HTTPS Proxy HTTP/2 Negotiation and Fallback
HTTP/2 negotiation through an HTTPS proxy involves more than the protocol label shown in a client setting. The client-to-proxy connection and the connection carrying the destination request can have different behavior. When testing Go2Proxy HTTP and HTTPS proxy connections, record the protocol observed at the boundary relevant to the application.
A client requesting HTTP/2 does not prove that every hop used it. A supported fallback can also be a valid outcome if it meets the workload's requirements.

Define the required capability
Decide whether HTTP/2 is a strict application dependency or a performance preference. A strict requirement needs a direct acceptance criterion. A preference needs a comparison that includes correctness and resource use, not just a version string.
Identify the client version, TLS backend and supported proxy mode. Different builds can expose different negotiation capabilities. Preserve these details when comparing a workstation with a deployed worker.
Use an owned or approved destination whose expected protocol behavior is known. A random page may not provide enough evidence to separate destination support from proxy or client behavior.
Keep protocol boundaries separate
A TLS-protected connection to a supported HTTPS proxy is not the same as the destination's TLS connection through a tunnel. Record which connection is being discussed in each diagnostic.
Do not assume an HTTP version reported for the destination describes the proxy control exchange. Equally, an initial gateway response does not establish the final application's negotiated protocol.
Keep the supported gateway scheme unchanged unless the product documentation calls for another mode. Changing schemes simply to obtain a desired protocol label can move the test onto an unsupported path.
Compare requested and observed behavior
Configure the client's documented version preference and inspect the actual result. Save requested behavior separately from negotiated behavior. The distinction matters when a client falls back transparently.
Validate the same content under each permitted test condition. A faster response that returns an error document is not a successful protocol comparison.
Keep connection reuse conditions consistent. A warm connection can avoid work that a fresh connection performs, so mixing them can exaggerate the effect attributed to HTTP/2.
Evaluate fallback deliberately
If HTTP1.1 is an acceptable supported fallback, test it as a separate labeled condition. Record correctness, task completion and relevant timing under comparable payloads.
If the application requires a specific capability, preserve an unsupported result when that capability cannot be demonstrated. Do not disable TLS verification or alter security policy merely to force a protocol preference.
Keep failure categories precise. A certificate problem, tunnel rejection and unexpected application response are not all negotiation failures. Diagnose the actual boundary before changing client preferences.
A misleading version comparison
Imagine an application update appears to improve proxy performance. The previous run used fresh connections, while the new run reused an existing connection and fetched a smaller response. The HTTP version changed too, but the test cannot isolate its contribution.
Repeat a bounded comparison with the same destination, response, connection state and worker environment. Preserve failures and fallback outcomes. Only then describe what the measured results support.
The application may choose the simpler supported configuration if it meets the task's requirements. The highest protocol number is not automatically the best operational choice for every client stack. The HTTP CONNECT tunneling guide explains the distinction between reaching the proxy and establishing the destination connection.
HTTP/2 proxy questions
Does requesting HTTP/2 guarantee it was negotiated?
No. Record the actual result for the relevant connection. Preferences, support and fallback behavior can differ.
Does destination HTTP/2 prove the proxy connection also used it?
No. These are separate protocol boundaries. Label each observation accordingly.
Is fallback always a failure?
Only when it violates the application's stated requirement. Otherwise evaluate it under a documented supported configuration and retain the observed result.
Measure the property the task needs
Keep protocol preference, negotiated behavior and usable-content results together. This makes compatibility decisions defensible without turning a version label into an unsupported performance or reliability promise.



