DNS uses both UDP and TCP. Many ordinary queries begin with UDP because it has low setup overhead, while TCP is used when responses are truncated, for operations such as zone transfer, and whenever the selected DNS protocol requires a reliable stream. Modern networks must permit and monitor both transports; treating TCP port 53 as optional creates subtle resolution failures.
A sound design connects architecture to daily operations. Teams should be able to explain which component is authoritative, how a decision propagates, what the client actually experiences, what evidence is retained, and how normal policy returns after failure. These questions keep a familiar network service from becoming an unexamined dependency.
This article follows the problem-to-control-to-validation rhythm of the cited Infoblox Blog article. It does not attribute Infoblox-specific functions to ZDNS. Product statements remain within verified ZDNS DNS, DHCP, IPAM, or NACS responsibilities.
Why UDP Is Common for Ordinary Queries
UDP lets a client send a compact request without establishing a connection. This is efficient for many exchanges, but the application must handle timeout, retransmission, source validation, and the possibility that a response will not fit the available message size.
Start by naming the decision owner, the source of truth, and the maximum acceptable age of every input. A dashboard can look healthy while identity, topology, lease, policy, or resolver data is stale. Operators need an explicit unknown state and a route for resolving conflicting evidence. ZDNS DNS supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
For validation, run the normal path once and retain a baseline. Then remove one evidence source, introduce one contradictory value, and compare the decision. The record should show what became uncertain, whether the user-facing result changed, and who owns the discrepancy. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
How Truncation Triggers TCP
When a UDP response cannot carry the full answer, a DNS server can set the truncated flag. A compliant client retries using TCP to receive the complete message. Blocking that retry can make some names fail while smaller answers continue to work.
Separate requested, accepted, applied, independently observed, failed, and rolled-back states. This distinction matters whenever a controller publishes policy to resolvers, DHCP services, switches, wireless systems, or external dependencies. An API success is not proof of the result experienced by a client. ZDNS IPAM supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Issue a valid change and observe every downstream layer from a representative client. Capture timestamps and versions for the request, application, verification, and any retry. This makes a quiet propagation failure visible before it becomes a prolonged incident. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
EDNS Changes Size, Not the Need for TCP
Extension mechanisms for DNS allow larger UDP messages, but path MTU, fragmentation, middleboxes, response growth, and security policy still matter. Larger advertised buffers are not a guarantee that every path delivers a large datagram reliably.
Apply least privilege to routine changes, bulk operations, emergency overrides, key material, and deletion. High-impact actions need stronger approval, a reason, a bounded scope, and a record that remains useful during audit or incident reconstruction. ZDNS DHCP supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Use separate operator roles to attempt an ordinary edit, a bulk operation, an emergency action, and deletion. Confirm that denied attempts and approved changes both appear in the audit trail with enough context to explain the control. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
TCP Is Part of Normal DNS Operation
Resolvers and authoritative servers need robust TCP support, capacity, timeouts, and connection management. Zone transfers use TCP, and ordinary resolution may use it. Firewalls should allow intended DNS TCP flows rather than assuming they indicate an attack.
Design for partial failure. Delayed collectors, unreachable upstream services, exhausted address pools, stale caches, rejected updates, and asymmetric network paths occur more often than clean total outages. A resilient service signals degraded confidence and follows a tested fallback policy. ZDNS DNS supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Delay one dependency, reject one update, and restore services in a different order. Confirm stale-state signaling, bounded fallback, retry behavior, and reconciliation. Recovery should not silently overwrite a newer authoritative value with an older queued change. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Encrypted DNS Introduces Additional Transports
DoT carries DNS over TLS on TCP, while DoH carries DNS in HTTPS. These protocols protect transport between client and resolver but introduce resolver-governance, certificate, observability, policy, connection, and fallback considerations.
Preserve event-time history. Addresses, users, device names, attachment points, and policies change, so current state may identify the wrong system during an investigation. DHCP lease history and IPAM context are especially valuable when an address has been reused. ZDNS IPAM supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Replay a historical event after an address, hostname, or endpoint has changed. An investigator should be able to recover the event-time lease, owner, network location, policy, and relevant DNS activity without relying on today's state. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Security Controls Must Be Transport-Aware
Rate limits, reflection defenses, recursion controls, response validation, logging, and anomaly detection should understand both UDP and TCP. A control applied to only one path can create a bypass or an availability asymmetry.
Treat exceptions as governed objects with an owner, business reason, scope, compensating control, review date, and expiration. Repeated renewal is operational evidence: it may expose an unsupported device class, an unrealistic baseline, or a dependency that the normal workflow cannot satisfy. ZDNS DHCP supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Create a temporary exception, let it approach expiration, and attempt renewal. The workflow should expose its owner, age, reachability, compensating controls, and renewal history. Expiration must not depend on someone remembering an informal note. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Capacity Planning Differs by Transport
UDP consumes little per-query connection state, while TCP requires connection tracking, buffers, timeouts, concurrency, and often connection reuse. Test normal peaks, retries, slow clients, upstream latency, and failover instead of sizing from average query rate.
Publish rollback and recovery steps before rollout. Restoration must verify the client-facing outcome, reconcile changes made during the incident, remove temporary access or bypasses, and retain evidence. Service availability alone does not mean normal policy has been restored. ZDNS DNS supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Trigger rollback during a staged deployment. Verify that ordinary service returns, temporary settings disappear, and evidence remains available for review. Test the bootstrap and management paths as well as the production data path. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Troubleshoot the Complete Resolution Path
Capture query type, response size, truncation, retry, latency, selected server, and firewall outcome. Compare client, recursive, forwarder, and authoritative observations to locate where the transport changed or failed.
Measure outcomes rather than interface activity. Useful measures include decision latency, stale-data age, failed enforcement, resolution success, address conflicts, exception age, audit completeness, and mean time to safe restoration. Every metric needs a clear source and denominator. ZDNS IPAM supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Calculate each metric from source records and sample both successes and failures. Check timestamps, exclusions, duplicates, and stale records. Presentation summaries should help interpretation but must not be the only evidence of control effectiveness. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Build a Transport Regression Test
Include small and large answers, DNSSEC-related responses, UDP loss, forced truncation, TCP retry, zone transfer where authorized, DoT, DoH, and blocked-path scenarios. Validate from different sites and client classes.
Repeat the exercise after material changes to topology, identity, software, upstream connectivity, certificates, integrations, or policy. A control that worked in the original design can become a hidden dependency after an architecture change. ZDNS DHCP supplies a relevant ZDNS capability within this workflow, while identity, endpoint, firewall, routing, application, and incident-response controls retain their own responsibilities.
Close the exercise only after normal policy is confirmed from the client path and all temporary states are removed. Document the root cause, update the runbook, and add the scenario to regression testing so the same failure produces a faster and safer response next time. For dns udp or tcp, retain the expected result, source timestamps, policy or data version, client observation, and next safe action as the acceptance record.
Put the Design into a Bounded Pilot
Choose a representative but noncritical scope. Record the baseline before changing policy, then include one expected case, one unsupported or legacy case, one stale-data case, and one dependency failure. Assign observers at the client, network, service, and management layers so the pilot can distinguish a control-plane message from the actual result.
Expand only after the same test produces repeatable outcomes and the operations team can explain discrepancies. Pause rollout when false decisions, resolution or allocation failures, support workload, exception age, or recovery time exceed agreed thresholds. Preserve the test evidence and make the most important failure scenarios part of regression testing.
Operational Checklist
- Allow intended DNS over both UDP and TCP.
- Test truncated responses and TCP retry.
- Tune EDNS size against real network paths.
- Capacity-plan TCP connection state and reuse.
- Monitor encrypted DNS as separate governed paths.
Conclusion
DNS: UDP or TCP? How Modern Resolvers Choose the Transport is successful when the service remains understandable under change and failure. Clear authority, current evidence, proportionate controls, verified outcomes, and deliberate recovery are more valuable than a console that reports only success or failure.
ZDNS DNS supports the central network-infrastructure role described here and can work with the related ZDNS products listed below. The final architecture should reflect the organization's own risk, topology, client behavior, operational capacity, and recovery objectives.