acedergren/infrastructure-as-code

OCI Infrastructure as Code - Expert Knowledge

NEVER Do This

NEVER hardcode OCIDs in Terraform (breaks portability)

# WRONG - breaks when moving between regions/tenancies
resource "oci_core_instance" "web" {
  compartment_id = "ocid1.compartment.oc1..aaaaaa..."  # Hardcoded!
  subnet_id      = "ocid1.subnet.oc1.phx.bbbbbb..."   # Hardcoded!
}

# RIGHT - variables or data sources
resource "oci_core_instance" "web" {
  compartment_id = var.compartment_ocid
  subnet_id      = data.oci_core_subnet.existing.id
}

NEVER hardcode availability domain names

# WRONG - AD names are tenant-specific (fMgC: prefix differs per tenancy)
availability_domain = "fMgC:US-ASHBURN-AD-1"

# RIGHT - query dynamically
data "oci_identity_availability_domains" "ads" {
  compartment_id = var.tenancy_ocid
}
resource "oci_core_instance" "web" {
  availability_domain = data.oci_identity_availability_domains.ads.availability_domains[0].name
}

NEVER use preserve_boot_volume = true in dev/test (default behavior is true!)

# WRONG - default! Orphans boot volumes when instance is destroyed
resource "oci_core_instance" "dev" {
  # preserve_boot_volume not set = defaults to true
}

# RIGHT - explicit cleanup in dev/test
resource "oci_core_instance" "dev" {
  preserve_boot_volume = false
}

Cost impact: dev team with 10 instances cycling through testing = $50–500/month in silent orphaned volumes.

NEVER skip lifecycle blocks on production databases

# RIGHT - protect production resources from accidental destroy
resource "oci_database_autonomous_database" "prod" {
  lifecycle {
    prevent_destroy = true
    ignore_changes  = [defined_tags]  # Ignore tag edits made via console
  }
}

Without this: a mistyped terraform destroy -target deletes a production database permanently.

NEVER use count for resources that shouldn't be replaced on list reorder

# WRONG - reordering instance_names from ["web1","web2"] to ["web0","web1","web2"]
#         causes Terraform to RECREATE all instances
resource "oci_core_instance" "web" {
  count        = length(var.instance_names)
  display_name = var.instance_names[count.index]
}

# RIGHT - for_each with stable keys
resource "oci_core_instance" "web" {
  for_each     = toset(var.instance_names)
  display_name = each.value
}

NEVER store Terraform state locally for team use

# WRONG - no locking, no collaboration
terraform { backend "local" {} }

# RIGHT - OCI Object Storage with S3-compatible backend
terraform {
  backend "s3" {
    bucket                      = "terraform-state"
    key                         = "prod/terraform.tfstate"
    region                      = "us-phoenix-1"
    endpoint                    = "https://<namespace>.compat.objectstorage.us-phoenix-1.oraclecloud.com"
    skip_region_validation      = true
    skip_credentials_validation = true
    skip_metadata_api_check     = true
    use_path_style              = true
  }
}

OCI Provider Authentication Gotchas

Authentication precedence (silent override is a common footgun):

1. Explicit provider block credentials
2. TF_VAR_* environment variables
3. ~/.oci/config file (DEFAULT profile)
4. Instance Principal (auth = "InstancePrincipal")

Common mistake: setting env vars but an explicit provider block overrides them without any warning.

Instance Principal for Terraform running on OCI compute:

provider "oci" {
  auth   = "InstancePrincipal"
  region = var.region
}

Critical: the instance must be in the dynamic group BEFORE Terraform runs. If added after, auth fails with the cryptic error: "authorization failed or requested resource not found".

State Management

Fixing State Drift

State drift happens when resources are modified outside Terraform (console, CLI, API).

terraform plan    # Shows unexpected changes — identifies drift
terraform refresh # Updates state to match actual OCI reality (safe read-only op)

# For new resources created outside Terraform:
terraform import oci_core_vcn.main ocid1.vcn.oc1.phx.xxxxx

Suppress drift from console tag edits (common source of noise):

lifecycle {
  ignore_changes = [defined_tags, freeform_tags]
}

Fixing "409 Conflict — Resource Already Exists"

Cause: resource exists in OCI but not in state file (e.g., created manually or previous import failure).

terraform import oci_core_vcn.main ocid1.vcn.oc1.phx.xxxxx
terraform plan   # Should now show no changes for that resource

State File Corruption Recovery

# 1. Backup first
cp terraform.tfstate terraform.tfstate.backup

# 2. Try state pull repair
terraform state pull > recovered.tfstate
mv recovered.tfstate terraform.tfstate

# 3. If that fails, restore from Object Storage versioning
# 4. Last resort: reconstruct with terraform import for each resource

Prevention: enable Object Storage bucket versioning on the state backend.

Destroy Failures (Dependency Order)

Error: Resource still in use

OCI enforces strict dependency order on destroy: instances must be terminated before subnets, subnets before VCN, etc.

# Visualize the dependency graph
terraform graph | dot -Tpng > graph.png

# Destroy in reverse dependency order
terraform destroy -target=oci_core_instance.web
terraform destroy -target=oci_core_subnet.private
terraform destroy -target=oci_core_vcn.main

Timeouts for Long-Running OCI Resources

OCI resource provisioning times vary significantly. Default Terraform timeouts often cause false failures:

# Autonomous Database: 15-30 min to provision (default 20m is borderline)
resource "oci_database_autonomous_database" "prod" {
  timeouts {
    create = "60m"
    update = "60m"
    delete = "30m"
  }
}

# Compute: usually fast, but capacity issues can cause retries
resource "oci_core_instance" "web" {
  timeouts {
    create = "30m"
  }
}

OCI Landing Zones

Use oracle-terraform-modules/terraform-oci-landing-zones for:

• Greenfield tenancy setup requiring CIS OCI Foundations Benchmark compliance
• Multi-environment (dev/test/prod) with hub-and-spoke networking
• Centralized logging, Cloud Guard, and Security Zones

Do NOT use Landing Zone for:

• Brownfield (existing infrastructure) — too opinionated, causes state conflicts
• Simple single-app deployments — the module overhead exceeds the value

Progressive Loading Reference

Load references/oci-terraform-patterns.md when:

• Setting up provider configuration (multi-region, auth methods)
• Resource Manager stack operations via CLI
• Common resource patterns with full HCL examples (VCN, compute, ADB)
• Landing Zone module usage examples

Do NOT load for NEVER-list gotchas, lifecycle management, or state troubleshooting — this file covers those.