HTTP 2.1 Data Flow — Sketch
This is a sketch of the SLeeLa HTTP 2.1 application protocol, a sibling to
../http-3.0/. HTTP 2.1 shares HTTP 3.0's basic design goals
but is an earlier generation: it carries the clean core only and omits the
3.0-era per-packet integrity substrate.
Same design goals as HTTP 3.0
| Goal | HTTP 2.1 | HTTP 3.0 |
|---|---|---|
| Compact application envelope | ✅ core fields | ✅ core + integrity fields |
| Fast service/operation naming | ✅ | ✅ |
| Response model (transport ≠ application) | ✅ | ✅ |
| Retry classes | ✅ (READ/IDEMPOTENT/MUTATING) | ✅ (+ STREAM) |
| Traceable processing pipeline | ✅ | ✅ |
What HTTP 2.1 deliberately omits (3.0-era additions)
- Per-packet keyed-MAC DIGEST (SipHash-2-4)
- INTACTX host-integrity identity / tamper reset
- Replay NONCE
- Goods & services BASKET
So a 2.1 packet is just: VERSION | FLAGS | SERVICE-ID | OP-ID | REQUEST-ID | PAYLOAD.
The data, and where it lives
| Concept | Type / symbol | File |
|---|---|---|
| Compact envelope | h21_envelope_t | h21_envelope.h |
| Textual/binary pack/unpack | h21_envelope_pack_text / _binary (+ unpack) | h21_envelope.c |
| Fast-naming dictionary | h21_naming_t | h21_naming.h |
| Response model | h21_response_t / h21_status_t | h21_envelope.h |
| Retry classes | h21_retry_class_t | h21_pipeline.h |
| Processing pipeline | h21_pipeline_dispatch / _handle_wire | h21_pipeline.c |
| Python parity | h21_flow.py | h21_flow.py |
Wire forms
envelope (textual): H21 <ver> <flags> <service_id> <op_id> <request_id> <len>:<payload>\n
response (textual): H21R <status> <request_id> <len>:<result>\n
envelope (binary) : [ver:1][flags:1][service_id:4][op_id:4][request_id:8][payload_len:4][payload:N]
Hot path (no integrity gate)
receive -> minimal parse (auto-detect textual "H21 " vs binary)
-> service-id lookup -> operation-id lookup
-> dispatch -> business logic -> pack response
Unlike HTTP 3.0, there is no integrity gate between parse and dispatch: 2.1 has no per-packet MAC, host-identity, replay, or basket checks.
Implementation status
- C envelope/response pack/unpack is implemented with strict version, flag, and exact-length checks; Python text envelope/response parsing and naming/ pipeline dispatch are implemented and covered by regression tests.
- Native server parsing/serialization is hardened but still synchronous, has no built-in TLS/authentication, and must be deployed behind a protected edge.
- HTTP/2 stream multiplexing is a design target only. The native server uses HTTP/1.x transport.
Build
cd http-2.0
make proto # compiles the protocol-core objects (compile check)
make syntax # syntax-only check of all C sources
Port and Stream Multiplexing
The logical-port model can sit above an HTTP/2 stream layer in a future transport implementation. The current server does not implement that layer.
one transport connection
|
+-- HTTP/2 stream 1 -- logical port -- service/op -- payload
+-- HTTP/2 stream 2 -- logical port -- service/op -- payload
+-- HTTP/2 stream N -- logical port -- service/op -- payload
The important separation is:
- Transport endpoint: the native TCP/TLS endpoint.
- HTTP/2 stream: the independently multiplexed request/response stream.
- SLeeLa logical port: the application-level service/channel/route identifier.
- SERVICE-ID / OP-ID: the compact dispatch identity carried by the SLeeLa envelope.
A logical port therefore does not require one operating-system socket per service. Multiple logical ports can share the same HTTP/2 connection and its stream pool.
HTTP/2 stream identifiers and SLeeLa logical ports are different namespaces and must not be conflated. The stream provides transport multiplexing; the logical port provides application routing.
Download Mode for Files > 50 MB
The design proposes HTTP/2 streams for multiplexing, while SLeeLa DOWNLOAD metadata describes resumable transfers for files larger than 50 MB. The current server does not implement an HTTP/2 stream engine or filesystem-backed transfer manager. A stream carries FILE-ID and transfer index; resume state survives stream completion.
SESSION-ID | DATETIME | FILE-ID | FILE-NAME | INDEX | OFFSET | TOTAL-SIZE
The client resumes from FILE-ID + INDEX/OFFSET and does not depend on the original HTTP/2 stream remaining available.