ESEN

Private object ingestion with expiration (TTL)

A workload uploads files to a private bucket, keyless, over a private network, and files delete themselves after N days.

Problem

You need to receive temporary files (exports, user uploads, ETL staging) that must not be exposed or pile up. Requirements: zero public access, zero static keys, private-network traffic, permissions scoped to the ingestion prefix, and automatic deletion after N days.

The topology is the same on every cloud; what changes is what implements each role.

Amazon Web Services

Private object ingestion with expiration (TTL) — AWSgets tokenPUT objectdeletes after N daysrestricts originProducer workloadSecretless identityIAM Role (ECS taskrole)Private access ≈S3 Gateway VPCEndpoint1 gotcha in this roleObject storageS3 bucket2 gotchas in this roleExpiration policyS3 Lifecycle rule3 gotchas in this roleData perimeterBucket policy1 gotcha in this role

Implementation

Secretless identity
IAM Role (ECS task role)s3:PutObject scoped to bucket/incoming/*Coming from Azure: there this role is User-assigned Managed Identity.Coming from GCP: there this role is Service Account attached to the workload.
Private access
S3 Gateway VPC Endpoint≈attached to the private route tableIt's a route-table entry, not a private IP: it isn't reachable from on-premises or another region (that takes an Interface Endpoint).Coming from Azure: there this role is Private Endpoint (subresource blob). It isn't a Private Endpoint: it's a free route-table entry that isn't reachable from on-premises or another region. The real equivalent is an Interface Endpoint, billed hourly.Coming from GCP: there this role is Private Google Access.
Object storage
S3 bucketBlock Public Access + BucketOwnerEnforced + SSE-KMSComing from Azure: there this role is Storage Account (StorageV2) + container.Coming from GCP: there this role is Cloud Storage bucket.
Expiration policy
S3 Lifecycle ruleExpiration.Days + AbortIncompleteMultipartUploadComing from Azure: there this role is Lifecycle management policy. The filter is just the prefix inside the bucket, and billing stops at the expiration date even if physical deletion takes days.Coming from GCP: there this role is Object Lifecycle Management.
Data perimeter
Bucket policyDeny if aws:SourceVpce ≠ endpointComing from Azure: there this role is Storage firewall.Coming from GCP: there this role is Bucket IP filtering (optional).

Cost model

Scales to zero

No fixed cost

No fixed cost: the Gateway Endpoint is free and the AWS-managed KMS key (aws/s3) has no monthly fee; Bucket Keys reduce the KMS calls that are billed per request.

Monthly floor with no traffic

USD 0/ mo

USD 0: no components with a fixed cost.

Fine print · 6 gotchas

Security● verified 34 days ago◌ no external reviewer

Block Public Access and disabled ACLs are default on new buckets (but declare them anyway)

By default, Object Ownership is BucketOwnerEnforced and all ACLs are disabled. AWS announced that, starting in April 2023, S3 would enable Block Public Access and disable ACLs on all new buckets, with no change for existing ones. Still declare them in IaC: declaring the state prevents drift if someone edits them in the console.

⚑ Report outdated
IAM● verified 34 days ago◌ no external reviewer

A Deny on aws:SourceVpce locks you out too

An explicit Deny beats any Allow, including the admin's. If the policy denies all s3:* outside the endpoint, the console and the CI pipeline (outside the VPC) lose access, and you can end up unable to edit the bucket policy itself. This module limits the Deny to object operations and lets you exclude break-glass principals via aws:PrincipalArn.

⚑ Report outdated
Networking● verified 34 days ago◌ no external reviewer

The Gateway Endpoint is free, but only works within the VPC and region

It's a route table entry with no hourly or per-GB charge, which is why it's the default for S3. But it's not reachable from on-premises (VPN/Direct Connect) or from another region. For those cases you need an Interface Endpoint, which is billed hourly per AZ and is the real equivalent of Azure Private Endpoint.

⚑ Report outdated
Cost● verified 34 days ago◌ no external reviewer

Expiration is asynchronous, but billing stops at the expiration date

S3 queues expirations and processes them asynchronously: anywhere from days to weeks can pass between the expiration date and the actual deletion. You're not charged for the expiration or for the storage time of an object that has already expired.

⚑ Report outdated
Cost● verified 34 days ago◌ no external reviewer

On a versioned bucket, Expiration deletes nothing

With versioning on, the Expiration action only adds a delete marker and the version becomes noncurrent, which is still billed. You also need NoncurrentVersionExpiration (and to clean orphaned delete markers). This pattern leaves the bucket unversioned on purpose: the TTL is the desired behavior.

⚑ Report outdated

Terraform

aws ~> 6.67✓ fmt · validate · tflint · checkov · terraform testGitHub ↗
main.tf
# Ingesta privada de objetos con expiración (TTL) — AWS
#
# Bucket S3 privado, cifrado con SSE-KMS, accesible por Gateway Endpoint,
# con escritura acotada a un prefijo y expiración a los N días.

# --- Red ----------------------------------------------------------------------

resource "aws_vpc" "this" {
  #checkov:skip=CKV2_AWS_11:Flow logs fuera de alcance del patrón (se cubren en el patrón de networking base).
  cidr_block           = var.vpc_cidr
  enable_dns_support   = true
  enable_dns_hostnames = true

  tags = { Name = "vpc-${var.name}" }
}

# Sin reglas: nada usa el SG default (CIS 5.4).
resource "aws_default_security_group" "this" {
  vpc_id = aws_vpc.this.id
}

resource "aws_route_table" "private" {
  vpc_id = aws_vpc.this.id

  tags = { Name = "rt-${var.name}-private" }
}

# Gateway endpoint: entrada en la route table, sin costo por hora ni por GB
# (gotcha: gateway-endpoint-scope).
resource "aws_vpc_endpoint" "s3" {
  vpc_id            = aws_vpc.this.id
  service_name      = "com.amazonaws.${var.region}.s3"
  vpc_endpoint_type = "Gateway"
  route_table_ids   = [aws_route_table.private.id]

  tags = { Name = "vpce-${var.name}-s3" }
}

# --- Bucket -------------------------------------------------------------------

resource "aws_s3_bucket" "ingest" {
  #checkov:skip=CKV_AWS_21:Sin versioning a propósito: con versioning, Expiration solo agrega un delete marker (gotcha: versioning-delete-marker).
  #checkov:skip=CKV_AWS_144:Replicación cross-region no aplica a datos efímeros con TTL.
  #checkov:skip=CKV_AWS_18:Access logging fuera de alcance; usar CloudTrail data events si se requiere auditoría.
  #checkov:skip=CKV2_AWS_62:Notificaciones de eventos fuera de alcance del patrón.
  bucket        = var.bucket_name
  force_destroy = false
}

# Default desde abril 2023 en buckets nuevos; se declara para evitar drift
# (gotcha: secure-defaults-since-2023).
resource "aws_s3_bucket_public_access_block" "ingest" {
  bucket                  = aws_s3_bucket.ingest.id
  block_public_acls       = true
  block_public_policy     = true
  ignore_public_acls      = true
  restrict_public_buckets = true
}

resource "aws_s3_bucket_ownership_controls" "ingest" {
  bucket = aws_s3_bucket.ingest.id

  rule {
    object_ownership = "BucketOwnerEnforced"
  }
}

resource "aws_s3_bucket_server_side_encryption_configuration" "ingest" {
  #checkov:skip=CKV2_AWS_67:Clave administrada por AWS (aws/s3): sin rotación manual que gestionar ni costo mensual.
  bucket = aws_s3_bucket.ingest.id

  rule {
    apply_server_side_encryption_by_default {
      sse_algorithm = "aws:kms"
    }
    # Bucket Keys: menos llamadas a KMS (que se cobran por request).
    bucket_key_enabled = true
  }
}

# --- Expiración (TTL) ---------------------------------------------------------

resource "aws_s3_bucket_lifecycle_configuration" "ingest" {
  bucket = aws_s3_bucket.ingest.id

  rule {
    id     = "expire-incoming"
    status = "Enabled"

    filter {
      prefix = var.object_prefix
    }

    expiration {
      days = var.ttl_days
    }
  }

  # Las partes huérfanas se cobran y no se listan (gotcha: abort-incomplete-multipart).
  rule {
    id     = "abort-incomplete-multipart"
    status = "Enabled"

    filter {}

    abort_incomplete_multipart_upload {
      days_after_initiation = 1
    }
  }
}

# --- Perímetro: solo TLS y solo desde el VPC endpoint -------------------------

locals {
  object_arns = "${aws_s3_bucket.ingest.arn}/*"

  # Excluye principals de break-glass del Deny por endpoint. Si la lista está vacía
  # no se agrega la condición (un ArnNotLike vacío no excluye a nadie).
  vpce_deny_condition = merge(
    { StringNotEquals = { "aws:SourceVpce" = aws_vpc_endpoint.s3.id } },
    length(var.break_glass_principal_arns) > 0 ? { ArnNotLike = { "aws:PrincipalArn" = var.break_glass_principal_arns } } : {},
  )
}

resource "aws_s3_bucket_policy" "ingest" {
  bucket = aws_s3_bucket.ingest.id

  # La public access block tiene que existir antes que la policy.
  depends_on = [aws_s3_bucket_public_access_block.ingest]

  policy = jsonencode({
    Version = "2012-10-17"
    Statement = [
      {
        Sid       = "DenyInsecureTransport"
        Effect    = "Deny"
        Principal = "*"
        Action    = "s3:*"
        Resource  = [aws_s3_bucket.ingest.arn, local.object_arns]
        Condition = { Bool = { "aws:SecureTransport" = "false" } }
      },
      {
        # Solo operaciones de objetos: la configuración del bucket sigue siendo
        # gestionable desde fuera de la VPC (gotcha: sourcevpce-lockout).
        Sid       = "DenyObjectAccessOutsideVpce"
        Effect    = "Deny"
        Principal = "*"
        Action    = ["s3:GetObject", "s3:PutObject", "s3:DeleteObject"]
        Resource  = local.object_arns
        Condition = local.vpce_deny_condition
      },
    ]
  })
}

# --- Identidad del productor --------------------------------------------------

resource "aws_iam_role" "producer" {
  name = "${var.name}-producer"

  # Ejemplo con ECS: cambiá el principal según dónde corra el workload
  # (ec2.amazonaws.com, lambda.amazonaws.com, o un OIDC provider para EKS).
  assume_role_policy = jsonencode({
    Version = "2012-10-17"
    Statement = [{
      Effect    = "Allow"
      Principal = { Service = "ecs-tasks.amazonaws.com" }
      Action    = "sts:AssumeRole"
    }]
  })
}

resource "aws_iam_role_policy" "producer_put" {
  name = "put-incoming-objects"
  role = aws_iam_role.producer.id

  # Acotado al prefijo: el equivalente de Azure necesita ABAC (gotcha: rbac-scope-is-container).
  policy = jsonencode({
    Version = "2012-10-17"
    Statement = [{
      Effect   = "Allow"
      Action   = ["s3:PutObject"]
      Resource = "${aws_s3_bucket.ingest.arn}/${var.object_prefix}*"
    }]
  })
}
variables.tf
variable "name" {
  description = "Prefijo para nombrar los recursos (ej. \"ingest-demo\")."
  type        = string

  validation {
    condition     = can(regex("^[a-z0-9-]{3,20}$", var.name))
    error_message = "name: 3-20 caracteres, minúsculas, números o guiones."
  }
}

variable "bucket_name" {
  description = "Nombre global único del bucket S3."
  type        = string

  validation {
    condition     = can(regex("^[a-z0-9][a-z0-9.-]{1,61}[a-z0-9]$", var.bucket_name))
    error_message = "bucket_name: 3-63 caracteres, minúsculas, números, puntos o guiones."
  }
}

variable "region" {
  description = "Región de AWS."
  type        = string
  default     = "us-east-1"
}

variable "ttl_days" {
  description = "Días desde la creación tras los cuales expira el objeto."
  type        = number
  default     = 7

  validation {
    condition     = var.ttl_days >= 1 && floor(var.ttl_days) == var.ttl_days
    error_message = "ttl_days debe ser un entero >= 1."
  }
}

variable "object_prefix" {
  description = "Prefijo de las keys a las que aplica el TTL (en S3 el prefijo NO incluye el bucket)."
  type        = string
  default     = "incoming/"
}

variable "vpc_cidr" {
  description = "CIDR de la VPC del workload."
  type        = string
  default     = "10.20.0.0/16"
}

variable "break_glass_principal_arns" {
  description = "ARNs (admite comodines) excluidos del Deny por VPC endpoint: consola, CI. Gotcha: sourcevpce-lockout."
  type        = list(string)
  default     = []
}

variable "tags" {
  description = "Tags aplicados a todos los recursos."
  type        = map(string)
  default     = {}
}
outputs.tf
output "bucket_arn" {
  description = "ARN del bucket."
  value       = aws_s3_bucket.ingest.arn
}

output "vpc_endpoint_id" {
  description = "ID del Gateway Endpoint (usado en la condición aws:SourceVpce)."
  value       = aws_vpc_endpoint.s3.id
}

output "producer_role_arn" {
  description = "ARN del rol que tiene que asumir el workload productor."
  value       = aws_iam_role.producer.arn
}
versions.tf
terraform {
  required_version = ">= 1.7.0"

  required_providers {
    aws = {
      source  = "hashicorp/aws"
      version = "~> 6.67"
    }
  }
}

provider "aws" {
  region = var.region

  default_tags {
    tags = var.tags
  }
}

Google Cloud

Private object ingestion with expiration (TTL) — GCPgets tokenPUT objectdeletes after N daysrestricts originProducer workloadSecretless identityService Accountattached to theworkload1 gotcha in this rolePrivate access ≈Private Google Access1 gotcha in this roleObject storageCloud Storage bucket2 gotchas in this roleExpiration policyObject LifecycleManagement2 gotchas in this roleData perimeter ≈Bucket IP filtering(optional)1 gotcha in this role

Implementation

Secretless identity
Service Account attached to the workloadroles/storage.objectCreator with a prefix IAM ConditionComing from Azure: there this role is User-assigned Managed Identity.Coming from AWS: there this role is IAM Role (ECS task role).
Private access
Private Google Access≈subnet flagReaches Google APIs without a public IP, but it isn't a private endpoint for the bucket and doesn't restrict who reaches it.Coming from Azure: there this role is Private Endpoint (subresource blob). It isn't a Private Endpoint: the bucket gets no private IP. Restricting the origin takes bucket IP filtering or VPC Service Controls.Coming from AWS: there this role is S3 Gateway VPC Endpoint.
Object storage
Cloud Storage bucketUniform bucket-level access + Public access prevention enforcedComing from Azure: there this role is Storage Account (StorageV2) + container. Soft delete is on by default (7 days): an ephemeral object is billed for an extra week unless you set it to 0.Coming from AWS: there this role is S3 bucket. Soft delete is on by default (7 days): an ephemeral object is billed for an extra week unless you set it to 0.
Expiration policy
Object Lifecycle ManagementDelete by age + AbortIncompleteMultipartUploadComing from Azure: there this role is Lifecycle management policy.Coming from AWS: there this role is S3 Lifecycle rule.
Data perimeter
Bucket IP filtering (optional)≈The full perimeter is VPC Service Controls, configured at organization level.Coming from Azure: there this role is Storage firewall.Coming from AWS: there this role is Bucket policy.

Cost model

Scales to zero

No fixed cost

No fixed cost: Private Google Access is free. Watch out for soft delete (see gotcha): it's this pattern's hidden cost on GCP.

Monthly floor with no traffic

USD 0/ mo

USD 0: no components with a fixed cost.

Fine print · 5 gotchas

Security● verified 34 days ago◌ no external reviewer

Uniform bucket-level access + public access prevention is the pair to declare

UBLA disables per-object ACLs (equivalent to BucketOwnerEnforced in S3) and is required for IAM Conditions. Public access prevention "enforced" blocks allUsers/allAuthenticatedUsers (equivalent to Block Public Access).

⚑ Report outdated
IAM● verified 34 days ago◌ no external reviewer

roles/storage.objectCreator is not enough to overwrite

Overwriting an existing object requires storage.objects.delete, which objectCreator lacks. For ingestion that's a feature (a compromised producer can't clobber files), but it breaks clients that retry with the same name. In S3, s3:PutObject does overwrite.

⚑ Report outdated
Networking● verified 34 days ago◌ no external reviewer

Private Google Access gives reach to Google APIs; filtering the bucket's origin is another control

VM instances that only have internal IP addresses can use Private Google Access to reach Google APIs and services. That solves reachability, not origin filtering. Bucket IP filtering lets you restrict access to the bucket based on client IP ranges and VPC networks, and works alongside IAM.

⚑ Report outdated
Cost● verified 34 days ago◌ no external reviewer

Soft delete is on by default (7 days) and makes ephemeral data more expensive

Every bucket has soft delete with 7-day retention unless you change it. An object that lives one hour is billed for that hour plus 7 days of soft delete. In a TTL ingestion bucket that can multiply cost. The module sets it to 0 (disabled) by default and exposes it as an explicit variable.

⚑ Report outdated

Terraform

google ~> 8.6✓ fmt · validate · tflint · checkov · terraform testGitHub ↗
main.tf
# Ingesta privada de objetos con expiración (TTL) — GCP
#
# Bucket con UBLA + public access prevention, soft delete explícito,
# acceso desde una subnet con Private Google Access, escritura acotada
# a un prefijo con IAM Conditions y borrado a los N días.

# --- Red ----------------------------------------------------------------------

resource "google_compute_network" "this" {
  name                    = "vpc-${var.name}"
  auto_create_subnetworks = false
}

# Denegar todo ingress explícitamente (las VPCs custom no traen reglas,
# pero declararlo deja la intención en el código).
resource "google_compute_firewall" "deny_all_ingress" {
  name      = "fw-${var.name}-deny-all-ingress"
  network   = google_compute_network.this.id
  direction = "INGRESS"
  priority  = 65534

  deny {
    protocol = "all"
  }

  source_ranges = ["0.0.0.0/0"]
}

resource "google_compute_subnetwork" "workloads" {
  name          = "snet-${var.name}-workloads"
  region        = var.region
  network       = google_compute_network.this.id
  ip_cidr_range = var.subnet_cidr

  # VMs/servicios sin IP pública llegan a storage.googleapis.com sin NAT.
  # No restringe quién accede al bucket (gotcha: pga-is-not-a-perimeter).
  private_ip_google_access = true

  log_config {
    aggregation_interval = "INTERVAL_10_MIN"
    flow_sampling        = 0.5
    metadata             = "INCLUDE_ALL_METADATA"
  }
}

# --- Bucket -------------------------------------------------------------------

resource "google_storage_bucket" "ingest" {
  #checkov:skip=CKV_GCP_62:Access logs fuera de alcance; usar Cloud Audit Logs (Data Access) si se requiere auditoría.
  #checkov:skip=CKV_GCP_78:Sin versioning a propósito: el TTL es el comportamiento buscado.
  name          = var.bucket_name
  location      = upper(var.region)
  storage_class = "STANDARD"
  force_destroy = false
  labels        = var.labels

  # Equivalente a BucketOwnerEnforced + Block Public Access (gotcha: ubla-and-pap).
  uniform_bucket_level_access = true
  public_access_prevention    = "enforced"

  versioning {
    enabled = false
  }

  # Default de GCS: 7 días. Acá es una decisión explícita (gotcha: soft-delete-default-7-days).
  soft_delete_policy {
    retention_duration_seconds = var.soft_delete_retention_seconds
  }

  lifecycle_rule {
    condition {
      age            = var.ttl_days
      matches_prefix = [var.object_prefix]
    }
    action {
      type = "Delete"
    }
  }

  lifecycle_rule {
    condition {
      age = 1
    }
    action {
      type = "AbortIncompleteMultipartUpload"
    }
  }

  dynamic "ip_filter" {
    for_each = var.restrict_to_vpc ? [1] : []
    content {
      mode = "Enabled"
      vpc_network_sources {
        network                = google_compute_network.this.id
        allowed_ip_cidr_ranges = [var.subnet_cidr]
      }
    }
  }
}

# --- Identidad del productor --------------------------------------------------

resource "google_service_account" "producer" {
  account_id   = "${var.name}-producer"
  display_name = "Productor de ingesta (${var.name})"
}

resource "google_storage_bucket_iam_member" "producer_create" {
  bucket = google_storage_bucket.ingest.name
  # objectCreator no puede sobrescribir (gotcha: objectcreator-cannot-overwrite).
  role   = "roles/storage.objectCreator"
  member = google_service_account.producer.member

  # Acotado al prefijo, como en S3. Requiere UBLA.
  condition {
    title       = "only-${trimsuffix(var.object_prefix, "/")}"
    description = "Solo objetos bajo ${var.object_prefix}"
    expression  = "resource.name.startsWith(\"projects/_/buckets/${var.bucket_name}/objects/${var.object_prefix}\")"
  }
}
variables.tf
variable "project_id" {
  description = "ID del proyecto de GCP."
  type        = string
}

variable "name" {
  description = "Prefijo para nombrar los recursos (ej. \"ingest-demo\")."
  type        = string

  validation {
    condition     = can(regex("^[a-z][a-z0-9-]{2,19}$", var.name))
    error_message = "name: 3-20 caracteres, empieza con letra, minúsculas, números o guiones."
  }
}

variable "bucket_name" {
  description = "Nombre global único del bucket."
  type        = string

  validation {
    condition     = can(regex("^[a-z0-9][a-z0-9_.-]{1,61}[a-z0-9]$", var.bucket_name))
    error_message = "bucket_name: 3-63 caracteres, minúsculas, números, puntos, guiones o guiones bajos."
  }
}

variable "region" {
  description = "Región de GCP (el bucket se crea regional en la misma región)."
  type        = string
  default     = "us-central1"
}

variable "ttl_days" {
  description = "Edad en días tras la cual se borra el objeto."
  type        = number
  default     = 7

  validation {
    condition     = var.ttl_days >= 1 && floor(var.ttl_days) == var.ttl_days
    error_message = "ttl_days debe ser un entero >= 1."
  }
}

variable "object_prefix" {
  description = "Prefijo de los objetos a los que aplica el TTL (en GCS el prefijo NO incluye el bucket)."
  type        = string
  default     = "incoming/"
}

variable "soft_delete_retention_seconds" {
  description = "Retención de soft delete. 0 lo deshabilita; el default de GCS es 7 días (gotcha: soft-delete-default-7-days)."
  type        = number
  default     = 0

  validation {
    condition     = var.soft_delete_retention_seconds == 0 || (var.soft_delete_retention_seconds >= 604800 && var.soft_delete_retention_seconds <= 7776000)
    error_message = "soft_delete_retention_seconds: 0 (deshabilitado) o entre 7 y 90 días (604800-7776000)."
  }
}

variable "subnet_cidr" {
  description = "CIDR de la subnet del workload."
  type        = string
  default     = "10.30.0.0/24"
}

variable "restrict_to_vpc" {
  description = "Activa bucket IP filtering para aceptar requests solo desde la VPC. Puede dejar afuera al pipeline si corre fuera de la VPC."
  type        = bool
  default     = false
}

variable "labels" {
  description = "Labels aplicados a los recursos que los soportan."
  type        = map(string)
  default     = {}
}
outputs.tf
output "bucket_url" {
  description = "URL gs:// del bucket."
  value       = google_storage_bucket.ingest.url
}

output "producer_service_account_email" {
  description = "Service Account que tiene que usar el workload productor."
  value       = google_service_account.producer.email
}
versions.tf
terraform {
  required_version = ">= 1.7.0"

  required_providers {
    google = {
      source  = "hashicorp/google"
      version = "~> 8.6"
    }
  }
}

provider "google" {
  project = var.project_id
  region  = var.region
}