Bound HTTP/2 streams with flow control and connection error handling

Treat this document as a focused evidence review: Bound HTTP/2 streams with flow control and connection error handling. Only the official source and traced locations below supply facts. Confirm applicability before acting.

Source fact:

The official RFC 9113 — HTTP/2 from RFC Editor / Internet Engineering Task Force supports the following bounded statements:

• HTTP/2 maintains receiver-controlled, hop-by-hop flow-control windows for each stream and for the connection; a sender must honor both advertised limits. The research record locates this support at Sections 5.2 (Flow Control) and 5.2.1 (Flow-Control Principles).
• Only DATA frames consume flow-control credit, and every received flow-controlled frame counts against the connection window even when the frame also has an error. The research record locates this support at Sections 5.2.1 (Flow-Control Principles) and 6.9 (WINDOW_UPDATE).
• A stream-local failure sends RST_STREAM without ending other streams; a connection-state failure sends GOAWAY when possible and then closes the TCP connection. The research record locates this support at Sections 5.4.1 (Connection Error Handling) and 5.4.2 (Stream Error Handling).

These statements are the factual basis for this document. Do not extend them into a broader assurance. Review clients, origin services, intermediaries, caches, proxies, gateways, protocol versions, and security-policy enforcement points only where the source and recorded environment align.

What the source does not establish

This RFC evidence supports only the named web or transport decision; it does not prove browser, intermediary, library, or service compatibility. It does not establish a deployment’s current state, authorize a production change, prove compliance, or show that DNS, certificates, identity providers, time, content delivery, network paths, and application ownership are healthy. Documented options are review inputs, not universal mandates.

Applicability questions

• For source statement 1 at Sections 5.2 (Flow Control) and 5.2.1 (Flow-Control Principles), which observable configuration, record, or test can confirm applicability here?
• For source statement 2 at Sections 5.2.1 (Flow-Control Principles) and 6.9 (WINDOW_UPDATE), which observable configuration, record, or test can confirm applicability here?
• For source statement 3 at Sections 5.4.1 (Connection Error Handling) and 5.4.2 (Stream Error Handling), which observable configuration, record, or test can confirm applicability here?
• What inventory proves which parts of clients, origin services, intermediaries, caches, proxies, gateways, protocol versions, and security-policy enforcement points are in and out of scope?
• Which condition in DNS, certificates, identity providers, time, content delivery, network paths, and application ownership must be healthy before evidence is trustworthy?
• What result would disprove the working assumption and return the issue to the owner?

DSE recommendation:

DSE recommends using the cited source as the evidence anchor for this decision. Start with applicability, then compare the observed state with the cited source. Record the source location, examined part of clients, origin services, intermediaries, caches, proxies, gateways, protocol versions, and security-policy enforcement points, observed and expected states, owner, and reason for deviation.

If the review warrants change, use a bounded implementation with prerequisites, test population, monitoring, abort criteria, and a rehearsed reversal. Sequence checks for DNS, certificates, identity providers, time, content delivery, network paths, and application ownership and sanitize protected material before retention.

Verification and evidence

Build a reproducible chain from Sections 5.2 (Flow Control) and 5.2.1 (Flow-Control Principles); Sections 5.2.1 (Flow-Control Principles) and 6.9 (WINDOW_UPDATE); Sections 5.4.1 (Connection Error Handling) and 5.4.2 (Stream Error Handling) to the observed environment. Useful domain evidence includes request and response captures, negotiated protocol details, headers, cache behavior, certificate state, and server or proxy logs; label every item with scope, timestamp, collector, and stable identifier.

Preserve both successful and failed observations, along with approval and rollback evidence. Avoid uncontrolled production experiments. Assign an expiry or event-driven recheck so the conclusion is not treated as permanent.

Official references

• RFC 9113 — HTTP/2 — RFC Editor / Internet Engineering Task Force