squall

README

squall

In-memory stand-ins for cloud APIs that fail on cue, for testing code that must survive partial failure.

Code that creates cloud resources has to cope with calls that fail after they took effect: an insert that succeeded but whose response was lost, an operation that ended in an error, a retry that finds the resource already there. Those paths are where leaked resources and stuck records come from, and they are hard to reach against a real cloud. A squall server answers the same API calls the official client libraries make, keeps the resources in memory, and fails exactly the calls you choose, or a seeded random share of them, so a failure replays from its seed.

Packages

Package	What it does
fault	Decides which calls fail and how: rejected, applied with the response lost, applied with an error, an operation that ends in an error, or slow. Scripted rules and a seeded random rate.
access	Grants a set of permissions (Google Cloud IAM permissions, AWS IAM actions or Azure RBAC actions) and refuses calls that need one it lacks, with the error the cloud answers, including those a request's fields add and Azure's linked checks. Reports which permissions were needed but missing and which were granted but never used.
aws/cur	AWS Cost and Usage Reports: report definitions (put with the API's field rules, a per-account limit and a check that the S3 bucket's policy allows report delivery, describe with pages, delete).
aws/ec2	Amazon EC2: instances (run with client tokens, describe with filters and pages, start, stop, terminate, change type or which attached volumes terminating deletes), images (describe, register from snapshots, deregister), instance types, EBS volumes (create, describe, attach, detach, delete), EBS snapshots (create, copy across regions, describe with filters and pages, delete), placement groups, tags, Elastic IP addresses (allocate, describe, release, associate with an interface or instance, disassociate), network interfaces (create with client tokens, describe, delete), and VPC networking: VPCs and their DNS attributes, subnets, availability zones, internet gateways (create, attach, detach, delete), NAT gateways on Elastic IP addresses, route tables with their routes and subnet associations, DHCP options, security groups (create, describe groups and rules with filters and pages, authorize and revoke ingress and egress rules, delete) with each VPC's default group, and VPC attachments to transit gateways the test sets up, refusing deletes with dependencies as EC2 does.
aws/route53	Amazon Route 53: private hosted zones (create with caller references, get, list by name with pages, delete), their VPC associations, tags, and record sets (change in atomic batches with CREATE, UPSERT and DELETE, list with pages), and changes (get).
aws/s3	Amazon S3: general purpose buckets (create, head, list with pages and filters, delete), their tags, versioning, policy and CORS rules, and objects (put, delete one or many, list with pages, prefixes and delimiters, list versions and delete markers).
azure/arm	Azure Resource Manager: resource groups (get, create or update, list with tag filters, delete with everything in them that nothing outside uses), virtual machines (create or update, get, list, instance view, deallocate, start, delete) and their extensions (create or update, get, list, delete), Uniform scale sets (create or update, update, get, list, delete, delete instances) and their VMs, managed disks (VM OS disks included) and snapshots, with cross-region incremental snapshot copies whose completionPercent climbs to 100, Azure Compute Gallery galleries, image definitions and image versions with their replication status, virtual networks and their subnets, NAT gateways, public load balancers with backend pools and outbound rules, network security groups, network interfaces and public IP addresses, the IP configurations of scale set VMs, private DNS zones, virtual network links and A record sets, resource provider registration, Cost Management exports at a subscription's scope, Log Analytics workspaces (create or update, recover, get, list, delete with soft deletion) and the default workspace Application Insights makes for a component that names none, Cognitive Services deployments with partner model attestations, and long-running operations to poll.
azure/graph	Microsoft Graph: the organization (list), a directory's domains (list) and users (create with Microsoft Entra's name and password rules, get by ID or user principal name, delete, permanently delete once deleted), with application permissions checked as Graph documents them.
gcp/bigquery	BigQuery: datasets (insert with the API's ID and label rules, get, list with pages, patch friendly name, description and labels with etag checks, delete refusing a dataset that holds tables unless deleteContents), optionally refused in a project where a gcp/serviceusage server has not enabled the API.
gcp/compute	Compute Engine: instances (bulkInsert, insert, get, list, aggregatedList, delete, stop, start, setLabels, setMachineType, addAccessConfig, attachDisk, detachDisk), disks (get, list, aggregatedList, insert, delete), snapshots (insert, get, list, delete), zones, regional addresses (get, list, insert, delete, setLabels), resource policies, instance templates, routes, custom mode VPC networks with their default route, subnetworks with secondary ranges, routers with Cloud NAT (insert, get, patch, delete, list, aggregatedList), firewalls, global addresses including ranges for private services access, and zone, region and global operations (get, wait).
gcp/lustre	Managed Lustre: instances (create with the API's capacity, tier, ID, filesystem, root squash and maintenance rules, get, list with filters on name, state, filesystem, network, description and labels, patch labels, description, gkeSupportEnabled and capacity, delete) and their long-running operations (get, list, delete, cancel).
gcp/resourcemanager	Cloud Resource Manager: projects (get by ID or number, getIamPolicy, setIamPolicy with etag checks, testIamPermissions answered from the server's access policy), seeded by the test.
gcp/servicenetworking	Service Networking: private service connections (create, list by network, patch with force, deleteConnection), checked against a gcp/compute server's networks and allocated ranges and peering the network there, and their long-running operations (get).
gcp/serviceusage	Service Usage: services (enable, disable, get, list by state) in projects seeded by the test and named by number or ID, and their long-running operations (get).
gcp/storage	Cloud Storage: buckets (insert, get, list, patch labels, versioning and CORS, delete) and objects (multipart and media uploads, list with versions, delete by generation).

Coverage grows with use: each stand-in implements the methods someone needed, answers anything else with 404, and documents how faithful it is. Contributions of more methods and services are welcome.

Example

faults := fault.New().Add(fault.Rule{
	Method: "compute.addresses.insert",
	Nth:    1,
	Fault:  fault.LoseResponse(), // takes effect, but the caller never hears back
})
s := compute.New(compute.WithFaults(faults))
srv := httptest.NewServer(s)
defer srv.Close()

client, _ := gcompute.NewAddressesRESTClient(ctx,
	option.WithEndpoint(srv.URL), option.WithoutAuthentication())

// Run the code under test with client, then check what it left behind:
for _, a := range s.Addresses() {
	fmt.Println(a.Name, a.Address, a.Labels)
}

For a property test, inject random faults and check that every seed converges:

for seed := range uint64(100) {
	faults := fault.New(fault.Seed(seed)).Random(0.3,
		fault.Unavailable(), fault.LoseResponse(), fault.FailAfter(500))
	// ... run the code under test against compute.New(compute.WithFaults(faults))
	// and on failure report the seed and faults.Injected().
}

To check that a list of required permissions is exact, grant only it and see what the code under test needed:

policy := access.Allow(requiredPermissions...)
s := compute.New(compute.WithPolicy(policy))
// ... run the code under test against s
for _, d := range policy.Denials() {
	t.Errorf("%s on %s needs %v", d.Method, d.Resource, d.Missing)
}
if unused := policy.Unused(); len(unused) > 0 {
	t.Logf("granted but never used: %v", unused)
}

Faithfulness

Each stand-in follows the real API's requests, response shapes and error codes for the methods it covers, and is tested through the official Go client library. Where it simplifies, the package documentation says so. For Compute Engine:

• Changes to instances, disks and snapshots finish before the call that makes them returns, so the SDK's Operation.Wait returns at once. WithSettleAfter(n) keeps them in progress, with their operation RUNNING, for n more reads of the resource or its operation, or until the test calls Settle: a new instance is PROVISIONING and then STAGING, a stopping or deleted one STOPPING, a starting one STAGING, a new snapshot CREATING and a deleted one DELETING. An operations.wait finishes the operation it waits for. A change to an instance that has another in progress finishes that one first. Every other change finishes at once.
• A call that repeats an earlier call's requestId in the same project, for the same method, gets that call's operation back without taking effect again, as the real API does for retries. The server keeps the last 10,000 operations.
• Images are registered by the test with PutImage, and an image family names the newest image in it. Snapshots are taken of a READY disk with snapshots.insert, or registered with PutSnapshot; they record the disk's size, not its data, and a disk from one that is not READY fails with resourceNotReady. Machine and disk types are any well-formed name in the call's zone. Zones come from a fixed set per region, which WithZones changes.
• Networks are custom mode only, and each gets the default route to 0.0.0.0/0 that goes with it. Subnetworks, routers, firewalls and VPC_PEERING global addresses must name a network that exists (404 otherwise), and subnetwork ranges must not overlap within the network or with its allocated ranges. Deleting a network or subnetwork that something still uses, a peering included, fails with 400 resourceInUseByAnotherResource. A router patch replaces its nats, as a JSON merge patch does. A VPC_PEERING range with no address gets the first free block of its size in 10.0.0.0/8.
• Instances, routes and regional addresses may name networks and subnetworks the stand-in does not hold, and service accounts and accelerators are not checked. Fields the stand-in only records keep the JSON the caller sent, and fields it does not know are rejected with 400. Instance group managers are not supported: their aggregated list is always empty.
• A bulkInsert skips each instance whose name, or a disk it would create, already exists, and creates the rest when they number at least minCount; its operation succeeds, with createdVmCount and failedToCreateVmCount counting the two, and lists no warning for what it skipped. With fewer than minCount, its operation fails with VM_MIN_COUNT_NOT_REACHED and a RESOURCE_ALREADY_EXISTS error per conflict, and nothing is created. It takes no fixed IP addresses or disk names, and ignores locationPolicy.
• List filters support field = value and field != value comparisons joined by AND, OR and parentheses (OR binds tighter than AND), and the regular expressions field eq re and field ne re, which must match the whole field. Lists are ordered by name. Anything else, such as a wildcard or orderBy=creationTimestamp desc, is rejected with 400 rather than ignored. An aggregatedList lists only the scopes that have items.
• IP addresses come from 100.64.0.0/10 (external) and 10.0.0.0/8 (internal), never from routable ranges. An external address is unique across every project and region; an internal one is unique within its region, which stands in for its subnetwork. An instance's ephemeral external address is released when it stops; a reserved one it takes is IN_USE until it goes, and stays with it until then if the address is deleted.
• A fault rule's Resource for an insert is the resource being created, so a rule can target the insert of one name; for a bulkInsert, the first instance it names.

For Cloud Resource Manager:

• Projects are seeded with WithProject or PutProject and are always ACTIVE. A project the server does not hold answers 403 PERMISSION_DENIED, as the API does for one that does not exist.
• A setIamPolicy with a stale etag answers 409 ABORTED. Conditional bindings need policy version 3. Audit configs are rejected with 400.
• testIamPermissions answers from the server's access.Policy, not the project's own policy, so a test can check code that tests its own permissions before acting.

For BigQuery:

• Point the client at srv.URL + "/bigquery/v2/".
• Datasets hold a friendly name, a description, labels and a location, which defaults to US. Other dataset fields are rejected with 400, and projects are not checked.
• Tables are not served; AddTable puts one in a dataset, so a delete without deleteContents is refused with 400 resourceInUse.
• WithServiceUsage refuses calls in a project where a gcp/serviceusage server has not enabled bigquery.googleapis.com, with 403 accessNotConfigured.

For Service Usage:

• Projects are seeded with WithProject and start with no service enabled. A project the server does not hold, or a service outside its catalog (extended with WithService), answers 403 PERMISSION_DENIED, as the API does for one that does not exist.
• Changes finish before the call returns. Enabling an enabled service answers with the done operation operations/noop.DONE_OPERATION; disabling a disabled one is refused with FAILED_PRECONDITION. Dependencies between services are not modeled.

For Service Networking:

• WithCompute checks a connection's network and reserved ranges against a gcp/compute server and adds the servicenetworking-googleapis-com peering to the network, so the network cannot be deleted until deleteConnection. WithProjectNumber lets a network be named by project number or ID; connections name it by number, as the API does.
• Changes finish before the call returns. A second create on a network, or one naming a range that is not a VPC_PEERING address on it, ends its operation in FAILED_PRECONDITION, the first with the API's "Please use UpdateConnection." message. A patch that drops a range needs force=true. Producer services are not modeled, so ranges are never in use by them.

For Managed Lustre:

• Creates, patches and deletes finish before the call that makes them returns, so the SDK's Wait returns at once. WithSettleAfter(n) keeps the instance CREATING or DELETING (a patched one keeps its state, since the API deprecates UPDATING), with its operation not done, for n more reads of it or its operation, or until the test calls Settle. The v1 API has no FAILED state: a failed create ends its operation in an error and leaves no instance.
• Instance IDs, filesystem names, networks, labels and root squash options are checked as the API documents them, and capacityGib against its throughput tier's minimum, maximum and step, the step growing above the tier's threshold. Locations are not checked, and networks, KMS keys and quotas are not looked up. A mount point's address comes from 10.0.0.0/8.
• List filters follow AIP-160 for name, state, filesystem, network, description and labels (labels.KEY = value, labels.KEY:*, labels:KEY), with AND, OR, NOT and parentheses; anything else is rejected with 400. A patch changes labels, description, gkeSupportEnabled or capacityGib, named by updateMask, by *, or without a mask by the fields the body holds; output-only fields and unchanged ones are ignored, as AIP-203 directs. The API also changes throughput, access rules, maintenance windows, placement policy and target version, which the stand-in rejects with 400. Names match exactly, in lowerCamelCase or snake_case, and options read back the same whether their enums were sent by name or number. Error messages are the stand-in's own.
• Fault method names are the API's method IDs, such as lustre.projects.locations.instances.create, and a rule's Resource is the path below /v1/, for a create the instance it makes.

For Amazon EC2:

• Signatures are not checked. The region comes from the SigV4 credential scope, and each region keeps its own resources.
• Key pair names are not validated, and there is no default VPC: RunInstances needs a subnet or a network interface. A subnet ID the stand-in does not hold is accepted there and by CreateNetworkInterface, with any security group IDs, as before VPCs existed; in a subnet created with CreateSubnet, interfaces and instances take the subnet's VPC, zone and an address from its CIDR block, and security groups of that VPC: the ones named, which must exist there, or the VPC's default group. Images are registered by the test with PutImage, or with RegisterImage from snapshots the stand-in holds; instance types come from a small built-in set or PutInstanceType.
• Snapshots record the volume's size and encryption, not its data, and every one is the caller's own: DescribeSnapshots with an owner other than self lists none. DeleteSnapshot fails with InvalidSnapshot.InUse while a registered image maps the snapshot.
• CopySnapshot, called in the destination region, copies a completed snapshot from SourceRegion into a new snapshot there that starts pending and completes like a new snapshot. A source that region does not hold fails with InvalidSnapshot.NotFound; a pending source fails with IncorrectState (inferred: AWS documents copying only completed snapshots). The copy takes the request's description and tags, not the source's tags, reports the arbitrary volume ID vol-ffffffff, and is independent of its source: deleting either leaves the other (inferred from copies being full copies with their own ID). It is encrypted if its source is or Encrypted is set; KMS keys are accepted and not modeled, and Outpost, Local Zone and time-based copies are rejected as unknown parameters. Encrypted=false, and a KmsKeyId without Encrypted=true, are accepted, though the API reference allows neither.
• ModifyInstanceAttribute changes the instance type of a stopped instance, or, with BlockDeviceMapping, the DeleteOnTermination of volumes attached to the instance in any state, true when a mapping leaves it out; DescribeVolumes, DescribeInstances and termination follow the new value (sourced: the API reference for ModifyInstanceAttribute and BlockDeviceMapping.N). A device with no volume of the instance attached, or a VolumeId other than the one there, fails with InvalidBlockDeviceMapping and changes nothing, and both attributes in one call fail with InvalidParameterCombination (inferred: the reference says only one attribute can change at a time and names no codes).
• RunInstances records metadata, private DNS name and CPU options, and the primary interface's type (interface or efa; with an existing interface, only that interface's type) and network card index (only 0, the same launch as leaving it out); DescribeInstances returns them, with metadata options pending until the instance first settles. EFA and AMD SEV-SNP are not checked against the instance type. CPU options take up to the type's default cores and threads per core, which DescribeInstanceTypes reports, and any core count from 1 up, where EC2 accepts only the counts it lists for each type.
• Instances, volumes and snapshots move through EC2's states when describes see them. By default the first DescribeInstances, DescribeVolumes or DescribeSnapshots after a change shows the state it settles in, so the SDK's waiters return at once; WithSettleAfter holds them in transition for more describes, or until the test calls Settle. Describes are consistent at once, not eventually.
• Terminated instances stay visible, as in EC2, until the test calls ForgetInstance. Termination deletes the interface and volumes marked to delete on termination and releases the rest.
• A CreateNetworkInterface, RunInstances or CreateVolume that repeats a ClientToken with the same parameters gets what the first call created, as the real API does for retries; with other parameters it fails with IdempotentParameterMismatch.
• Filters take EC2's * and ? wildcards. Filter names and parameters the stand-in does not support are rejected with 400 rather than ignored.
• Public IP addresses come from 100.64.0.0/10 and private ones from 10.0.0.0/8, or from the subnet's CIDR block for subnets the stand-in holds. There is no address quota unless the test sets one.
• VPCs and subnets take IPv4 CIDR blocks from /16 to /28; IPv6, secondary CIDR blocks and IPAM pools are not supported. Each VPC gets a main route table with the local route, a default security group, and the region's default DHCP options set. A region has the zones WithAvailabilityZones sets, or the region with a, b and c appended, with zone IDs such as use1-az1.
• Security groups belong to a VPC. The default group allows all traffic from its members and all outbound traffic, is deleted with its VPC and cannot be deleted on its own (CannotDelete); a new group starts with the allow-all outbound rule. Rules take IPv4 and IPv6 CIDR blocks and groups in the same VPC as peers, one rule with an sgr- ID per peer. Duplicates fail with InvalidPermission.Duplicate and revoking a missing rule with InvalidPermission.NotFound, as in a nondefault VPC. Prefix lists, the flat CidrIp parameters and rule tags are not supported. Group names stand in for IDs only in DescribeSecurityGroups, where they match groups of that name in any VPC.
• Deletes fail with DependencyViolation while something depends on the resource, as in EC2: a VPC with subnets, an attached internet gateway, extra route tables or security groups other than its default one; a security group an interface or instance uses or another group's rule refers to; a subnet with network interfaces (a NAT gateway's included until it is deleted) or a transit gateway attachment; an attached internet gateway; a main or associated route table; DHCP options in use. Detaching an internet gateway while the VPC has public addresses mapped fails the same way.
• NAT gateways and transit gateway attachments move from pending to available and from deleting to deleted as describes see them, like instances, and stay visible once deleted. A NAT gateway on an Elastic IP address that is already associated, or in a VPC with no internet gateway, ends failed with EC2's failure code. Routes whose target is gone read blackhole. Transit gateways themselves are set up by the test with PutTransitGateway; the stand-in has no API to create them.
• EC2 has no long-running operations, so a failed operation fails the call before it takes effect. A fault's reason is the EC2 error code it answers with, and the generic ones become InternalError, Unavailable and RequestLimitExceeded. A fault rule's Resource is the region and ID, such as us-east-1/eipalloc-…, and for a create, the resource being created (the first instance, for RunInstances). A rejected call has no effect, so a rejected describe does not move an instance along.

For Amazon S3:

• Requests are path-style, which the SDK uses for an IP endpoint such as httptest's. Signatures are not checked; the region comes from the SigV4 credential scope. A call on a bucket in another region fails with 301 PermanentRedirect and x-amz-bucket-region, as S3 answers it.
• There is one account. PutForeignBucket holds a name for another one, so creating it fails with BucketAlreadyExists and calls on it with AccessDenied (403 for HeadBucket). Creating an owned bucket again succeeds in us-east-1 and fails with BucketAlreadyOwnedByYou elsewhere.
• ListBuckets with any parameter pages (10,000 buckets by default) and reports each bucket's region, as S3 does; with none it lists every bucket without regions. ListObjectsV2 lists each key whose newest version is not a delete marker, in UTF-8 binary order, and its continuation tokens are positions. ListObjectVersions orders by key and then newest first, and its markers are positions, so the next page still follows one whose versions were just deleted. DeleteBucket fails with BucketNotEmpty while any version or delete marker remains.
• Objects keep their size and ETag, not their data. Checksums are not verified, and multipart uploads, copies, object tags and lifecycle are not supported. Policies and CORS rules are recorded, not enforced. S3 documents in-progress multipart uploads as blocking DeleteBucket only for directory buckets.
• A fault rule's Resource is the bucket name, or bucket/key for an object call. A lost HeadBucket breaks off inside its status line, since a sent header would read as an answer.

For Amazon Route 53:

• Only private hosted zones are supported: CreateHostedZone needs a VPC. VPCs are any ID and region unless WithVPCLookup says which exist, so a test can back it with another stand-in. Signatures are not checked.
• A reused CallerReference fails with HostedZoneAlreadyExists, even after its zone is deleted. A zone name already associated with the same VPC fails with ConflictingDomainExists, on create and on AssociateVPCWithHostedZone; removing a zone's last VPC fails with LastVPCAssociation.
• Each zone starts with its SOA and NS record sets, which cannot be deleted, and DeleteHostedZone fails with HostedZoneNotEmpty while any other remains. Changes are INSYNC at once.
• A change batch applies in order and atomically. A failed one answers InvalidChangeBatch with a message per failed change: creating a set that exists, deleting one that is missing or whose TTL and values differ, a name outside the zone, a bad A or AAAA value, or a CNAME that shares its name. Record sets are simple or alias ones: set identifiers, routing policies and health checks are rejected with InvalidInput.
• Names are lowercased with a trailing dot, and other characters are \ooo escapes, so * lists as \052. Lists order by name with its labels reversed, then by zone ID or record type.
• A fault rule's Resource is the bare hosted zone ID, or the zone's name for CreateHostedZone.

For Cloud Storage:

• Point the client at srv.URL + "/storage/v1/", or set STORAGE_EMULATOR_HOST. There is one caller, which reaches every project; PutForeignBucket holds a name it cannot reach, so inserting it fails with 409 conflict and calls on it with 403 forbidden. A repeated insert of the caller's own bucket fails with 409, as the client's retry of a create that took effect does.
• Bucket names and labels follow the API's rules; locations are upper-cased but not checked. Fields the server does not model, such as lifecycle rules or retention, are rejected with 400. A patch with a field set to null clears it, and a label set to null is deleted.
• Objects keep their size, MD5 and CRC32C, which an upload's checksums must match. With versioning on, overwrites and deletes keep the old generation as noncurrent, and a bucket with any generation left cannot be deleted. Page tokens are positions, so a listing survives deletes between pages. Resumable uploads, reads, IAM policies and HMAC keys are not supported.
• A fault rule's Resource is the path below /storage/v1/, such as b/my-bucket/o/my-object, and for an insert the bucket or object it makes.

For Azure Resource Manager:

• Long-running operations finish within the call that starts them, so the SDK's pollers are done at once. WithSettleAfter(n) answers with an Azure-AsyncOperation URL instead (and a Location URL for actions and deletes, the only one for a resource group deletion), keeping the resource Creating, Updating or Deleting for n more reads of it or its operation, or until the test calls Settle. Polls carry a retry-after-ms of 10. A fault in FailOperation mode ends the operation in its error without the call taking effect.
• PUT and DELETE are idempotent, as in Azure: a retried create answers 200 with the resource, and deleting what is not there answers 204. A change to a resource with another in progress is answered as its provider answers it: network resources fail it with 429 RetryableError and virtual network links with 409 Conflict, counting the call as a read of that change, and a VM or scale set change preempts the other, which ends Canceled with OperationPreempted, or fails with 409 OperationNotAllowed if that one is a deletion. A change in a resource group being deleted fails with ResourceGroupBeingDeleted.
• Deleting a resource group deletes what nothing outside it uses. When something outside still uses a resource in it, such as a VM in another group attached to its network interface, the rest is deleted and the operation fails with 409 ResourceGroupDeletionBlocked, leaving the group with the resources in use. Each resource left is a detail whose message is the error deleting it alone answers, such as NicInUse, as Azure reports it; the status, and that the rest is deleted first, are inferred.
• Resource names, subscription IDs, locations and tags are not validated, and names match without regard to case. Fields it does not know are rejected with InvalidRequestContent; values it does not model, such as a Flexible scale set or a VM with networkInterfaceConfigurations, with BadRequest.
• A subnet's address prefixes must lie in its virtual network's address space (NetcfgSubnetRangeOutsideVnet), not overlap another subnet's (NetcfgSubnetRangesOverlap) and be no smaller than a /29; a virtual network PUT replaces its subnets, and a subnet, virtual network or NAT gateway in use cannot be deleted. Interfaces in a subnet the server holds take addresses from its prefix. A subnet the server does not hold is recorded, not checked: its private addresses come from 10.0.0.0/8, unique within the subnet ID, and public ones from 100.64.0.0/10, assigned when created; WithPrivateRange and WithPublicRange change the ranges. Once a range is used up, a change that needs an address fails with SubnetIsFull or PublicIPCountLimitReached.
• A public load balancer's outbound rule gives SNAT ports in multiples of 8 (SpecifiedAllocatedOutboundPortsForOutboundRuleIsNotAMultipleOfEight), and the write that would ask more of its frontends than 64,000 ports per public IP address, shared among the primary IP configurations in its backend pool (each scale set VM counts), fails with SpecifiedAllocatedOutboundPortsForOutboundRuleExceedsTotalNumberOfAvailablePorts: the load balancer's, or the scale set or interface write that grows the pool. A load balancer whose pool a scale set uses cannot be deleted (LoadBalancerInUseByVirtualMachineScaleSet). Load balancing rules, probes and inbound NAT rules are not modeled.
• A disk or snapshot change in progress answers 202 with the resource, showing Updating even while it is created, and polling URLs under locations/{location}/DiskOperations, the Location URL adding monitor=true; once done, the status carries the resource as properties.output. That a delete is polled the same way is inferred.
• A VM made from an image makes its OS disk a served disk, Attached and managed by the VM, Reserved while the VM is deallocated. Deleting the VM deletes the disk when its deleteOption is Delete and detaches it otherwise. Its default SKU and size, Standard_LRS and 30 GB, are inferred. An attached disk cannot be deleted (409 OperationNotAllowed). Data disks and scale set VM disks are not modeled.
• A VM extension create answers 201 with only an Azure-AsyncOperation URL under Microsoft.Compute/locations/{location}/operations, showing Creating, then Succeeded, and a delete 202 with a Location URL as well, the same path adding monitor=true, as Azure answers; that an update shows Updating is inferred. protectedSettings are recorded, as VMExtensions shows, but not returned, as Azure leaves them out. The handler does not run, so an extension never ends Failed, and its instance view is not served. A write under a missing VM fails with ParentResourceNotFound and the message Microsoft documents; its 404 status is inferred, as are 404 ResourceNotFound for a get or list under a missing VM or a get of a missing extension, and 204 for deleting either. Deleting the VM or its group deletes its extensions, as Microsoft documents deleting a parent deleting its children.
• A copy from a disk or snapshot that is missing or in another location fails with 404 NotFound. An in-region Copy snapshot shows no completionPercent. An incremental CopyStart copy, in any location, succeeds while its background copy goes on: completionPercent climbs from 0 to 100 in WithCopySteps steps (2 by default: 50, then 100), one per get of the snapshot (unless WithSettleAfter is negative) or per Settle, a pace the stand-in sets. Using it as a source before then fails with 409 Conflict "Source incremental snapshot {name} copy is still in progress", as Azure answers a disk; for a snapshot or image version it is inferred. One copy of a disk's snapshots runs at a time (409 Conflict, message inferred). Deleting the source mid-copy succeeds and the copy shows copyCompletionError CopySourceNotFound; its message is inferred. Copy order and the ZRS default for incremental snapshots are not modeled.
• Galleries, image definitions and image versions are polled under locations/{location}/capsOperations. A version is Creating, then Succeeded; $expand=ReplicationStatus shows each target region's state. A version from a copy still in progress is accepted and ends Failed, its operation and replication status giving the copy's Conflict, which is inferred. A missing source fails with 400 InvalidParameter, the code inferred. Deleting a gallery or image definition that holds others fails with 409 CannotDeleteResource, the code inferred. Replication, sharing and soft deletion are not modeled.
• A snapshot of a disk needs Microsoft.Compute/disks/beginGetAccess/action on the disk, or fails with LinkedAuthorizationFailed; linked checks on other disk, snapshot and gallery sources are not modeled.
• Scale set VMs are numbered from 0 without reuse; raising sku.capacity creates them with their interfaces and addresses, and lowering it deletes the newest. A PUT to an existing scale set that leaves out sku keeps the capacity. One that gives a sku without a capacity fails, as Azure answers one that changes the model, unless it repeats the create of a scale set made without a capacity with nothing changed since: then it succeeds and changes nothing, so a retry after a lost response converges, which is inferred from ARM PUTs being idempotent. A new scale set needs a sku with a name. One created without a capacity starts with no VMs, as a Flexible one does in Azure; for Uniform, both are inferred.
• Resource SKUs list none, so availability checks see no restrictions.
• Every resource provider namespace is Registered unless WithUnregisteredProviders names it; Azure registers only some by default, such as Microsoft.Resources and Microsoft.CostManagement but not Microsoft.CostManagementExports, as Microsoft documents. A call to a NotRegistered namespace fails with 409 MissingSubscriptionRegistration, naming it as the detail's target, as an Azure CLI recording shows. Registering answers 200 with the provider Registering and no operation to poll, as Azure answers; it becomes Registered after the reads WithSettleAfter sets (gets of the provider, and export writes it turns away) or on Settle, and at once by default. A namespace being registered takes calls, as Microsoft documents that callers need not wait for it. ProviderState and ProviderRegistrations show the states and how many register calls each got. Registering a Registered namespace leaves it Registered, which is inferred. A provider lists no resource types and unregistering is not served.
• A Cost Management export create answers 201 and an update 200, each with a new eTag, as Azure CLI recordings show. An update needs the current eTag in the body: without one it fails with 400 "Etag value required for update operation of report: {name}.Please get the latest to avoid overwrite.", as Azure answers; with a stale one, 412, which is inferred, as is its message. Cost Management errors carry the HTTP status as their code, such as 404, and end in a request ID, as Azure answers. A missing export answers 404 "No export found for {name}, Subscription: {subscription}.", as Azure answers a get; that a delete of one answers the same is inferred (the SDK accepts only 200 from a delete). An export write while Microsoft.CostManagementExports is Registering fails with 503, as Microsoft documents; the message is inferred. One while it is NotRegistered is not refused, since no source shows how Azure answers it. The destination storage account and container are recorded, not checked. The answer leaves out the destination's type, shows a null rootFolderPath when none was given, and writes recurrence dates in RFC 3339 UTC, as Azure does. Exports live at a subscription's scope, so deleting a resource group, even the destination's, leaves them. Runs, run history, $expand and other scopes are not served. Exports shows each export's definition, destination, schedule and eTag.
• A Log Analytics workspace create answers 201 and an update 200, polled through an Azure-AsyncOperation URL under Microsoft.OperationalInsights/locations/{location}/operationstatuses, and a delete 200, or 202 with a Location URL under workspaces/{name}/operationresults as well, as Azure CLI recordings (test_monitor_log_analytics_workspace_default) and the REST API's WorkspacesDelete example show; deleting a missing workspace answers 204, as Microsoft documents. A workspace shows a customerId, and the PerGB2018 SKU, 30 days of retention, public network access and the features and workspaceCapping a recording shows unless set: 30 days is the default Microsoft documents and PerGB2018 its pay-as-you-go tier, which as the default when a create names no SKU is inferred. Deleting a workspace, or its group, keeps it as deleted for 14 days, as Microsoft documents, unless the delete passes force=true; DeletedLogAnalyticsWorkspaces lists them. A forced delete of a workspace that is already deleted answers 204 and leaves it deleted, as Microsoft documents that it must be recovered first. A create with the deleted workspace's group, name and location recovers it, keeping its customerId, showing Restoring while in progress (201, as a recording of az monitor log-analytics workspace recover shows), and keeping the properties the create leaves out, as Microsoft documents that recovery brings back its configuration. A create in another location while the name is held fails with 409 Conflict; the code and message are inferred. Polling a delete's operationresults URL needs Microsoft.OperationalInsights/workspaces/operations/read, which is inferred, as Azure defines no operationresults action for workspaces. Tables, data sources, deleted workspace listing and shared keys are not served.
• An Application Insights component records and echoes properties.WorkspaceResourceId as sent, and shows IngestionMode LogAnalytics when linked unless set, as the REST API's ComponentsCreate example shows. The workspace it names is not checked. A component created without one gets a workspace made for it, as Microsoft Learn documents ("a new Log Analytics resource is created automatically"); it is DefaultWorkspace-{subscription}-{code} in the group DefaultResourceGroup-{code}, both in the component's location, where {code} is the location's laRegionCode in the Application Insights portal's region mapping, the names the Azure CLI uses for the same default (appservice/_create_util.py). Making them in Application Insights rather than the client, their location, and using the location's name as the code for one the mapping gives none are inferred. Later components reuse them, an update that leaves out the link keeps it (inferred), and they outlive the component's group, as Resources and ResourceGroups show. A deleted default workspace is recovered by the next component that needs it, as a create recovers one (inferred). No RBAC action is checked on them.
• A Cognitive Services deployment records and echoes properties.modelProviderData (industry, organizationName, countryCode), which clients send for partner models such as Anthropic's on newer api-versions than the SDK's, as the Azure-Samples Claude starter kit and an azure-rest-api-specs issue (#43610) show; the published specs do not describe it. Other members fail with InvalidRequestContent. Values are not checked, and a partner model deployed without it is not failed, though those sources report Azure failing it with AnthropicOrganizationCreationFailed. Deployments shows it.
• Fault method names are the Azure RBAC action, such as Microsoft.Compute/virtualMachines/write, and a rule's Resource is the resource ID in the call's path.

For Microsoft Graph:

• Permanently deleting a deleted user (DELETE /directory/deletedItems/{id}) answers 204 and moves it from DeletedUsers to PurgedUsers, as Microsoft documents. It needs User.DeleteRestore.All, the only application permission Graph lists for a user; Directory.ReadWrite.All is documented as not allowing user deletion, and without it the call fails with the stand-in's usual 403 Authorization_RequestDenied. An ID that is not a deleted user's fails with 404 Request_ResourceNotFound, the status and code a public report records for the call; the message is inferred.
• GET /organization lists one organization, as Graph documents, in the shape of its example: WithOrganization sets its ID (the tenant ID), display name, countryLetterCode and defaultUsageLocation, which default to 00000000-0000-0000-0000-000000000001, Squall, US and US, and verifiedDomains are the verified domains WithDomains sets. It needs Organization.Read.All, Directory.Read.All, Organization.ReadWrite.All or Directory.ReadWrite.All, as Graph documents. Each verified domain's capabilities and type are inferred.
• The initial domain, which domains and the organization's verified domains mark with isInitial, is the directory's onmicrosoft.com domain, whichever domain is the default, as Microsoft Entra documents; that it is the first such domain WithDomains sets is inferred.

Requirements

Go 1.27 or later. The module requires nothing beyond the standard library; the tests that use the official client libraries live in the separate sdktest module.

Contributing

See CONTRIBUTING.md. Security issues: SECURITY.md.

License

Apache 2.0.