Permissions
Every permission qloudpulse asks for, in your AWS accounts and your Kubernetes clusters, statement by statement. This page is made from the same policy files as the scripts the product hands you, and as qloudpulse's own check of your roles, so it cannot say more or less than they do. Read it before you create anything, and check it with tools you trust.
Check it yourself
You should not have to take our word for any of this. Each of these takes minutes, and none needs anything from us.
- Ask an AI assistant. Paste a role's policy, from this page or from the Connect AWS page in the product, into ChatGPT, Claude, Gemini or the assistant your company uses, with the question below. The policies hold no secrets, only placeholders such as
<ACCOUNT_ID>, so they are safe to paste. An assistant can be wrong in either direction, so treat its answer as a second opinion and check what it flags with AWS's own tools. - Ask AWS. IAM Access Analyzer checks a policy as you create it in the IAM console and flags grants that are broad or wrong, and the IAM policy simulator answers whether a role may make a given call. Both are free.
- Watch it. Once a role exists, CloudTrail records every call made through it, under the role's name, so what qloudpulse reads is in your own account's event history.
- Take it back. Delete a role and qloudpulse's access through it ends at once. Uninstall the agent and it stops sending.
- Ask us. Write to support@qloudpulse.com about any statement, and we explain why it is there.
A question for your assistant
I am asked to create this AWS IAM role for a cloud cost management tool. Go through the policy statement by statement and tell me: 1. Every allowed action that can read stored data, such as S3 objects, database items or rows, secret values, parameters, log events or queue messages. 2. Every allowed action that can create, change or delete anything. 3. Any grant broader than reading what runs, its settings and its cost needs, such as a wildcard action, or a resource of * where the action supports a narrower one. 4. Whether the Deny statement really stops the reads it names, even if a broader policy is attached to the role later. 5. Whether the trust policy lets anyone other than the named principal, with the ExternalId, assume the role. Name the statement (its Sid) for each finding, and say plainly if you find nothing. <paste the policy here>
For the cluster agent:
I am asked to install a pod with this Kubernetes ClusterRole for a cost management tool. Tell me what it can read, whether it reaches Secrets, ConfigMaps, pod logs, exec, port-forward or nodes/proxy, and whether it can create, change or delete anything. Say plainly if you find nothing. <paste the ClusterRole here>
Who can use the roles
Every AWS role below trusts one principal, the role qloudpulse reads through, and only with your company's own ExternalId, so another qloudpulse customer cannot point qloudpulse at your account. The Connect AWS page fills in both values for your company. sts:TagSession stands apart, without the ExternalId condition: on its own it lets nobody in, and it lets the session carry tags.
The trust policy, as JSON
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "QloudpulseAssumes",
"Effect": "Allow",
"Principal": {
"AWS": "<QLOUDPULSE_ROLE_ARN>"
},
"Action": "sts:AssumeRole",
"Condition": {
"StringEquals": {
"sts:ExternalId": "<EXTERNAL_ID>"
}
}
},
{
"Sid": "QloudpulseSessionTags",
"Effect": "Allow",
"Principal": {
"AWS": "<QLOUDPULSE_ROLE_ARN>"
},
"Action": "sts:TagSession"
}
]
}
The account role: cost-collector
In each AWS account you want read, the payer included. It reads what runs, how much it is used, how it is connected, and names and settings, never contents. Its last statement denies reading what is stored, and a Deny wins over any Allow, so a broader policy attached to the role later still cannot open your files, rows, secrets or logs.
50 statements, 127 actions allowed, 52 denied. If you choose not to collect resource tags, the script leaves out tag:GetResources.
| Statement | Actions | On |
|---|---|---|
UsageMetricsThe last full hour's usage of what is billed by use (on-demand DynamoDB, NAT gateway and transit gateway bytes, load balancer capacity units, ...), from metrics AWS publishes at no charge, read inside CloudWatch's free API tier. Not GetMetricData, which is billed per metric read. No resource-level permissions exist. |
|
|
DynamoDbTablesDynamoDB: billing mode, provisioned RCU/WCU including each GSI's, and table size. |
|
|
DynamoDbTableDetailsDescribeContinuousBackups: whether point-in-time recovery is on, which bills the table's size. |
|
|
DynamoDbBackupsDynamoDB backups: each backup's size, AWS Backup's copies included (no resource-level permissions). |
|
|
EksClustersEKS: the clusters, and whether each one's Kubernetes version is in standard or extended support (DescribeClusterVersions has no resource-level permissions). |
|
|
EksClusterDetails |
|
|
Ec2InstancesEC2: the running instances and their types, the spot price a spot instance pays in its Availability Zone, and each VPC flow log's destination (flow logs delivered to CloudWatch Logs cost more than to S3). None of the calls has resource-level permissions. |
|
|
ElastiCacheNodesElastiCache: every node's type, engine and engine version, replicas included. |
|
|
NetworkAndStorageThe VPC and EC2 resources billed by the hour: EBS volumes (type, size, provisioned IOPS and throughput), NAT gateways, transit gateway attachments, site-to-site VPN connections, VPC endpoints, Client VPN endpoints and their subnet associations, and public IPv4 addresses on network interfaces or held as Elastic IPs. None of these calls has resource-level permissions. |
|
|
LoadBalancersLoad balancers: one action for Application, Network and Classic load balancers alike. |
|
|
ResolverEndpointsRoute 53 Resolver endpoints, each with its IP address count, and for the self-hosted forwarder each endpoint's addresses' subnets and the forwarding rules and their VPC associations: domain names and target addresses, which are the company's own systems. |
|
|
KinesisStreamsKinesis: each stream's mode and open shard count. |
|
|
KinesisStreamDetails |
|
|
RdsInstancesRDS: each instance's class, engine, Multi-AZ setting and allocated storage. |
|
|
RdsClustersAurora clusters: each cluster's engine and mode, members and writer, Serverless v2 scaling, storage type, backtrack window and Global Database membership, for the Aurora twins. A global cluster's ARN names no region. Configuration only: rds-data stays denied below. |
|
|
OpenSearchDomainsOpenSearch: the domains (ListDomainNames has no resource-level permissions) and each one's node classes and counts, dedicated masters, warm and cold storage, EBS and version. Never a document: es:ESHttp* is denied below. |
|
|
OpenSearchDomainDetails |
|
|
MskClustersMSK: the clusters (ListClustersV2 has no resource-level permissions) and each one's broker class and count, volume size, storage mode and version. Never a message: kafka-cluster:* is denied below. |
|
|
MskClusterDetails |
|
|
MqBrokersAmazon MQ: the brokers (ListBrokers has no resource-level permissions) and each one's engine, version, instance type, deployment mode and storage type. DescribeBroker answers with user names, which the collector does not keep; the brokers' protocols have no IAM data plane. |
|
|
MqBrokerDetails |
|
|
MonthlyFeesThe small monthly fees: secrets, customer-managed KMS keys, log group storage, alarms and directories. Names, sizes and states only. |
|
|
LogFiltersWhether a log group has a metric filter or a subscription filter, which an Infrequent Access log group cannot have. Only that a filter exists is kept, never its pattern. |
|
|
KmsKeyDetailsWhether a key is customer-managed (billed) or AWS-managed (free). |
|
|
S3BucketsS3 buckets, each with its region, whose CloudWatch storage metrics give its size. ListBuckets is a global call, allowed only at the us-east-1 endpoint. |
|
Only when |
InfraNetworkThe network, which nothing bills, for the infrastructure inventory: VPCs, subnets, route tables, security groups with their rules, network ACLs, internet gateways and VPC peerings. None of these calls has resource-level permissions. |
|
|
InfraTagsThe tags of the resources the bill names, asked by ARN (no resource-level permissions). |
|
|
InfraBucketPolicyStatusWhether a bucket's policy makes it public: IsPublic only, never the policy itself. |
|
|
S3BucketLifecycleWhich storage classes a bucket's lifecycle moves objects to, for the S3 storage-class savings: only whether rules exist and their target classes are kept, never a rule's filter. |
|
|
InfraLoadBalancerListenersWhat the internet reaches, for the internet exposure report: the listeners of internet-facing load balancers, their ports and protocols (no resource-level permissions for a describe call). |
|
|
InfraLoadBalancerTargetsWhat those listeners forward to: target groups with their protocol and port, each registered target's own port and health, and an Application Load Balancer listener's rules beyond its default action. Configuration only, never a request that passed through (no resource-level permissions). |
|
|
InfraBucketPublicAccessWhether a bucket is public by its ACL, or kept private by its Block Public Access settings: the ACL's grants and the four switches only, never an object. |
|
|
InfraAccountPublicAccessThe account's own Block Public Access settings, which override every bucket's. |
|
|
InfraPublicImagesSnapshots and images shared with everyone: EBS snapshots restorable by all and public AMIs, asked for this account's own (no resource-level permissions). |
|
|
InfraPublicDbSnapshotsManual RDS snapshots, instance and cluster, and whether their restore attribute names "all". |
|
|
InfraApisPublic API Gateway APIs: REST APIs with their endpoint type and resource policy, their methods' authorization and their stages; HTTP and WebSocket APIs with their routes' authorization and their stages. apigateway:GET reads by path, so it is allowed on these six only: never API keys, exports, SDKs or client certificates. API Gateway's ARNs name no account; they are the caller's. |
|
|
InfraAccessAnalyzersIAM Access Analyzer's analyzers, in an account that has one on (no resource-level permissions). They also find Lambda function URLs anyone can call: a URL with auth type NONE needs a function policy open to everyone, which an external access analyzer reports as a public finding. The role has no Lambda read, since lambda:ListFunctions answers with each function's environment variables; for an account with no analyzer the report still says Lambda function URLs are not read. |
|
|
InfraAccessFindingsTheir findings: each resource shared outside the account, whether it is public, and its status. Findings only: no analyzer, archive rule or finding is changed. |
|
|
EcrRepositoriesECR images nobody uses: each repository's settings, each image's digest, tags, size, push time and last recorded pull time, and whether the repository has a lifecycle policy, whose not-found answer is the finding. Names, sizes and dates only: no layer, no manifest and no login. BatchGetImage and GetDownloadUrlForLayer stay denied below, and GetAuthorizationToken is not allowed. ECR scan findings: the CVE ids, severities and package names and versions ECR's own scan found in the worst image in use of a repository, so a finding names them. What AWS already found, never a scan. |
|
|
EcrRegistryScanningThe registry's scanning rules and type, basic or enhanced, so a repository a registry rule scans is not reported as unscanned. A registry-level call with no resource-level permissions. Never changed. |
|
|
LoadBalancerSettingsApplication load balancers that could be one: each load balancer's attributes and each listener's certificate ARNs, beside the listeners, rules and target groups above. Settings only, never a request or an access log (no resource-level permissions for a describe call). |
|
|
LoadBalancerWebAclWhich web ACL protects a load balancer, since two behind different web ACLs cannot be merged: its name and rules, never a request it sampled or logged. On any resource, as the other describe calls here: the answer names a web ACL the role does not know before it asks. |
|
|
CloudFrontDistributionsCloudFront distributions: each distribution's id, aliases, origins, price class, web ACL and enabled state, for the cheaper-price-class advice and to join a web ACL's AWS WAF lines to its distribution. Configuration only, never content, logs or keys. CloudFront is global, so no region condition, and the list call has no resource-level permissions. |
|
|
EcsCapacityECS clusters' idle capacity: the clusters and their capacity providers, each container instance's registered and remaining CPU, memory and ENIs, and each service's desired and running count and launch type. Reservations only, never a task or a task definition: DescribeTasks and DescribeTaskDefinition return the containers' environment and command overrides, so they are denied below with ExecuteCommand. The service metrics come from UsageMetrics, behind the company's usage-metrics option. Free describe and list calls, at any resource, as the other describe calls here. |
|
|
EcsAutoScalingGroupsThe Auto Scaling group behind each ECS capacity provider: its minimum and desired size, the minimum a node pool's poolMinimums is for Kubernetes. No resource-level permissions for a describe call. |
|
|
SavingsReadsThe savings recommendations, in one statement since the role's inline policy is near IAM's 10,240-character limit. Free describe and list calls with no resource-level permissions for listing; settings, counts and names only. None answers with a user, a configuration value or a key: transfer:ListUsers, airflow:GetEnvironment, ec2:DescribeClientVpnConnections and ec2:DescribeLaunchTemplateVersions are left out on purpose, and the data planes are denied below. MemoryDB clusters' engine and node type, for the Valkey price. VPCs that on-premises networks reach, whose S3 interface endpoints are not advised away. Trails' event selectors and logging state, to find a second copy of management events. Ways into an instance other than its public address. Which attachments a transit gateway's route tables join, and which VPCs reach one another through it. Task ARNs per service, whose sizes the bill gives; DescribeTasks stays denied. Config recorders' mode, resource counts per type and rules; never a configuration item. EFS size per storage class, mount targets' AZs and lifecycle. Snapshots' storage tier and AWS Backup's own recovery points. Instance types' baseline and burst bandwidth, to size a NAT instance. Managed Prometheus and Grafana workspaces. Transfer Family servers' protocols, identity provider and storage. Private CAs and whether ACM issues from them. Network Firewall firewalls, policies and whether their rule groups are AWS managed. MWAA environments by name; their size comes from the bill. |
|
|
SecretPoliciesWhether a secret's resource policy names another account, which Parameter Store could not do. Secrets Manager bills its API calls, so the collector calls it only behind the usage-metrics option. The policy, never the value: GetSecretValue stays denied below. |
|
|
CarbonEmissionsAWS's estimate of the account's carbon emissions per month, region and service: what a management account answers covers its member accounts too. The AWS Sustainability API is free and has no resource-level permissions. Only the emissions: water allocation and the dimension listings are not read. |
|
|
OwnPolicyThe role's own inline policy, for the account service's role check: the collector reads it as the role and compares its actions with the ones this file grants today, so a role made by an older script shows on the company's AWS accounts page as "re-run the role script" with what it lacks, instead of failing each run on AccessDenied. This one role's one policy, nothing else in IAM; iam:GetRolePolicy is free. A role made before this statement cannot read itself, which the check reports as the same thing. |
|
|
DenyDataPlane DenyEvery data-plane read, denied: an explicit Deny stays on top, so attaching a broad read policy later still cannot open table items, objects, log events, secret values, stream records, database rows, instance user data, console output, files, snapshot blocks, metric series, private keys or who did what in CloudTrail. The role sees what runs and how much it is used, never what is stored in it. |
|
|
The whole policy of cost-collector, as JSON
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "UsageMetrics",
"Effect": "Allow",
"Action": [
"cloudwatch:GetMetricStatistics"
],
"Resource": "*"
},
{
"Sid": "DynamoDbTables",
"Effect": "Allow",
"Action": [
"dynamodb:ListTables"
],
"Resource": "*"
},
{
"Sid": "DynamoDbTableDetails",
"Effect": "Allow",
"Action": [
"dynamodb:DescribeTable",
"dynamodb:DescribeContinuousBackups"
],
"Resource": "arn:aws:dynamodb:*:<ACCOUNT_ID>:table/*"
},
{
"Sid": "DynamoDbBackups",
"Effect": "Allow",
"Action": [
"dynamodb:ListBackups"
],
"Resource": "*"
},
{
"Sid": "EksClusters",
"Effect": "Allow",
"Action": [
"eks:ListClusters",
"eks:DescribeClusterVersions"
],
"Resource": "*"
},
{
"Sid": "EksClusterDetails",
"Effect": "Allow",
"Action": [
"eks:DescribeCluster"
],
"Resource": "arn:aws:eks:*:<ACCOUNT_ID>:cluster/*"
},
{
"Sid": "Ec2Instances",
"Effect": "Allow",
"Action": [
"ec2:DescribeInstances",
"ec2:DescribeSpotPriceHistory",
"ec2:DescribeFlowLogs"
],
"Resource": "*"
},
{
"Sid": "ElastiCacheNodes",
"Effect": "Allow",
"Action": [
"elasticache:DescribeCacheClusters"
],
"Resource": "*"
},
{
"Sid": "NetworkAndStorage",
"Effect": "Allow",
"Action": [
"ec2:DescribeVolumes",
"ec2:DescribeNatGateways",
"ec2:DescribeTransitGatewayAttachments",
"ec2:DescribeVpnConnections",
"ec2:DescribeVpcEndpoints",
"ec2:DescribeClientVpnEndpoints",
"ec2:DescribeClientVpnTargetNetworks",
"ec2:DescribeNetworkInterfaces",
"ec2:DescribeAddresses"
],
"Resource": "*"
},
{
"Sid": "LoadBalancers",
"Effect": "Allow",
"Action": [
"elasticloadbalancing:DescribeLoadBalancers"
],
"Resource": "*"
},
{
"Sid": "ResolverEndpoints",
"Effect": "Allow",
"Action": [
"route53resolver:ListResolverEndpoints",
"route53resolver:ListResolverEndpointIpAddresses",
"route53resolver:ListResolverRules",
"route53resolver:ListResolverRuleAssociations"
],
"Resource": "*"
},
{
"Sid": "KinesisStreams",
"Effect": "Allow",
"Action": [
"kinesis:ListStreams"
],
"Resource": "*"
},
{
"Sid": "KinesisStreamDetails",
"Effect": "Allow",
"Action": [
"kinesis:DescribeStreamSummary"
],
"Resource": "arn:aws:kinesis:*:<ACCOUNT_ID>:stream/*"
},
{
"Sid": "RdsInstances",
"Effect": "Allow",
"Action": [
"rds:DescribeDBInstances"
],
"Resource": "arn:aws:rds:*:<ACCOUNT_ID>:db:*"
},
{
"Sid": "RdsClusters",
"Effect": "Allow",
"Action": [
"rds:DescribeDBClusters",
"rds:DescribeGlobalClusters"
],
"Resource": [
"arn:aws:rds:*:<ACCOUNT_ID>:cluster:*",
"arn:aws:rds::<ACCOUNT_ID>:global-cluster:*"
]
},
{
"Sid": "OpenSearchDomains",
"Effect": "Allow",
"Action": [
"es:ListDomainNames"
],
"Resource": "*"
},
{
"Sid": "OpenSearchDomainDetails",
"Effect": "Allow",
"Action": [
"es:DescribeDomains"
],
"Resource": "arn:aws:es:*:<ACCOUNT_ID>:domain/*"
},
{
"Sid": "MskClusters",
"Effect": "Allow",
"Action": [
"kafka:ListClustersV2"
],
"Resource": "*"
},
{
"Sid": "MskClusterDetails",
"Effect": "Allow",
"Action": [
"kafka:DescribeClusterV2"
],
"Resource": "arn:aws:kafka:*:<ACCOUNT_ID>:cluster/*"
},
{
"Sid": "MqBrokers",
"Effect": "Allow",
"Action": [
"mq:ListBrokers"
],
"Resource": "*"
},
{
"Sid": "MqBrokerDetails",
"Effect": "Allow",
"Action": [
"mq:DescribeBroker"
],
"Resource": "arn:aws:mq:*:<ACCOUNT_ID>:broker:*"
},
{
"Sid": "MonthlyFees",
"Effect": "Allow",
"Action": [
"secretsmanager:ListSecrets",
"kms:ListKeys",
"logs:DescribeLogGroups",
"cloudwatch:DescribeAlarms",
"ds:DescribeDirectories"
],
"Resource": "*"
},
{
"Sid": "LogFilters",
"Effect": "Allow",
"Action": [
"logs:DescribeMetricFilters",
"logs:DescribeSubscriptionFilters"
],
"Resource": "*"
},
{
"Sid": "KmsKeyDetails",
"Effect": "Allow",
"Action": [
"kms:DescribeKey"
],
"Resource": "arn:aws:kms:*:<ACCOUNT_ID>:key/*"
},
{
"Sid": "S3Buckets",
"Effect": "Allow",
"Action": [
"s3:ListAllMyBuckets"
],
"Resource": "*",
"Condition": {
"StringEquals": {
"aws:RequestedRegion": "us-east-1"
}
}
},
{
"Sid": "InfraNetwork",
"Effect": "Allow",
"Action": [
"ec2:DescribeVpcs",
"ec2:DescribeSubnets",
"ec2:DescribeRouteTables",
"ec2:DescribeSecurityGroups",
"ec2:DescribeNetworkAcls",
"ec2:DescribeInternetGateways",
"ec2:DescribeVpcPeeringConnections"
],
"Resource": "*"
},
{
"Sid": "InfraTags",
"Effect": "Allow",
"Action": [
"tag:GetResources"
],
"Resource": "*"
},
{
"Sid": "InfraBucketPolicyStatus",
"Effect": "Allow",
"Action": [
"s3:GetBucketPolicyStatus"
],
"Resource": "arn:aws:s3:::*"
},
{
"Sid": "S3BucketLifecycle",
"Effect": "Allow",
"Action": [
"s3:GetLifecycleConfiguration",
"s3:GetIntelligentTieringConfiguration"
],
"Resource": "arn:aws:s3:::*"
},
{
"Sid": "InfraLoadBalancerListeners",
"Effect": "Allow",
"Action": [
"elasticloadbalancing:DescribeListeners"
],
"Resource": "*"
},
{
"Sid": "InfraLoadBalancerTargets",
"Effect": "Allow",
"Action": [
"elasticloadbalancing:DescribeTargetGroups",
"elasticloadbalancing:DescribeTargetHealth",
"elasticloadbalancing:DescribeRules"
],
"Resource": "*"
},
{
"Sid": "InfraBucketPublicAccess",
"Effect": "Allow",
"Action": [
"s3:GetBucketAcl",
"s3:GetBucketPublicAccessBlock"
],
"Resource": "arn:aws:s3:::*"
},
{
"Sid": "InfraAccountPublicAccess",
"Effect": "Allow",
"Action": [
"s3:GetAccountPublicAccessBlock"
],
"Resource": "*"
},
{
"Sid": "InfraPublicImages",
"Effect": "Allow",
"Action": [
"ec2:DescribeSnapshots",
"ec2:DescribeImages"
],
"Resource": "*"
},
{
"Sid": "InfraPublicDbSnapshots",
"Effect": "Allow",
"Action": [
"rds:DescribeDBSnapshots",
"rds:DescribeDBSnapshotAttributes",
"rds:DescribeDBClusterSnapshots",
"rds:DescribeDBClusterSnapshotAttributes"
],
"Resource": "arn:aws:rds:*:<ACCOUNT_ID>:*"
},
{
"Sid": "InfraApis",
"Effect": "Allow",
"Action": [
"apigateway:GET"
],
"Resource": [
"arn:aws:apigateway:*::/restapis",
"arn:aws:apigateway:*::/restapis/*/resources",
"arn:aws:apigateway:*::/restapis/*/stages",
"arn:aws:apigateway:*::/apis",
"arn:aws:apigateway:*::/apis/*/routes",
"arn:aws:apigateway:*::/apis/*/stages"
]
},
{
"Sid": "InfraAccessAnalyzers",
"Effect": "Allow",
"Action": [
"access-analyzer:ListAnalyzers"
],
"Resource": "*"
},
{
"Sid": "InfraAccessFindings",
"Effect": "Allow",
"Action": [
"access-analyzer:ListFindings"
],
"Resource": "arn:aws:access-analyzer:*:<ACCOUNT_ID>:analyzer/*"
},
{
"Sid": "EcrRepositories",
"Effect": "Allow",
"Action": [
"ecr:DescribeRepositories",
"ecr:DescribeImages",
"ecr:GetLifecyclePolicy",
"ecr:DescribeImageScanFindings"
],
"Resource": "arn:aws:ecr:*:<ACCOUNT_ID>:repository/*"
},
{
"Sid": "EcrRegistryScanning",
"Effect": "Allow",
"Action": [
"ecr:GetRegistryScanningConfiguration"
],
"Resource": "*"
},
{
"Sid": "LoadBalancerSettings",
"Effect": "Allow",
"Action": [
"elasticloadbalancing:DescribeLoadBalancerAttributes",
"elasticloadbalancing:DescribeListenerCertificates"
],
"Resource": "*"
},
{
"Sid": "LoadBalancerWebAcl",
"Effect": "Allow",
"Action": [
"wafv2:GetWebACLForResource"
],
"Resource": "*"
},
{
"Sid": "CloudFrontDistributions",
"Effect": "Allow",
"Action": [
"cloudfront:ListDistributions"
],
"Resource": "*"
},
{
"Sid": "EcsCapacity",
"Effect": "Allow",
"Action": [
"ecs:ListClusters",
"ecs:DescribeClusters",
"ecs:ListContainerInstances",
"ecs:DescribeContainerInstances",
"ecs:ListServices",
"ecs:DescribeServices",
"ecs:DescribeCapacityProviders"
],
"Resource": "*"
},
{
"Sid": "EcsAutoScalingGroups",
"Effect": "Allow",
"Action": [
"autoscaling:DescribeAutoScalingGroups"
],
"Resource": "*"
},
{
"Sid": "SavingsReads",
"Effect": "Allow",
"Action": [
"memorydb:DescribeClusters",
"directconnect:DescribeDirectConnectGatewayAssociations",
"cloudtrail:DescribeTrails",
"cloudtrail:GetEventSelectors",
"cloudtrail:GetTrailStatus",
"ec2:DescribeInstanceConnectEndpoints",
"ssm:DescribeInstanceInformation",
"ec2:DescribeTransitGatewayRouteTables",
"ec2:GetTransitGatewayRouteTableAssociations",
"ec2:GetTransitGatewayRouteTablePropagations",
"ecs:ListTasks",
"config:DescribeConfigurationRecorders",
"config:DescribeConfigurationRecorderStatus",
"config:GetDiscoveredResourceCounts",
"config:DescribeConfigRules",
"elasticfilesystem:DescribeFileSystems",
"elasticfilesystem:DescribeMountTargets",
"elasticfilesystem:DescribeLifecycleConfiguration",
"ec2:DescribeSnapshotTierStatus",
"backup:ListRecoveryPointsByResource",
"ec2:DescribeInstanceTypes",
"aps:ListWorkspaces",
"aps:DescribeWorkspace",
"grafana:ListWorkspaces",
"transfer:ListServers",
"transfer:DescribeServer",
"acm-pca:ListCertificateAuthorities",
"acm-pca:DescribeCertificateAuthority",
"acm:ListCertificates",
"network-firewall:ListFirewalls",
"network-firewall:DescribeFirewall",
"network-firewall:DescribeFirewallPolicy",
"network-firewall:ListRuleGroups",
"airflow:ListEnvironments"
],
"Resource": "*"
},
{
"Sid": "SecretPolicies",
"Effect": "Allow",
"Action": [
"secretsmanager:GetResourcePolicy"
],
"Resource": "arn:aws:secretsmanager:*:<ACCOUNT_ID>:secret:*"
},
{
"Sid": "CarbonEmissions",
"Effect": "Allow",
"Action": [
"sustainability:GetEstimatedCarbonEmissions"
],
"Resource": "*"
},
{
"Sid": "OwnPolicy",
"Effect": "Allow",
"Action": [
"iam:GetRolePolicy"
],
"Resource": "arn:aws:iam::<ACCOUNT_ID>:role/cost-collector"
},
{
"Sid": "DenyDataPlane",
"Effect": "Deny",
"Action": [
"secretsmanager:GetSecretValue",
"secretsmanager:BatchGetSecretValue",
"kms:Decrypt",
"kms:GenerateDataKey*",
"s3:GetObject",
"s3:GetObjectVersion",
"ecr:BatchGetImage",
"ecr:GetDownloadUrlForLayer",
"dynamodb:GetItem",
"dynamodb:BatchGetItem",
"dynamodb:Query",
"dynamodb:Scan",
"dynamodb:PartiQLSelect",
"ssm:GetParameter*",
"logs:GetLogEvents",
"logs:FilterLogEvents",
"logs:StartQuery",
"logs:GetQueryResults",
"logs:StartLiveTail",
"kinesis:GetRecords",
"kinesis:GetShardIterator",
"kinesis:SubscribeToShard",
"rds:DownloadDBLogFilePortion",
"rds:DownloadCompleteDBLogFile",
"rds-data:ExecuteStatement",
"rds-data:BatchExecuteStatement",
"sqs:ReceiveMessage",
"ec2:DescribeInstanceAttribute",
"ec2:GetConsoleOutput",
"ec2:GetConsoleScreenshot",
"ec2:GetPasswordData",
"ec2:ExportClientVpnClientConfiguration",
"eks:AccessKubernetesApi",
"ecs:DescribeTasks",
"ecs:DescribeTaskDefinition",
"ecs:ExecuteCommand",
"elasticache:Connect",
"es:ESHttp*",
"kafka-cluster:*",
"memorydb:Connect",
"cloudtrail:LookupEvents",
"elasticfilesystem:ClientMount",
"ebs:ListSnapshotBlocks",
"ebs:ListChangedBlocks",
"ebs:GetSnapshotBlock",
"aps:QueryMetrics",
"aps:GetSeries",
"aps:GetLabels",
"aps:GetMetricMetadata",
"acm:ExportCertificate",
"airflow:CreateCliToken",
"airflow:CreateWebLoginToken"
],
"Resource": "*"
}
]
}
The bill's role: qloudpulse-cost-reader
In the payer account, and only once you connect the bill. It runs Athena queries over your bill's CUR 2.0 export, in the bucket the bill script makes or one you name, and reads AWS's own month forecast from Cost Explorer. Its only write is the query results Athena saves under athena-results/ in that bucket, which Athena makes as the caller. The bucket below is the default name; with a bucket of your own, the Connect AWS page shows that name instead.
9 statements, 26 actions allowed.
| Statement | Actions | On |
|---|---|---|
AthenaQuery |
|
|
GlueCatalogRead |
|
|
CurDataRead |
|
|
CurDataList |
|
Only when |
CurBucketLocation |
|
|
AthenaResultsReadWrite |
|
|
AthenaResultsList |
|
Only when |
CostForecastRead |
|
|
OwnPolicy |
|
|
The whole policy of qloudpulse-cost-reader, as JSON
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AthenaQuery",
"Effect": "Allow",
"Action": [
"athena:GetDataCatalog",
"athena:GetDatabase",
"athena:GetQueryExecution",
"athena:GetQueryResults",
"athena:GetQueryResultsStream",
"athena:GetTableMetadata",
"athena:GetWorkGroup",
"athena:ListDatabases",
"athena:ListDataCatalogs",
"athena:ListTableMetadata",
"athena:StartQueryExecution",
"athena:StopQueryExecution"
],
"Resource": [
"arn:aws:athena:eu-west-1:<ACCOUNT_ID>:workgroup/qloudpulse-cost",
"arn:aws:athena:eu-west-1:<ACCOUNT_ID>:datacatalog/*"
]
},
{
"Sid": "GlueCatalogRead",
"Effect": "Allow",
"Action": [
"glue:GetDatabase",
"glue:GetDatabases",
"glue:GetTable",
"glue:GetTables",
"glue:GetPartition",
"glue:GetPartitions"
],
"Resource": [
"arn:aws:glue:eu-west-1:<ACCOUNT_ID>:catalog",
"arn:aws:glue:eu-west-1:<ACCOUNT_ID>:database/qloudpulse_cur",
"arn:aws:glue:eu-west-1:<ACCOUNT_ID>:table/qloudpulse_cur/*"
]
},
{
"Sid": "CurDataRead",
"Effect": "Allow",
"Action": [
"s3:GetObject"
],
"Resource": "arn:aws:s3:::qloudpulse-cur-<ACCOUNT_ID>/cur2/*"
},
{
"Sid": "CurDataList",
"Effect": "Allow",
"Action": [
"s3:ListBucket"
],
"Resource": "arn:aws:s3:::qloudpulse-cur-<ACCOUNT_ID>",
"Condition": {
"StringLike": {
"s3:prefix": [
"cur2",
"cur2/*"
]
}
}
},
{
"Sid": "CurBucketLocation",
"Effect": "Allow",
"Action": [
"s3:GetBucketLocation"
],
"Resource": "arn:aws:s3:::qloudpulse-cur-<ACCOUNT_ID>"
},
{
"Sid": "AthenaResultsReadWrite",
"Effect": "Allow",
"Action": [
"s3:GetObject",
"s3:PutObject",
"s3:AbortMultipartUpload"
],
"Resource": "arn:aws:s3:::qloudpulse-cur-<ACCOUNT_ID>/athena-results/*"
},
{
"Sid": "AthenaResultsList",
"Effect": "Allow",
"Action": [
"s3:ListBucket"
],
"Resource": "arn:aws:s3:::qloudpulse-cur-<ACCOUNT_ID>",
"Condition": {
"StringLike": {
"s3:prefix": [
"athena-results",
"athena-results/*"
]
}
}
},
{
"Sid": "CostForecastRead",
"Effect": "Allow",
"Action": [
"ce:GetCostAndUsage",
"ce:GetCostForecast"
],
"Resource": "*"
},
{
"Sid": "OwnPolicy",
"Effect": "Allow",
"Action": [
"iam:GetRolePolicy"
],
"Resource": "arn:aws:iam::<ACCOUNT_ID>:role/qloudpulse-cost-reader"
}
]
}
The flow log role: qloudpulse-flow-sampler
Only if you turn on flow log sampling, in each account and region you sample, for the VPCs you choose. With the usage sampling role, it is one of two roles that change anything: it turns VPC flow logs on and off for short sampling windows, only on those VPCs and only for flow logs tagged for qloudpulse, and queries the records they write to a bucket of its own. The script fills in your VPCs where the policy shows <VPC_ID>.
10 statements, 19 actions allowed.
| Statement | Actions | On |
|---|---|---|
CreateOnTheseVpcs |
|
|
CreateOnlyTagged |
|
Only when |
TagWhatItCreates |
|
Only when |
DeleteOnlyTagged |
|
Only when |
FindItsFlowLogs |
|
|
S3Delivery |
|
|
QueryTheRecords |
|
|
ReadTheTable |
|
|
ReadTheRecords |
|
|
WriteQueryResults |
|
|
The whole policy of qloudpulse-flow-sampler, as JSON
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "CreateOnTheseVpcs",
"Effect": "Allow",
"Action": [
"ec2:CreateFlowLogs"
],
"Resource": [
"arn:aws:ec2:<REGION>:<ACCOUNT_ID>:vpc/<VPC_ID>"
]
},
{
"Sid": "CreateOnlyTagged",
"Effect": "Allow",
"Action": [
"ec2:CreateFlowLogs"
],
"Resource": "arn:aws:ec2:<REGION>:<ACCOUNT_ID>:vpc-flow-log/*",
"Condition": {
"StringEquals": {
"aws:RequestTag/Purpose": "qloudpulse-flow-sample"
}
}
},
{
"Sid": "TagWhatItCreates",
"Effect": "Allow",
"Action": [
"ec2:CreateTags"
],
"Resource": "arn:aws:ec2:<REGION>:<ACCOUNT_ID>:vpc-flow-log/*",
"Condition": {
"StringEquals": {
"ec2:CreateAction": "CreateFlowLogs"
}
}
},
{
"Sid": "DeleteOnlyTagged",
"Effect": "Allow",
"Action": [
"ec2:DeleteFlowLogs"
],
"Resource": "arn:aws:ec2:<REGION>:<ACCOUNT_ID>:vpc-flow-log/*",
"Condition": {
"StringEquals": {
"aws:ResourceTag/Purpose": "qloudpulse-flow-sample"
}
}
},
{
"Sid": "FindItsFlowLogs",
"Effect": "Allow",
"Action": [
"ec2:DescribeFlowLogs"
],
"Resource": "*"
},
{
"Sid": "S3Delivery",
"Effect": "Allow",
"Action": [
"logs:CreateLogDelivery",
"logs:DeleteLogDelivery"
],
"Resource": "*"
},
{
"Sid": "QueryTheRecords",
"Effect": "Allow",
"Action": [
"athena:GetWorkGroup",
"athena:StartQueryExecution",
"athena:GetQueryExecution",
"athena:GetQueryResults",
"athena:StopQueryExecution"
],
"Resource": "arn:aws:athena:<REGION>:<ACCOUNT_ID>:workgroup/qloudpulse-flow"
},
{
"Sid": "ReadTheTable",
"Effect": "Allow",
"Action": [
"glue:GetDatabase",
"glue:GetTable",
"glue:GetPartitions"
],
"Resource": [
"arn:aws:glue:<REGION>:<ACCOUNT_ID>:catalog",
"arn:aws:glue:<REGION>:<ACCOUNT_ID>:database/qloudpulse_flow",
"arn:aws:glue:<REGION>:<ACCOUNT_ID>:table/qloudpulse_flow/records"
]
},
{
"Sid": "ReadTheRecords",
"Effect": "Allow",
"Action": [
"s3:GetObject",
"s3:ListBucket",
"s3:GetBucketLocation"
],
"Resource": [
"arn:aws:s3:::qloudpulse-flow-<ACCOUNT_ID>-<REGION>",
"arn:aws:s3:::qloudpulse-flow-<ACCOUNT_ID>-<REGION>/AWSLogs/*"
]
},
{
"Sid": "WriteQueryResults",
"Effect": "Allow",
"Action": [
"s3:GetObject",
"s3:PutObject",
"s3:AbortMultipartUpload"
],
"Resource": "arn:aws:s3:::qloudpulse-flow-<ACCOUNT_ID>-<REGION>/athena-results/*"
}
]
}
The usage sampling role: qloudpulse-usage-sampler
Only if you turn on usage sampling, in each account whose buckets and instances may be sampled. It switches two paid AWS signals on and off for one resource at a time, while a saving needs the figure: S3 request metrics on a bucket of that account, under the configuration id qloudpulse-sample, and EC2 detailed monitoring on an instance of that account. It reads nothing; the samples are CloudWatch metrics, read through the account role. An instance tagged qloudpulse-usage-sample=never is refused to it by its own Deny.
3 statements, 4 actions allowed, 2 denied.
| Statement | Actions | On |
|---|---|---|
RequestMetricsOnThisAccountsBuckets |
|
Only when |
DetailedMonitoringOnThisAccountsInstances |
|
|
NeverOnInstancesTaggedNever Deny |
|
Only when |
The whole policy of qloudpulse-usage-sampler, as JSON
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "RequestMetricsOnThisAccountsBuckets",
"Effect": "Allow",
"Action": [
"s3:GetMetricsConfiguration",
"s3:PutMetricsConfiguration"
],
"Resource": "arn:aws:s3:::*",
"Condition": {
"StringEquals": {
"aws:ResourceAccount": "<ACCOUNT_ID>"
}
}
},
{
"Sid": "DetailedMonitoringOnThisAccountsInstances",
"Effect": "Allow",
"Action": [
"ec2:MonitorInstances",
"ec2:UnmonitorInstances"
],
"Resource": "arn:aws:ec2:*:<ACCOUNT_ID>:instance/*"
},
{
"Sid": "NeverOnInstancesTaggedNever",
"Effect": "Deny",
"Action": [
"ec2:MonitorInstances",
"ec2:UnmonitorInstances"
],
"Resource": "*",
"Condition": {
"StringEquals": {
"aws:ResourceTag/qloudpulse-usage-sample": "never"
}
}
}
]
}
Kubernetes: qloudpulse-agent
One pod per cluster, installed from the public qloudpulse-public repository, where its ClusterRole is defined. A ServiceAccount and ClusterRole from the chart or the plain YAML. It has no access to Secrets, ConfigMaps, pod logs, exec or nodes/proxy. It only reads, and it sends what each pod requests and uses out over HTTPS: nothing connects into your cluster.
| Statement | Verbs | Resources |
|---|---|---|
ReadTheCluster |
|
|
ReadUsage |
|
|
Something here you would rather we did not ask for? Write to support@qloudpulse.com. The rest of how qloudpulse works is in the docs.