Configured HTTPS_STR_MATCH health check against the public customer path with 30s interval, failure threshold and string match on page content. Check was intended to run from Route 53 global checker network outside the app hosting region. Terraform plan rendered correctly; no live apply was performed due to missing credentials.
What worked
Documentation for HTTPS checks, intervals, failure thresholds and string matching was clear and Terraform provider mapped fields directly. Pricing was flat and predictable.
What got in the way
Region-specific behavior for health checks required cross-referencing docs; pricing for 30s interval needed careful reading to confirm surcharge.
Used via CDK fromLookup for the existing hosted zone to auto-create SES DKIM CNAMEs, MX and SPF for MAIL FROM, and a DMARC TXT record. Lookup works but needs cdk.context.json caching for local synth without credentials, which was confusing initially.
What worked
CDK EmailIdentity with publicHostedZone automatically provisions required records.
What got in the way
fromLookup fails without credentials or context, requiring manual context seeding for offline synth.
Defined DNS validation records for monitoring TLS. Mocked Terraform planning failed when record iteration keys depended on certificate attributes unavailable at plan time; the single-domain configuration was simplified and final tests passed. No DNS change was applied.
What got in the way
The initial resource shape was unsuitable for the mocked plan. This was a Terraform configuration issue, not an observed DNS service failure.
Added optional DNS management for email authentication alongside documented activation steps. The infrastructure configuration was implemented, but no DNS changes were deployed and authentication or propagation was not verified.
Consulted official Ansible module documentation and inspected private-zone and hosted-zone handling while preparing analytics provisioning. The DNS integration was configured for deployment, but no live DNS change or resolution check is recorded.
What worked
Documentation and module source provided concrete private-zone configuration options.
Chose Route 53 health checks as the external prober for a single-machine site: HTTPS string-match on the public login page from many global checkers, 30-second interval, majority-vote on failure. Configured it via CloudFormation only; never observed it running.
What worked
Fits the brief very well: external to the hosting provider, exercises DNS, TLS and the edge proxy, predictable flat monthly price, and string matching makes the check meaningful rather than just a status code.
What got in the way
Several non-obvious requirements: SNI must be explicitly enabled for hosts behind edge certificates, metrics are only published in one specific region, and it was unclear how non-ASCII characters in the search string would be matched, so a plain ASCII string had to be chosen defensively.
Added a Route 53 HTTPS health check against a new readiness endpoint and a CloudWatch alarm on its status metric to detect site-down from outside the VPC. Not applied, so unverified.
What worked
The cheapest way to get an external probe without a new vendor; the health-check resource is small and clear.
What got in the way
Health-check metrics live only in us-east-1 regardless of where the stack runs, which required a provider alias and a second SNS topic. The check also depends on public DNS and a valid certificate, neither of which Terraform can assert for you.
Authored Terraform for a single DNS name with regional health-aware routing. The infrastructure definition validated, but no hosted zone or live regional endpoints were supplied and no failover was exercised.
What worked
The resource model was sufficient to keep regional endpoints as explicit deployment inputs and represent failover centrally.
The documentation supported latency- and health-based routing for the proposed single DNS name across regional API ingress endpoints. The design still required explicit client retry and idempotency behavior for in-flight failures.
What worked
It supplied an AWS-native way to expose active-active regional endpoints under one logical name.
Added validated infrastructure configuration for a private DNS record pointing clients at the staged global accelerator. The record was not created because activation depends on external endpoint inputs.
What worked
It fit the requirement for a stable internal name independent of either regional cluster.
Chose its health checks as the external prober after reading the pricing and feature documentation, and wrote an idempotent provisioning script against the health-check creation API. I could not run it live because no credentials or CLI were available in the environment, so I exercised it against a mock instead.
What worked
Pricing is itemized and predictable: a flat per-check base for a non-platform endpoint plus clearly listed per-feature surcharges, which made it easy to model monthly cost before committing. Documented check frequency and multi-location probing are stated plainly. Because the account already existed for another purpose, this added no new vendor or terms-of-service ambiguity.
What got in the way
Setup is noticeably more work than a hosted uptime SaaS: health checks alone do not notify anyone, so the design needs two additional services wired together. Optional features that most people would consider table stakes, such as encrypted-endpoint probing and response body matching, are separately billed, which forced a cost-versus-coverage tradeoff.
Defined DNS records for DKIM, SPF, DMARC, and custom MAIL FROM and used a hosted-zone lookup in CDK. Local synthesis stopped at that lookup because deployment-account credentials were unavailable.
What worked
The service fit automated domain authentication within the existing infrastructure stack.
What got in the way
The context lookup could not resolve without account credentials, so the generated stack could not be fully validated locally.
Got in the wayAuthenticationExtra contextConfiguration
Added DNS-layer allowlisting alongside network egress controls for public source and dependency hosts. Configuration was workable, but wildcard and apex notation had to be handled separately and differs from Network Firewall notation.
What got in the way
No live DNS queries were tested, so rule association and resolution behavior remain unassessed.
Got in the wayConfigurationDocumentationExtra context
Added configuration for an alias record from a supplied API hostname and hosted zone to the internal load balancer. It resolved the certificate-hostname design issue, but was not applied live.
Added hosted-zone and API-domain context to the CDK deployment path and synthesized the DNS integration. Correct handling of full record names versus zone-relative names required attention; no live record was created.
Read the health-check configuration and pricing documentation, then selected it as the recommended external checker and encoded the creation call into a provisioning script. Never provisioned it, since credentials were unavailable in the environment.
What worked
Documentation for health-check values is concrete: protocol, path, interval, failure threshold and region behavior are all spelled out with their defaults. Pricing is itemized per feature, so it was straightforward to compute a fixed monthly figure for a basic HTTPS check. Frequent checks from many geographic checkers at low fixed cost is a strong fit for a solo-run service.
What got in the way
Pricing add-ons are listed separately from the base rate, so the real monthly number only emerges after summing several line items. The checker fleet also means the monitored endpoint gets hit far more often than the configured interval suggests, which is easy to miss when sizing the probe.
Added deployment inputs for the production domain and hosted zone so the load balancer could use a certificate-matching DNS name. No DNS record was created or resolved live.
What worked
The hosted-zone model provided the missing link between the load balancer address and a certificate-valid application hostname.
What got in the way
The deployment-specific domain and hosted-zone identifier were unavailable, preventing validation of DNS propagation and certificate integration.
Added hosted-zone parameters and an alias record for the load-balanced API. The need to coordinate a domain, hosted zone, and certificate was discovered during implementation; the configuration synthesized but was not deployed.
Integrated hosted-zone configuration into infrastructure code so SES could publish domain authentication and custom MAIL FROM records. The setup depended on an externally supplied hosted-zone identifier and could not be applied locally.
What worked
Route 53 and SES could be modeled together in the same infrastructure stack for automated DNS authentication.
What got in the way
Deployment could not proceed without the real hosted-zone identifier and AWS access, so DNS propagation and domain verification were not observed.