feat(supervisor): stream HTTP session hook uploads on the plain-HTTP forward proxy
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Newbie friendliness
- 32/100
- Issue type
- Feature
- Clarity
- Mostly clear
- Activity status
- Active
- Tech stack
- rust
- Domain
- backend, networking
Research direction
Start by reading ForwardMiddlewarePipeline::apply in crates/openshell-supervisor-network/src/proxy.rs and compare it with how request_body_delivery(config) in l7/relay.rs picks a delivery per route. Read docs/extensibility/supervisor-middleware/operations.mdx for the current 4 MiB wording. Write the forward-proxy denial test first (body_over_limit, upstream sees nothing), then stream middleware output for routes that need no body-wide work. Done means the new tests pass and the limits text matches actual behavior.
Written by the indexing model from the issue text.
Description
User Story
As an operator running an HTTP session hook that streams request bodies (for example a signing or redaction stage), I want plain-HTTP requests sent through the forward proxy to behave like requests on the L7 relay, so that uploads larger than 4 MiB are not denied depending on how the sandbox reached the endpoint.
Problem Statement
HTTP session hooks have no total size cap on STREAM request bodies on the L7 relay paths. The plain-HTTP forward proxy (ForwardMiddlewarePipeline::apply in crates/openshell-supervisor-network/src/proxy.rs) always passes RequestBodyDelivery::WITHHOLD, because it writes the request upstream as one buffer. The preflight therefore advertises limits.max_output_body_bytes of 4 MiB (MAX_MIDDLEWARE_PAYLOAD_BYTES), and a larger upload that matches a session request hook fails with failure_code: body_over_limit.
The limit is documented in Supported Operations (forward-proxy requests buffer output, 4 MiB), but the same hook behaves differently depending on the path a request takes. There is also no test covering the forward-proxy denial.
Raised in review of #4359: https://github.com/NVIDIA/OpenShell/pull/4359#discussion_r4381863330
Impact / Why This Matters
- Inconsistent behavior for one middleware config: an upload over 4 MiB succeeds via the L7 relay and fails via the forward proxy.
- It undercuts the "no total size cap" goal of HTTP session hooks for clients that use
HTTP_PROXYwith plain HTTP.
Proposed Design
- Add a test now: a forward-proxy upload over 4 MiB that matches a session request hook is denied with
failure_code: body_over_limit, and the upstream sees nothing. - Stream middleware output on the forward proxy, like the REST relay, when the route needs no body-wide work (no SigV4 signing, body credential rewrite, or body-aware L7 re-evaluation such as JSON-RPC/MCP/GraphQL). Keep
WITHHOLDwith its route limit for those cases. - Update the Supported Operations limits text once forward-proxy streaming lands.
Acceptance Criteria
- A test covers the 4 MiB forward-proxy denial (
body_over_limit, nothing forwarded). - A plain-HTTP forward-proxy upload over 4 MiB through a STREAM session request stage succeeds on routes that need no body-wide work, with
max_output_body_bytesunset in the preflight. - Routes that still need the whole body keep their documented limits.
- Docs describe the forward-proxy behavior.
Alternatives Considered
- Keep buffering and only document the limit. That leaves the path-dependent behavior in place.
Agent Investigation
proxy.rsForwardMiddlewarePipeline::applycallsapply_request_with_delivery(..., RequestBodyDelivery::WITHHOLD)with the comment "The forward proxy writes the request as one buffer, so middleware output is withheld and inlined."- The L7 REST relay chooses delivery per route via
request_body_delivery(config)inl7/relay.rs. - Docs:
docs/extensibility/supervisor-middleware/operations.mdx(request output buffering paragraph) already lists the forward proxy among buffered paths.
- Dominant language
- Rust
- Stars
- 20.6k
- Forks
- 2k
- Avg merge
- 2d 13h
- Merged PRs (30d)
- 262
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from NVIDIA/OpenShell
-
config preflight exits 0 on allow_unauthenticated_users = truePossibly taken @ericcurtin claimed this today. Openstate:triage-needed
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
Maintainers usually reply within 1 day
-
state:triage-needed
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
Maintainers usually reply within 1 day
-
bug(python): OIDC refresh drops the gateway's oidc_scopesMay be free again @jyotir07 claimed this 7 days ago, and no pull request is open. Openstate:triage-needed
Difficulty 2/5 1-3 hours Newbie friendliness 84/100
NVIDIA/OpenShell#4919 · 3 comments ·
Maintainers usually reply within 1 day
-
bug: workspace access hint suggests a --subject command that fails for hyphen-leading OIDC subjectsMay be free again A pull request for this issue was closed without being merged. Open
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
Maintainers usually reply within 1 day
All issues in NVIDIA/OpenShell
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
DioxusLabs/blitz#1371 ·
Maintainers usually reply within 1 day
-
follow-up
Difficulty 2/5 Half a day Newbie friendliness 62/100
carrick-sh/carrick#1494 ·
Maintainers usually reply within 1 day
-
test(hydro_std): track_membership join and leave are untestedPossibly taken A pull request linked to this issue is open or already merged. Open
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
hydro-project/hydro#3289 ·
Maintainers usually reply within 1 day
-
diagram-bug good first issue help wanted
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
enhapu-afk/omaspec#4 ·