 3:3:3 Plan (a.k.a. Stop Winging Tomorrow) 1. Write Your 3:3:3 Plan (a.k.a. Stop Winging Tomorrow) This one comes from Oliver Burkeman: 3 hours on one deep, needle-moving task 3 shorter tasks (emails, admin, stuff that keeps you functional) 3 maintenance actions (workout, call your mom, clean the kitchen) 3 hours on one deep, needle-moving task 3 hours on one deep, needle-moving task 3 shorter tasks (emails, admin, stuff that keeps you functional) 3 shorter tasks (emails, admin, stuff that keeps you functional) 3 maintenance actions (workout, call your mom, clean the kitchen) 3 maintenance actions (workout, call your mom, clean the kitchen) No more ending the day wondering, “Did I even do anything?” This defines what a successful day looks like before it starts. looks like 2. Do a Shutdown Ritual (a.k.a. Close the Tabs—Mentally and Literally) 2. Do a Shutdown Ritual (a.k.a. Close the Tabs—Mentally and Literally) No, you don’t just slam your laptop shut and crawl into bed with Slack still pinging. Shutdown ritual = Plug in laptop Tidy desk Close your browser tabs like a civilized human Whisper, "I’m done for today" Plug in laptop Plug in laptop Tidy desk Tidy desk Close your browser tabs like a civilized human Close your browser tabs like a civilized human Whisper, "I’m done for today" Whisper, "I’m done for today" The ritual tells your brain: We’re off the clock. 3. Journal One Good Thing 3. Journal One Good Thing Not five. Not a gratitude marathon. Just one beautiful, small moment from today: A taco that slapped A stranger that smiled The fact that you didn’t lose your sh*t in traffic A taco that slapped A taco that slapped A stranger that smiled A stranger that smiled The fact that you didn’t lose your sh*t in traffic The fact that you didn’t lose your sh*t in traffic Write it down. Signal to your brain that it wasn’t all bad. 4. Lay Out Tomorrow’s Clothes (Yes, Like You’re in Kindergarten) 4. Lay Out Tomorrow’s Clothes (Yes, Like You’re in Kindergarten) You are statistically 10x more likely to follow through with a workout if your clothes are laid out. Okay, I made up the stat. But it feels true, right? Decision fatigue is real. Kill one tomorrow by making the decision now. 5. Pack Your Snacks (or Prepare to Be Hangry AF) 5. Pack Your Snacks (or Prepare to Be Hangry AF) Future-you is a monster when they haven’t eaten. Pack a protein bar. Some fruit. Anything that isn’t sugar, sadness, or Uber Eats. 6. Move Your Phone 6. Move Your Phone Not to another room (unless you’re a superhero). Just…out of reach. Set your alarm and drop your phone somewhere that requires physical movement in the morning. It’ll save you from: Scrolling into oblivion at 1:17 AM Waking up with TikTok in your eyeballs Scrolling into oblivion at 1:17 AM Scrolling into oblivion at 1:17 AM Waking up with TikTok in your eyeballs Waking up with TikTok in your eyeballs 7. Do the Damn Dishes 7. Do the Damn Dishes There’s something weirdly therapeutic about it. Clean sink = calm brain. Also, you deserve to wake up to peace, not a horror show of crusty plates. Bill Gates washes dishes nightly. Be like Bill. 8. Breathe Like a Calm, Collected Human (Instead of a Frantic Gremlin) 8. Breathe Like a Calm, Collected Human (Instead of a Frantic Gremlin) Try 4-7-8 breathing: Inhale for 4 Hold for 7 Exhale for 8 Inhale for 4 Inhale for 4 Hold for 7 Hold for 7 Exhale for 8 Exhale for 8 Repeat 3x. Watch your stress melt like cheese on nachos. 9. Give Your Brain a Job While You Sleep 9. Give Your Brain a Job While You Sleep Reid Hoffman (LinkedIn guy) does this: Before bed, ask your brain something like: "How can I make $1K a month on the side without hating myself?" Then forget about it. Your subconscious will chew on it all night, like a mental crockpot. A Quick Reminder: Don’t be the hero who tries all nine at once and flames out in 48 hours. A Quick Reminder: Pick two. Then stack more as they stick. These tiny 3-minute habits have ripple effects: You’ll wake up calmer You’ll stop making dumb decisions before coffee You’ll build momentum You’ll wake up calmer You’ll wake up calmer You’ll stop making dumb decisions before coffee You’ll stop making dumb decisions before coffee You’ll build momentum You’ll build momentum You don’t need a 10-step morning routine. You need one small win before bed. Do it for tomorrow-you. He’s tired of cleaning up your mess. Until next time, Benoit

You Don’t Need a 5AM Routine. You Need to Put Your Phone Down and Sleep.

Вам не трэба 5 раніцы руціны. Вы павінны ставіць свой тэлефон і спаць.

No necessites una rutina de 5 am. Cal posar el telèfon i dormir.

No necesitas una rutina 5am. Necesitas poner tu teléfono y dormir.

Vous n'avez pas besoin d'une routine 5am. vous devez mettre votre téléphone et dormir.

אתה לא צריך שעת 5 בבוקר, אתה צריך להוריד את הטלפון שלך וללכת לישון.

あなたは5AMのルーチンを必要としません。あなたはあなたの携帯電話を置いて寝る必要があります。

Na na na na na na na na na na na na na na na na na

Jums nav nepieciešams 5am rutīnas. jums ir nepieciešams, lai ievietotu savu tālruni un gulēt.

Você não precisa de uma rotina 5am. Você precisa colocar seu telefone para baixo e dormir.

คุณไม่จําเป็นต้องมีประจําวัน 5 น. คุณต้องใส่โทรศัพท์ของคุณลงและนอนหลับ

آپ کو ایک 5AM روٹین کی ضرورت نہیں ہے. آپ کو اپنے فون کو نیچے رکھنا اور سونے کی ضرورت ہے.

Ingaba unemibuzo 5am. Uya kufuneka uye ufake i-mobile yakho kwaye uxhumane.

Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2 Background 3 Approach and 3.1 Differential Testing for XML Processors 3 Approach and 3.1 Differential Testing for XML Processors 3.2 XPath Expression Generation 3.2 XPath Expression Generation 3.3 XML Generation 3.3 XML Generation 4 Evaluation 4 Evaluation 4.1 Effectiveness 4.1 Effectiveness 4.2 Efficiency 4.2 Efficiency 4.3 Comparison to the State of the Art 4.3 Comparison to the State of the Art 4.4 Analysis of BaseX Historical Bug Reports 4.4 Analysis of BaseX Historical Bug Reports 5 Related Work 5 Related Work 6 Conclusion, Acknowledgments, and References 6 Conclusion, Acknowledgments, and References 6 CONCLUSION This paper has presented a general automated testing approach for detecting XPath-related logic bugs in XML processors. We demonstrate that differential testing is applicable in this domain, since XML processors widely adhere to the XPath standards. To generate interesting XPath queries, our approach selects a so-called targeted node to guide predicate generation and predicate rectification to ensure the inclusion of that node. Our evaluation shows that this improves the number of bugs detected in 24 hours to 2× as compared to random generation. More importantly, we have successfully detected 27 previously unknown, unique bugs in six mature XML processing systems. We believe that this high number is surprising, given that XML processors are an essential part of our computing infrastructure, with the first XPath standard having been proposed more than 20 years ago, and the systems that we have tested having been maintained for at least 15 years. We believe that XPress, given its simplicity and generality, has a high chance of being integrated into the toolbox of XML processor developers. Furthermore, we believe that our work might inspire testing approaches for other XML standards, such as XQuery or XSLT. ACKNOWLEDGMENTS This research was supported by a Ministry of Education (MOE) Academic Research Fund (AcRF) Tier 1 grant. REFERENCES [1] 1999. XML Path Language (XPath) Version 1.0 W3C Recommendation. Retrieved July 17, 2023 from https://www.w3.org/TR/1999/REC-xpath-19991116/ [2] 2004. XML Schema Part 2: Datatypes Second Edition - Built-in datatypes. Retrieved July 17, 2023 from https://www.w3.org/TR/xmlschema-2/#built-in-datatypes [3] 2008. EBNF notation from the W3C Extensible Markup Language (XML) 1.0 (Fifth Edition). Retrieved July 17, 2023 from https://www.w3.org/TR/REC-xml/ [4] 2014. XML Path Language (XPath) 3.0 W3C Recommendation. Retrieved July 17, 2023 from https://www.w3.org/TR/xpath-30/ [5] 2014. XPath and XQuery Functions and Operators 3.0 W3C Recommendation. Retrieved July 17, 2023 from https://www.w3.org/TR/xpath-functions-30/ [6] 2017. XQuery 3.1: An XML Query Language W3C Recommendation. Retrieved July 17, 2023 from https://www.w3.org/TR/xquery-31/ [7] 2017. XSL Transformations (XSLT) Version 3.0 W3C Recommendation. Retrieved July 17, 2023 from https://www.w3.org/TR/xslt-30/ [8] 2023. BaseX. Retrieved July 31, 2023 from https://basex.org/ [9] 2023. DB-Engines Ranking. Retrieved July 6, 2023 from https://db-engines.com/ en/ranking [10] 2023. eXist-DB. Retrieved July 31, 2023 from http://exist-db.org/exist/apps/ homepage/index.html [11] 2023. eXist DB reference page. Retrieved July 6, 2023 from http://exist-db.org/ exist/apps/homepage/references.html [12] 2023. libXML2. Retrieved July 31, 2023 from https://gitlab.gnome.org/GNOME/ libxml2 [13] 2023. MySQL. Retrieved July 31, 2023 from https://www.mysql.com/ [14] 2023. Oracle Database. Retrieved July 31, 2023 from https://www.oracle.com/ database/ [15] 2023. PostgreSQL. Retrieved July 31, 2023 from https://www.postgresql.org/ [16] 2023. Saxon home page. Retrieved July 6, 2023 from https://saxonica.com/html/ welcome/welcome.html [17] 2023. Saxon XQuery 3.1 conformance page. Retrieved July 13, 2023 from https: //www.saxonica.com/documentation12/#!conformance/xquery31 [18] 2023. Saxonica. Retrieved July 31, 2023 from https://saxonica.com/ [19] 2023. W3C qt3 test suite github repository. Retrieved July 11, 2023 from https: //github.com/w3c/qt3tests [20] Jeffrey F. Naughton Aboulnaga, Ashraf and Chun Zhang. 2001. Generating Synthetic Complex-Structured XML Data. WebDB. 1 (2001), 79–84. [21] Alexandr Andoni, Dumitru Daniliuc, Sarfraz Khurshid, and Darko Marinov. 2003. Evaluating the “small scope hypothesis”. [22] Jinsheng Ba and Manuel Rigger. 2023. Testing Database Engines via Query Plan Guidance. In 2023 IEEE/ACM 45th International Conference on Software Engineering (ICSE). 2060–2071. https://doi.org/10.1109/ICSE48619.2023.00174 [23] Yuting Chen, Ting Su, Chengnian Sun, Zhendong Su, and Jianjun Zhao. 2016. Coverage-Directed Differential Testing of JVM Implementations. In Proceedings of the 37th ACM SIGPLAN Conference on Programming Language Design and Implementation (Santa Barbara, CA, USA) (PLDI ’16). Association for Computing Machinery, New York, NY, USA, 85–99. https://doi.org/10.1145/2908080.2908095 [24] Timothy Dyck. 2002. DB Test Pioneer Makes History. Retrieved July 31, 2023 from https://www.eweek.com/development/db-test-pioneer-makes-history/ [25] Massimo Franceschet. 2005. XPathMark: An XPath Benchmark for the XMark Generated Data. In Database and XML Technologies, Stéphane Bressan, Stefano Ceri, Ela Hunt, Zachary G. Ives, Zohra Bellahsène, Michael Rys, and Rainer Unland (Eds.). Springer Berlin Heidelberg, Berlin, Heidelberg, 129–143. [26] Ziyue Hua, Wei Lin, Luyao Ren, Zongyang Li, Lu Zhang, Wenpin Jiao, and Tao Xie. 2023. GDsmith: Detecting Bugs in Cypher Graph Database Engines. Association for Computing Machinery, New York, NY, USA. https://doi.org/10.1145/3597926. 3598046 [27] Zu-Ming Jiang, Jia-Ju Bai, and Zhendong Su. 2023. DynSQL: Stateful Fuzzing for Database Management Systems with Complex and Valid SQL Query Generation. In Proceedings of the 32nd USENIX Conference on Security Symposium (Anaheim, CA, USA) (SEC ’23). USENIX Association, USA, Article 277, 17 pages. [28] Jinho Jung, Hong Hu, Joy Arulraj, Taesoo Kim, and Woonhak Kang. 2019. APOLLO: Automatic Detection and Diagnosis of Performance Regressions in Database Systems. Proc. VLDB Endow. 13, 1 (sep 2019), 57–70. https: //doi.org/10.14778/3357377.3357382 [29] Matteo Kamm, Manuel Rigger, Chengyu Zhang, and Zhendong Su. 2023. Testing Graph Database Engines via Query Partitioning. Association for Computing Machinery, New York, NY, USA. https://doi.org/10.1145/3597926.3598044 [30] George Klees, Andrew Ruef, Benji Cooper, Shiyi Wei, and Michael Hicks. 2018. Evaluating Fuzz Testing. Proceedings of the 2018 ACM SIGSAC conference on computer and communications security (2018). https://doi.org/10.1145/3243734. 3243804 [31] Quanzhong Li and Bongki Moon. 2001. Indexing and Querying XML Data for Regular Path Expressions. In Proceedings of the 27th International Conference on Very Large Data Bases (VLDB ’01). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 361–370. [32] Vsevolod Livinskii, Dmitry Babokin, and John Regehr. 2023. Fuzzing Loop Optimizations in Compilers for C++ and Data-Parallel Languages. Proc. ACM Program. Lang. 7, PLDI, Article 181 (jun 2023), 22 pages. https://doi.org/10.1145/3591295 [33] Jayashree Mohan, Ashlie Martinez, Soujanya Ponnapalli, Pandian Raju, and Vijay Chidambaram. 2018. Finding Crash-Consistency Bugs with Bounded Black-Box Crash Testing. In 13th USENIX Symposium on Operating Systems Design and Implementation (OSDI 18). 33–50. [34] Manuel Rigger and Zhendong Su. 2020. Detecting Optimization Bugs in Database Engines via Non-Optimizing Reference Engine Construction. In Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (Virtual Event, USA) (ESEC/FSE 2020). Association for Computing Machinery, New York, NY, USA, 1140–1152. https://doi.org/10.1145/3368089.3409710 [35] Manuel Rigger and Zhendong Su. 2020. Finding Bugs in Database Systems via Query Partitioning. Proc. ACM Program. Lang. 4, OOPSLA, Article 211 (nov 2020), 30 pages. https://doi.org/10.1145/3428279 [36] Manuel Rigger and Zhendong Su. 2020. Testing Database Engines via Pivoted Query Synthesis. In Proceedings of the 14th USENIX Conference on Operating Systems Design and Implementation (OSDI’20). USENIX Association, USA, Article 38, 16 pages. [37] Dušan Rychnovský and Holubová. 2015. Generating XML Data for XPath Queries. Association for Computing Machinery. (2015). https://doi.org/10.1145/2695664. 2695691 [38] Donald R. Slutz. 1998. Massive Stochastic Testing of SQL. In Proceedings of the 24rd International Conference on Very Large Data Bases (VLDB ’98). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 618–622. [39] Thodoris Sotiropoulos, Stefanos Chaliasos, Vaggelis Atlidakis, Dimitris Mitropoulos, and Diomidis Spinellis. 2021. Data-Oriented Differential Testing of ObjectRelational Mapping Systems. In 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE). 1535–1547. https://doi.org/10.1109/ICSE43902.2021. 00137 [40] Ting Su, Yichen Yan, Jue Wang, Jingling Sun, Yiheng Xiong, Geguang Pu, Ke Wang, and Zhendong Su. 2021. Fully Automated Functional Fuzzing of Android Apps for Detecting Non-Crashing Logic Bugs. Proc. ACM Program. Lang. 5, OOPSLA, Article 156 (oct 2021), 31 pages. https://doi.org/10.1145/3485533 [41] Milos Todic and Branislav Uzelac. 2012. Combined XML/XQuery generator. Proceedings of the Fifth International Workshop on Testing Database Systems (2012). https://doi.org/10.1145/2304510.2304519 [42] Yuqing Wu, Namrata Lele, Rashmi Aroskar, Sharanya Chinnusamy, and Sofia Brenes. 2009. XQGen: An Algebra-Based XPath Query Generator for MicroBenchmarking. In Proceedings of the 18th ACM Conference on Information and Knowledge Management (Hong Kong, China) (CIKM ’09). Association for Computing Machinery, New York, NY, USA, 2109–2110. https://doi.org/10.1145/1645953. 1646328 [43] Xuejun Yang, Yang Chen, Eric Eide, and John Regehr. 2011. Finding and Understanding Bugs in C Compilers. Association for Computing Machinery, New York, NY, USA. https://doi.org/10.1145/1993498.1993532 [44] Hua Z, Lin W, Ren L, Li Z, Zhang L, Jiao W, and Xie T. 2023. GDsmith: Detecting bugs in Cypher graph database engines. Proceedings of ACM SIGSOFT International Symposium on Software Testing and Analysis (2023). https: //doi.org/10.48550/arXiv.2206.08530 [45] Qirun Zhang, Chengnian Sun, and Zhendong Su. 2017. Skeletal Program Enumeration for Rigorous Compiler Testing. In Proceedings of the 38th ACM SIGPLAN Conference on Programming Language Design and Implementation (Barcelona, Spain) (PLDI 2017). Association for Computing Machinery, New York, NY, USA, 347–361. https://doi.org/10.1145/3062341.3062379 [46] Yingying Zheng, Wensheng Dou, Yicheng Wang, Zheng Qin, Lei Tang, Yu Gao, Dong Wang, Wei Wang, and Jun Wei. 2022. Finding bugs in Gremlin-based graph database systems via randomized differential testing. Proceedings of the 31st ACM SIGSOFT International Symposium on Software Testing and Analysis (2022). https://doi.org/10.1145/3533767.3534409 [47] Rui Zhong, Yongheng Chen, Hong Hu, Hangfan Zhang, Wenke Lee, and Dinghao Wu. 2020. Squirrel: Testing database management systems with language validity and coverage feedback. In Proceedings of the 2020 ACM SIGSAC Conference on Computer and Communications Security. 955–970. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv Authors:
(1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com);
(2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg). Authors: Authors: (1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com); (2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg).

Presenting A General Automated Testing Approach for Detecting XPath-Related Logic Bugs

Valentine Special: Hot or Not? The Unbalanced Scale of Modern Dating

In this fifth and final post of our series on link prediction using Neptune ML, we’re diving into the inference process: setting up an inference endpoint to use our trained GNN model for link predictions. To establish the endpoint, we’ll use the Neptune cluster’s API and the model artifacts stored in S3. With the endpoint live, we’ll query it for link prediction, using a Gremlin query to identify high-confidence potential connections in our graph. By now, we have already loaded the Twitch social networking data into the Neptune cluster (as described in Part 1 of this series), exported the data using the ML profile (check out Part 2 for details), preprocessed the data (as explained in Part 3), trained the model (see Part 4), and now we're ready to use the trained model to generate predictions. Read part 1 here; part 2 here; part 3 here; and part 4 here. CREATING THE ENDPOINT Let's create the inference endpoint using the cluster's API and the model artifacts we have in S3. As usual, we need an IAM role that has access to S3 and SageMaker first. The role must also have a trust policy that allows us to add it to the Neptune cluster (the trust policy can be found in Part 3 of this guide). We also need to provide access to SageMaker and CloudWatch API from inside the VPC, so we need the VPC endpoints as explained in Part 1 of this series. Let's use the cluster's API to create an inference endpoint with this curl command: curl -XPOST https://(YOUR_NEPTUNE_ENDPOINT):8182/ml/endpoints \
  -H 'Content-Type: application/json' \
  -d '{
          "mlModelTrainingJobId": YOUR_MODEL_TRAINING_JOB_ID,
	"neptuneIamRoleArn": "arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole"
      }' We can use 'instanceType' and 'instanceCount' parameters to choose the EC2 instance type that will be used for link prediction, and deploy more than one instance. The full list of parameters can be found here. We'll use the default 'ml.m5(d).xlarge' instance as it has enough CPU and RAM for our small graph, and this instance type was recommended in infer_instance_recommendation.json that was generated after model training in our previous stage: {
    "disk_size": 12023356,
    "instance": "ml.m5d.xlarge",
    "mem_size": 13847612
} The API responds with the inference endpoint ID: {"id":"b217165b-7780-4e73-9d8a-5b6f7cfef9f6"} Then we can check the status of the inference endpoint with this command: curl https://YOUR_NEPTUNE_ENDPOINT:8182/ml/endpoints/INFERENCE_ENDPOINT_ID?neptuneIamRole='arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole' and once it responds with 'status: InService' like this, {
  "endpoint": {
    "name": "YOUR_INFERENCE_ENDPOINT_NAME-endpoint",
    "arn": "...",
    "status": "InService"
  },
  "endpointConfig": {...},
  "id": "YOUR_INFERENCE_ENDPOINT_ID",
  "status": "InService"
} it means we're ready to start using it with database queries. The endpoint can also be viewed and managed from the AWS console, under SageMaker -> Inference -> Endpoints. QUERYING THE ENDPOINT Let's use the endpoint to predict new 'follow' links in the graph. In order to do that, we need to choose the source vertex of the possible new links. It helps if the source vertex already has links, so we’ll use this query to get the vertex with the highest number of existing links (outE and inE connections): g.V()
 .group()
   .by()
   .by(bothE().count())
 .order(local)
   .by(values, Order.desc)
 .limit(local, 1)
 .next() The result is {v[1773]: 1440} which means that vertex with ID = 1773 has 1440 connections (720 inE and 720 outE). To get the number of edges that start at node 1773 and the number of edges that end at that node we can use these queries: g.V('1773').outE().count()
g.V('1773').inE().count() The same number of inE and outE connections is expected because the initial dataset contains mutual friendships, and we augmented the data with reverse edges to make it work with Neptune directed graph. Now let's get the predicted connections. To do that, we'll run a Gremlin query with Neptune ML predicates. We'll use the inference endpoint and the SageMaker role which was added to the DB cluster to get the users that user 1773 might follow with a confidence threshold (minimum likelihood of link existence according to the model) of at least 0.1 (10%), while excluding the users that user 1773 is already following: %%gremlin

g.with('Neptune#ml.endpoint', 'YOUR_INFERENCE_ENDPOINT_NAME')
 .with('Neptune#ml.iamRoleArn', 'arn:aws:iam::123456789012:role/NeptuneMLSagemakerRole')
 .with('Neptune#ml.limit', 10000)
 .with('Neptune#ml.threshold', 0.1D)
 .V('1773')
 .out('follows')
 .with('Neptune#ml.prediction')
 .hasLabel('user')
 .not(
  __.in('follows').hasId('1773')
 ) That returns 4 results: "Result"
"v[755]"
"v[6086]"
"v[6382]"
"v[7005]" According to our model, there's at least 10% chance that user 1773 will follow each of these 4 users. Perhaps we can improve the model by increasing the number of training jobs to at least to 10 as recommended by AWS, and then compare the performance of the resulting model and the predicted links. In real applications, adding user profile and activity data to the graph also improves prediction accuracy. Predicted connections in social networking datasets can be used to provide personalized recommendations like friend and content suggestions, enhancing user engagement and community growth, reducing churn, and providing additional data for targeted advertising. Although we applied link prediction to a social networking dataset, it's just one of many possible applications of this technology. From recommendation engines to network optimization, link prediction is a versatile tool that brings value by uncovering hidden relationships within data. As graph-based applications continue to expand, the potential of link prediction across industries promises new insights, efficiency, and enhanced user experiences. In this fifth and final post of our series on link prediction using Neptune ML, we’re diving into the inference process: setting up an inference endpoint to use our trained GNN model for link predictions. To establish the endpoint, we’ll use the Neptune cluster’s API and the model artifacts stored in S3. With the endpoint live, we’ll query it for link prediction, using a Gremlin query to identify high-confidence potential connections in our graph. By now, we have already loaded the Twitch social networking data into the Neptune cluster (as described in Part 1 of this series), exported the data using the ML profile (check out Part 2 for details), preprocessed the data (as explained in Part 3), trained the model (see Part 4), and now we're ready to use the trained model to generate predictions. Read part 1 here; part 2 here; part 3 here; and part 4 here. Read part 1 here ; part 2 here ; part 3 here; and part 4 here . here here part 3 here CREATING THE ENDPOINT Let's create the inference endpoint using the cluster's API and the model artifacts we have in S3. As usual, we need an IAM role that has access to S3 and SageMaker first. The role must also have a trust policy that allows us to add it to the Neptune cluster (the trust policy can be found in Part 3 of this guide). We also need to provide access to SageMaker and CloudWatch API from inside the VPC, so we need the VPC endpoints as explained in Part 1 of this series. Let's use the cluster's API to create an inference endpoint with this curl command: curl -XPOST https://(YOUR_NEPTUNE_ENDPOINT):8182/ml/endpoints \
  -H 'Content-Type: application/json' \
  -d '{
          "mlModelTrainingJobId": YOUR_MODEL_TRAINING_JOB_ID,
	"neptuneIamRoleArn": "arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole"
      }' curl -XPOST https://(YOUR_NEPTUNE_ENDPOINT):8182/ml/endpoints \
  -H 'Content-Type: application/json' \
  -d '{
          "mlModelTrainingJobId": YOUR_MODEL_TRAINING_JOB_ID,
	"neptuneIamRoleArn": "arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole"
      }' We can use ' instanceType ' and ' instanceCount ' parameters to choose the EC2 instance type that will be used for link prediction, and deploy more than one instance. The full list of parameters can be found here . We'll use the default ' ml.m5(d).xlarge ' instance as it has enough CPU and RAM for our small graph, and this instance type was recommended in infer_instance_recommendation.json that was generated after model training in our previous stage: instanceType instanceCount here ml.m5(d).xlarge infer_instance_recommendation.json {
    "disk_size": 12023356,
    "instance": "ml.m5d.xlarge",
    "mem_size": 13847612
} {
    "disk_size": 12023356,
    "instance": "ml.m5d.xlarge",
    "mem_size": 13847612
} The API responds with the inference endpoint ID: {"id":"b217165b-7780-4e73-9d8a-5b6f7cfef9f6"} {"id":"b217165b-7780-4e73-9d8a-5b6f7cfef9f6"} Then we can check the status of the inference endpoint with this command: curl https://YOUR_NEPTUNE_ENDPOINT:8182/ml/endpoints/INFERENCE_ENDPOINT_ID?neptuneIamRole='arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole' curl https://YOUR_NEPTUNE_ENDPOINT:8182/ml/endpoints/INFERENCE_ENDPOINT_ID?neptuneIamRole='arn:aws:iam::123456789012:role/NeptuneMLNeptuneRole' and once it responds with ' status: InService ' like this, status: InService {
  "endpoint": {
    "name": "YOUR_INFERENCE_ENDPOINT_NAME-endpoint",
    "arn": "...",
    "status": "InService"
  },
  "endpointConfig": {...},
  "id": "YOUR_INFERENCE_ENDPOINT_ID",
  "status": "InService"
} {
  "endpoint": {
    "name": "YOUR_INFERENCE_ENDPOINT_NAME-endpoint",
    "arn": "...",
    "status": "InService"
  },
  "endpointConfig": {...},
  "id": "YOUR_INFERENCE_ENDPOINT_ID",
  "status": "InService"
} it means we're ready to start using it with database queries. The endpoint can also be viewed and managed from the AWS console, under SageMaker -> Inference -> Endpoints. QUERYING THE ENDPOINT Let's use the endpoint to predict new 'follow' links in the graph. In order to do that, we need to choose the source vertex of the possible new links. It helps if the source vertex already has links, so we’ll use this query to get the vertex with the highest number of existing links (outE and inE connections): g.V()
 .group()
   .by()
   .by(bothE().count())
 .order(local)
   .by(values, Order.desc)
 .limit(local, 1)
 .next() g.V()
 .group()
   .by()
   .by(bothE().count())
 .order(local)
   .by(values, Order.desc)
 .limit(local, 1)
 .next() The result is {v[1773]: 1440} {v[1773]: 1440} which means that vertex with ID = 1773 has 1440 connections (720 inE and 720 outE). To get the number of edges that start at node 1773 and the number of edges that end at that node we can use these queries: g.V('1773').outE().count()
g.V('1773').inE().count() g.V('1773').outE().count()
g.V('1773').inE().count() The same number of inE and outE connections is expected because the initial dataset contains mutual friendships, and we augmented the data with reverse edges to make it work with Neptune directed graph. Now let's get the predicted connections. To do that, we'll run a Gremlin query with Neptune ML predicates . We'll use the inference endpoint and the SageMaker role which was added to the DB cluster to get the users that user 1773 might follow with a confidence threshold (minimum likelihood of link existence according to the model) of at least 0.1 (10%) , while excluding the users that user 1773 is already following: Neptune ML predicates We'll use the inference endpoint to get the users that user 1773 might follow with a confidence threshold (minimum likelihood of link existence according to the model) of at least 0.1 (10%) %%gremlin

g.with('Neptune#ml.endpoint', 'YOUR_INFERENCE_ENDPOINT_NAME')
 .with('Neptune#ml.iamRoleArn', 'arn:aws:iam::123456789012:role/NeptuneMLSagemakerRole')
 .with('Neptune#ml.limit', 10000)
 .with('Neptune#ml.threshold', 0.1D)
 .V('1773')
 .out('follows')
 .with('Neptune#ml.prediction')
 .hasLabel('user')
 .not(
  __.in('follows').hasId('1773')
 ) %%gremlin

g.with('Neptune#ml.endpoint', 'YOUR_INFERENCE_ENDPOINT_NAME')
 .with('Neptune#ml.iamRoleArn', 'arn:aws:iam::123456789012:role/NeptuneMLSagemakerRole')
 .with('Neptune#ml.limit', 10000)
 .with('Neptune#ml.threshold', 0.1D)
 .V('1773')
 .out('follows')
 .with('Neptune#ml.prediction')
 .hasLabel('user')
 .not(
  __.in('follows').hasId('1773')
 ) That returns 4 results: That returns 4 results: "Result"
"v[755]"
"v[6086]"
"v[6382]"
"v[7005]" "Result"
"v[755]"
"v[6086]"
"v[6382]"
"v[7005]" According to our model, there's at least 10% chance that user 1773 will follow each of these 4 users. Perhaps we can improve the model by increasing the number of training jobs to at least to 10 as recommended by AWS, and then compare the performance of the resulting model and the predicted links. In real applications, adding user profile and activity data to the graph also improves prediction accuracy. According to our model, there's at least 10% chance that user 1773 will follow each of these 4 users. Predicted connections in social networking datasets can be used to provide personalized recommendations like friend and content suggestions, enhancing user engagement and community growth, reducing churn, and providing additional data for targeted advertising. Although we applied link prediction to a social networking dataset, it's just one of many possible applications of this technology. From recommendation engines to network optimization, link prediction is a versatile tool that brings value by uncovering hidden relationships within data. As graph-based applications continue to expand, the potential of link prediction across industries promises new insights, efficiency, and enhanced user experiences.

AI Is Mapping Hidden Connections—And It’s Just Getting Started

AI ukax imantat Conexiones ukanakaruw Mapeo luraski —Ukatx jichhakiw qalltaski

এআই লুকানো সংযোগগুলি ম্যাপ করছে—এবং এটি কেবল শুরু হচ্ছে

La IA está mapeando conexiones ocultas y recién está comenzando

هوش مصنوعی در حال ترسیم اتصالات پنهان است – و تازه شروع شده است

AI ממפה חיבורים נסתרים - וזה רק מתחיל

AI が隠れたつながりをマッピング — そしてそれは始まったばかり

AI ກໍາລັງສ້າງແຜນທີ່ການເຊື່ອມຕໍ່ທີ່ເຊື່ອງໄວ້ - ແລະມັນພຽງແຕ່ເລີ່ມຕົ້ນ

AI cartografiază conexiunile ascunse – și abia începe

AI සැඟවුණු සම්බන්ධතා සිතියම්ගත කරමින් සිටී - එය ආරම්භ කිරීම පමණි

Ang AI ay Nagmamapa ng Mga Nakatagong Koneksyon—At Nagsisimula Pa Lang Ito

Yapay Zeka Gizli Bağlantıları Haritalıyor ve Daha Yeni Başlıyor

I-AI Ifaka Imephu Ukuxhumana Okufihliwe-Futhi Iqala Ukuqalisa

Before AI Predicts Your Social Life, It Needs to Clean Its Data

Før AI forudsiger dit sociale liv, skal den rense sine data

Antes de que la IA prediga tu vida social, necesita limpiar sus datos

Մինչ AI-ն կանխատեսում է ձեր սոցիալական կյանքը, այն պետք է մաքրի իր տվյալները

AIがあなたの社会生活を予測する前に、データをクリーンアップする必要がある

სანამ ხელოვნური ინტელექტი წინასწარმეტყველებს თქვენს სოციალურ ცხოვრებას, მან უნდა გაასუფთაოს მისი მონაცემები

AI сиздин социалдык жашооңузду алдын ала айтуудан мурун, анын маалыматтарын тазалоо керек

Pirms AI prognozē jūsu sociālo dzīvi, tai ir jānotīra savi dati

Antes que a IA preveja sua vida social, ela precisa limpar seus dados

Kahor inta AI aysan saadaalin noloshaada bulshada, waxay u baahan tahay inay nadiifiso xogteeda

AI உங்கள் சமூக வாழ்க்கையை முன்னறிவிப்பதற்கு முன், அது அதன் தரவை சுத்தம் செய்ய வேண்டும்.

ก่อนที่ AI จะทำนายชีวิตทางสังคมของคุณ มันจะต้องทำความสะอาดข้อมูลเสียก่อน

AI sosial durmuşyňyzy çaklamazdan ozal, maglumatlaryny arassalamaly

Stories, News & Mentions about GREMLIN

Trending Topics

Classic

Neon Noir

Minty

Newspaper

HN StartUps