Skip to content

feat(middleware): introduce improved HTTP request/response hooks for the middleware - #4359

Merged
pimlock merged 56 commits into
mainfrom
feat/2431-http-middleware-protocol-2/pimlock
Nov 12, 2026
Merged

pimlock merged 56 commits into
mainfrom
feat/2431-http-middleware-protocol-2/pimlock

Conversation

@pimlock

@pimlock pimlock commented Oct 9, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

Introduce HTTP hooks: streaming HTTP request and response hooks that replace the legacy HTTP hooks.

Why

The legacy hooks were built for small, buffered checks. The use cases we want next, such as signing a Git push or a large AWS request, need the whole body regardless of size and need to change the body and its headers together.

  • Large bodies. EvaluateHttpRequest is unary: the whole body in one message, at most 4 MiB, under one RPC deadline. A larger request fails with request_body_over_capacity. HTTP hooks can stream a body with no total size or time cap.
  • Headers that depend on the body. Signing reads the whole body, then rewrites it along with headers such as Content-Length or a SigV4 payload hash. An HTTP hook stage can send header mutations after it has read the complete body.
  • Free-form response processing. HttpResponsePreReturn is lockstep: each body unit gets exactly one result, and a service can't hold bytes for a later unit. In DUPLEX mode, input and output chunks are independent, so a service can redact across chunk boundaries or re-frame a stream.
  • One model for both directions. Requests and responses share one lifecycle (preflight, body, finish or deny) with the same body modes and limits.

Why not extend the legacy hooks. A unary RPC can't become a stream without breaking every deployed service. Lockstep is a semantic guarantee, not a field, so relaxing it on the same RPC would change behavior that buf breaking can't detect. Older peers would also ignore new fields and call an extended service with legacy semantics. A separate service gated by a required capability makes older peers refuse it, and the legacy hooks keep working unchanged until they are removed.

What this PR adds

  • New streaming API. HttpHooks.EvaluateRequest and HttpHooks.EvaluateResponse, a new service served on the same endpoint as SupervisorMiddleware, with one stream per HTTP message. A preflight decides whether to continue, inspect the body, reject a request, or block response delivery with block_delivery. A stage inspects the body in one of two modes: WHOLE_BODY, one complete body and one result, or DUPLEX, independent input and output streams.
  • Opt in with a capability, no new HTTP operations. A service keeps its HTTP_REQUEST and HTTP_RESPONSE bindings and lists openshell.supervisor-middleware.http-hooks in its extension required_capabilities. Gateways and supervisors have rejected unmet required capabilities at Describe since v0.1.0, so older peers refuse the service instead of calling it through the legacy RPCs. No RPC, message, or enum value carries a version.
  • No fail_open. HTTP hooks are always fail-closed, which simplifies the UX, the API, and the code.
  • Legacy hooks are deprecated. EvaluateHttpRequest, HttpResponsePreReturn, and the types only they use are marked deprecated in the protobuf contract and are up for removal in the next version bump (0.2.0). Migration notes are in the new docs page. Both hook versions work side by side until then.

Tip

The interactive walkthrough covers the motivation, the new API, version selection, and migration in more detail.

Related Issue

Part of #2431 and #3307.

Closes #4269.

Changes

  • Gateways and supervisors advertise the openshell.supervisor-middleware.http-hooks capability. They reject a manifest that supports it without requiring it, because older peers would call that service through the legacy hooks.
  • HTTP hook services can't use on_error: fail_open, and one chain can't mix hook versions.
  • Middleware selectors can't contain a port. A host:port selector never matched anything, so the entry was silently skipped. It now fails validation.
  • Every hook with a RequestContext, including legacy and WebSocket hooks, now receives originating_process.
  • The content guard example uses HTTP hooks.
  • Docs and RFC 0009 cover HTTP hooks, the body modes, and migrating from the legacy hooks.

Release Notes

HTTP middleware can now use the new HTTP hooks, which stream request and response bodies of any size. A stage inspects the whole body (WHOLE_BODY, up to 4 MiB) or streams it (DUPLEX, with no total size or time limit), and can change headers after reading the body. A service opts in by requiring the openshell.supervisor-middleware.http-hooks capability. HTTP hook services are always fail-closed.

The legacy HTTP hooks (EvaluateHttpRequest and HttpResponsePreReturn) are deprecated and will be removed in 0.2.0.

Upgrade note: middleware selectors can no longer contain a port. Such a selector never matched, so remove the port.

Testing

  • mise run pre-commit passes
  • Unit tests added/updated
  • E2E tests added/updated: new mise run e2e:middleware-http-hooks suite
  • buf breaking against v0.1.2 reports no breaking change

Validation. Two prototypes on an earlier head of this PR exercise the API: Git commit signing streams a push larger than 4 MiB through an external signer that keeps the key outside the sandbox, and SigV4 signing signs the final request after HTTP hook stages transform its body.

Checklist

  • Follows Conventional Commits
  • Commits are signed off (DCO)

@github-actions

github-actions Bot commented Oct 9, 2026

Copy link
Copy Markdown

@pimlock
pimlock marked this pull request as draft October 9, 2026 01:31
@copy-pr-bot

copy-pr-bot Bot commented Oct 9, 2026

Copy link
Copy Markdown

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

@pimlock pimlock changed the title feat(middleware): add HTTP middleware protocol 2 feat(middleware): introduce improved HTTP request/response hook for the middleware Oct 9, 2026
@pimlock pimlock changed the title feat(middleware): introduce improved HTTP request/response hook for the middleware feat(middleware): add v2 HTTP request and response hooks Oct 9, 2026
@pimlock pimlock changed the title feat(middleware): add v2 HTTP request and response hooks feat(middleware): introduce improved HTTP request/response hooks for the middleware Oct 10, 2026
@pimlock

pimlock commented Oct 10, 2026

Copy link
Copy Markdown
Collaborator Author

/ok to test 7f09efe

@pimlock
pimlock marked this pull request as ready for review October 10, 2026 01:45
pimlock

This comment was marked as outdated.

@pimlock pimlock added the gator:in-review Gator is reviewing or awaiting PR review feedback label Oct 10, 2026
@pimlock

This comment was marked as outdated.

@pimlock pimlock added the test:e2e Requires end-to-end coverage label Oct 10, 2026
@github-actions

This comment was marked as outdated.

pimlock

This comment was marked as outdated.

@pimlock pimlock added gator:blocked Gator is blocked by process or repository gates and removed gator:in-review Gator is reviewing or awaiting PR review feedback labels Oct 10, 2026
Add EvaluateHttp, a streaming HTTP middleware protocol with BUFFERED and
STREAM body modes, next to the released HTTP protocol 1, which keeps its
engines unchanged and is deprecated for removal in 0.2.0.

- One EvaluateHttp stream per HTTP message and stage, or per uninspectable
  connection. A preflight selects continue, inspect, or reject; late header
  mutations apply before the head commits.
- A service implements one HTTP protocol, chosen per binding with
  http_protocol_version, and a protocol 2 service requires the
  openshell.supervisor-middleware.http-v2 capability. The gateway delivers
  each service's protocol to supervisors.
- Protocol 2 is always fail-closed. The gateway rejects fail_open on
  protocol 2 services and policies whose overlapping selectors mix
  protocols; a mixed chain fails closed at runtime.
- Protocol 2 middleware decides about uninspectable traffic (tls: skip,
  h2c, unsupported tunnels, raw TCP, SQL passthrough) at a preflight, and is
  exempt from the static tls: skip conflict rule.
- Request bodies stream to the upstream unless a later step needs the whole
  body; responses stream with chunked framing or buffer before commit.
- Migrate the content guard example to protocol 2, add a Docker e2e suite,
  and document the protocol.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
Replace the http-v2 capability and per-binding protocol fields with new
binding operations, and stop delivering the protocol to supervisors.

- HTTP_REQUEST_V2 and HTTP_RESPONSE_V2 select HTTP protocol 2. Released
  gateways and supervisors already reject unknown operation/phase pairs at
  Describe, so the capability, extension_metadata_with_requirements, and
  MiddlewareBinding.http_protocol_version are removed.
- Drop supported_http_body_modes. The preflight offers every mode the
  message is eligible for, and the stage picks one or continues or
  rejects. A protocol 2 binding with a zero payload limit is offered no
  body modes.
- Move the tls: skip conflict rule out of the shared policy validator into
  the gateway's safety validation, which knows each service's protocol.
  This removes the delivered SupervisorMiddlewareService protocol field,
  the process-wide protocol 2 set in openshell-policy, and the supervisor's
  undescribed-service tracking. An undescribed entry follows on_error,
  which the gateway requires to be fail_closed for protocol 2 services.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
Replace EvaluateHttp with EvaluateHttpRequestV2 and EvaluateHttpResponseV2,
one RPC per binding operation, as for HTTP protocol 1 and WebSocket. Both
keep the shared HttpEvent and HttpResult messages, so the stage pipeline is
unchanged; OpenShell picks the RPC from the preflight subject.

Traffic OpenShell cannot inspect stays on the request RPC. Only services
that bind HTTP_REQUEST_V2 decide about it, so only they are exempt from the
gateway's tls: skip conflict rule; a response-only protocol 2 service keeps
the rule and its fail-closed entry denies such traffic at runtime.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
Rename "HTTP protocol 1/2" to v1 and v2 HTTP hooks across code, proto
comments, docs, skills, and the example. The docs page moves to
extensibility/supervisor-middleware/http-hooks-v2, the protocol2 modules
become http_v2, HttpProtocol becomes HttpHookVersion, and the mixed-chain
reason becomes middleware_hook_versions_mixed.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
Rename the streaming HTTP hook RPCs to EvaluateHttpRequestSession and
EvaluateHttpResponseSession, and drop the unreleased HTTP_REQUEST_V2 and
HTTP_RESPONSE_V2 operations. A service keeps its HTTP_REQUEST and
HTTP_RESPONSE bindings and selects HTTP session hooks for all of them by
listing openshell.supervisor-middleware.http-session in its extension
required_capabilities. Released gateways and supervisors already reject
unmet required capabilities at Describe, so they never call such a
service through the v1 RPCs. Registration rejects a manifest that
supports the capability without requiring it.

Mark the v1 HTTP hook RPCs, the HttpResponsePreReturn service, and the
types only they use deprecated in the protobuf contract. Rename the
docs page, module, and e2e suite to HTTP session hooks, and update RFC
0009 so capability-gated contracts and removing deprecated elements in a
minor release match the release stability policy.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>

@matthewgrossman matthewgrossman left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Follow-up review of 1c982fa, from the same multi-agent review panel as the round on 97f049c. The panel inferred your responses from your replies and commits, and then had the original reviewers re-check each claimed fix.

All six fixes confirmed:

  • IPv6-glob/host:port selectors are rejected, with tests and an upgrade note (702284c).
  • The JSON-RPC and GraphQL RequestTimeout paths return 408 (7714bdb).
  • request_timeout is documented, and the skill pairs each OCSF reason with its failure_code (0732227).
  • originating_process is filled consistently for every hook (25dc2a5).

Still open as a follow-up: forward-proxy streaming for hooked uploads, tracked in #5086.

No new findings in the PR's own changes. Most of this range is merged-in main, so the reviewers focused on the PR's own commits (the http_session → http_hooks rename and the body-mode renames). They didn't build or run tests.

Contract-level comments are on #5087, as the PR description asks; any agreed changes would land here.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
State how zero-limit HTTP hooks bindings interact with the operator payload limit, align the HTTP/1.0 DUPLEX rule with the proto, document the 403 for a DUPLEX request denied after upstream output starts, and make failure_code wording consistent across the docs.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
…X request denial

Pin that a zero-limit HTTP hooks binding is excluded from the operator payload limit comparison, and that a DUPLEX request denied after its output started upstream still gets the canonical 403 and an unterminated upstream body.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
…ware-protocol-2/pimlock

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>

# Conflicts:
#	crates/openshell-supervisor-network/src/l7/middleware.rs
#	crates/openshell-supervisor-network/src/proxy.rs
@github-actions

github-actions Bot commented Nov 6, 2026

Copy link
Copy Markdown

Compatibility review (advisory)

New or updated breaking changes; see the full report for all findings.

1. policy — Risk explicitly accepted

network_middlewares.*.endpoints.include/exclude previously accepted IPv6 globs such as 2001:4860:*; colon-containing patterns now require complete IPv6 literals.

Review details
  • Execution: complete
  • Candidate: PR #4359 (11c5e586700807bac6fa25b5457f78123df6b1d3)
  • Baseline: d7da0c7da2d9c6058f5fd2b69b7ea14536e777c3 (d7da0c7da2d9c6058f5fd2b69b7ea14536e777c3)
  • Reviewed: rust-sdk, python-sdk, go-sdk, typescript-sdk, api-behavior, cli, configuration, policy, helm, state

Advisory review; publication gates are unchanged.

Branch check run

Rename HttpHooks.EvaluateRequest and EvaluateResponse to
EvaluateHttpRequestHook and EvaluateHttpResponseHook, and document in
supervisor_middleware.proto that RPC names must be unique across the
package's services, with the naming pattern for future hook services.
Python and Go servers name handler methods after the RPC alone, so one
object implementing several services needs distinct names.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
@pimlock

pimlock commented Nov 6, 2026

Copy link
Copy Markdown
Collaborator Author

Pushed bc90d94: renamed the HTTP hook RPCs so their names are unique across the middleware services.

Before After
HttpHooks.EvaluateRequest HttpHooks.EvaluateHttpRequestHook
HttpHooks.EvaluateResponse HttpHooks.EvaluateHttpResponseHook

Why: a middleware implements SupervisorMiddleware and the hook services it wants on one endpoint, often on one object. Python (grpcio) and Go generate handler methods named after the RPC alone, so two services with the same RPC name can't share an implementation. For example, a Python class that extends both SupervisorMiddlewareServicer and HttpResponsePreReturnServicer already exposes a bare Evaluate; a future service with EvaluateRequest would collide with HttpHooks the same way. The names are unreleased, so renaming now costs nothing.

Rule, documented at the top of proto/supervisor_middleware.proto: every RPC name is unique across all services in openshell.middleware.v1, including deprecated ones. A new protocol gets a <Protocol>Hooks service with RPCs named Evaluate<Protocol><Unit>Hook and messages <Protocol><Unit>HookEvent / <Protocol><Unit>HookResult.

Wire behavior is otherwise unchanged. The rename also updates the tonic trait methods in openshell-supervisor-middleware and the content-guard example, plus the docs, RFC 0009, and the debug skill reference.

pimlock added a commit that referenced this pull request Nov 6, 2026
Mirror bc90d94 from #4359: rename HttpHooks.EvaluateRequest and
EvaluateResponse to EvaluateHttpRequestHook and EvaluateHttpResponseHook,
and document in supervisor_middleware.proto that RPC names must be unique
across the package's services.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>

@matthewgrossman matthewgrossman left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Follow-up review of bc90d94 from the same multi-agent review panel, covering only your changes since 1c982fa (the RPC rename, the zero-limit and late-DUPLEX tests and docs, and the merge-conflict resolutions). The rename is applied consistently and the conflict resolutions keep the hook logic the same. Two small suggestions are inline. The forward-proxy follow-up stays tracked in #5086.

Set ActionId::Denied on the HTTP hooks http_response_blocked finding and
on the after-commit delivery failure when a stage stopped delivery, so
both hook versions emit the same OCSF for the same decision. Assert that
a DUPLEX request denied after upstream start leaves the chunked upstream
body unterminated.

Signed-off-by: Piotr Mlocek <pmlocek@nvidia.com>
@matthewgrossman

Copy link
Copy Markdown
Member

Review notes. I read the proto, capability negotiation, selector validation, header-authority rules and core trait changes. I only skimmed the streaming pipeline, relay and body-handling code and did not run the tests.

The required-capability gating, the always-fail-closed model and the two slot budgets look sound.

Concerns

  • host:port selectors now fail validation. Before this PR, a host:port selector was silently ignored, so the middleware did not run for that host. With this PR, the policy is rejected at load, so an upgrade can block a policy update or sandbox creation until the port is removed. The release note covers it. Please decide deliberately whether this needs a deprecation window, or at least an error that names the offending entry. (crates/openshell-policy/src/middleware.rs)
  • originating_process now goes to every hook with a RequestContext, including legacy and WebSocket hooks. Before this PR, remote middleware got no workload process identity. With this PR, third-party services receive binary and pid details. Please confirm that sharing this with remote middleware is intended and that the docs say so. (proto/supervisor_middleware.proto, RequestContext)
  • DUPLEX has no total size or time cap. A stuck or slow service can hold a sandbox stream and one of the 256 streaming slots until the idle timeout. Please confirm that idle-timeout enforcement and slot release on stage failure are covered by tests. (http_hooks/pipeline.rs)
  • Review surface. The PR is about 16.5k lines, and the pipeline (about 2.5k lines) and relay changes sit on the enforcement path. A second reviewer should walk through body re-framing (Content-Length, chunked, trailers) after stage mutations.

Context: #4269 is labelled state:triage-needed, but the author has admin access, so the acceptance gate does not apply.

@pimlock
pimlock added this pull request to the merge queue Nov 12, 2026
Merged via the queue into main with commit f37687a Nov 12, 2026
151 of 152 checks passed
pimlock pushed a commit that referenced this pull request Nov 12, 2026
Retain typed local deadline and remote failure causes through HTTP hook diagnostics and request handling. Track local upstream write progress through the existing live writer, including partial failures. Preserve response selection, diagnostic text and request retries.

Refs #4144

Build on PR #4359. Caller and event exposure remains with the companion tracing work.

Signed-off-by: Shiju <shiju@nvidia.com>
@pimlock
pimlock deleted the feat/2431-http-middleware-protocol-2/pimlock branch November 12, 2026 21:38
@matthewgrossman

Copy link
Copy Markdown
Member

Review

Read the full proto diff and sampled the core, policy, server, and pipeline code. I did not read all ~16.5k added lines. gh pr diff fails on this PR because it's over GitHub's 20k-line limit, so I reviewed a local fetch.

Overview

Adds the HttpHooks streaming service (EvaluateHttpRequestHook / EvaluateHttpResponseHook) for supervisor middleware, with WHOLE_BODY and DUPLEX body modes. The legacy unary and lockstep RPCs are deprecated for removal in 0.2.0.

Key design decisions

  • Capability gating instead of versioned RPCs. openshell.supervisor-middleware.http-hooks in required_capabilities makes older peers refuse the service at Describe (extension_protocol.rs, with a negotiation test).
  • Separate service instead of extending the legacy RPCs. Lockstep to free-form is a semantic change that buf breaking can't detect.
  • Preflight reports why a body mode is unavailable (HttpBodyUnavailableReason), so a stage can deny with a specific reason instead of silently skipping.
  • Always fail-closed. fail_open is rejected at policy validation (openshell-server/src/middleware.rs).
  • Two process-wide admission budgets. 64 buffering slots and 256 streaming slots, so long-lived SSE streams can't starve short requests (openshell-core/src/middleware.rs).

Potential concerns

  • Existing policies with a port in a middleware selector are now rejected. Before this PR, a host:port selector applied but silently matched nothing. With this PR, the policy fails validation with invalid_argument, so a policy update or re-apply that used to succeed now errors and needs editing. It's noted as an upgrade note, but it's an independent behavior change inside a very large PR. Worth confirming no stored or default policies hit it, or splitting it out. crates/openshell-policy/src/middleware.rs (selector_pattern).
  • HTTP hooks can't share an endpoint with legacy built-ins. Before this PR, a service with HTTP_REQUEST bindings could be chained with the built-in regex middleware on overlapping endpoints. With this PR, overlapping selectors must use one hook version per operation (overlapping_selectors_must_use_one_http_protocol_per_operation uses BUILTIN_REGEX as the legacy side). Operators adopting an HTTP-hooks service have to drop or move overlapping built-in redaction. Please confirm that's intended until the built-ins migrate, and that the docs say so. crates/openshell-server/src/middleware.rs (validate_http_hook_rules).
  • Legacy middleware services now receive originating_process. Before this PR, existing legacy services saw no process identity. With this PR, they receive the workload process path and pid without opting in. Probably fine for operator-run services, but it's a data-exposure change that isn't tied to the capability gate. proto/supervisor_middleware.proto (RequestContext).
  • Memory and slot starvation. Each of the 64 buffering slots can hold an input body plus a replacement, each up to the stage cap. At 4 MiB that's roughly 512 MiB worst case per process, and a slow middleware can hold a slot for up to the 2-minute HTTP_WHOLE_BODY_TIMEOUT. I didn't trace how the slots are shared when one supervisor serves many sandboxes. pipeline.rs; slot constants in openshell-core/src/middleware.rs.
  • Reviewability. pipeline.rs alone is 2.5k lines. The selector-port fix and the originating_process change are independent and could land separately.

No crypto or deny.toml changes. Skills and docs are updated (debug-openshell-cluster, generate-sandbox-policy, RFC 0009).

cv added a commit to NVIDIA/NemoClaw that referenced this pull request Nov 18, 2026
Moves NemoClaw's OpenShell pins from v0.1.8 to v0.1.9: the native CLI,
the gateway, supervisor and sandbox images, the gateway chart, the
OpenShell crates, and the vendored SDK. A published managed Kubernetes
gateway now disables the chart's metrics with `metrics.enabled: false`,
which v0.1.9 requires, and the live middleware fixture moves to
OpenShell's HTTP hooks. Deployments built from this revision run
OpenShell v0.1.9, whose chart applies `caConfigMap`'s CA only to the
OIDC issuer.

## Problem

NemoClaw pins OpenShell v0.1.8 (`f3d9765`). v0.1.9 (`ce58cf8`, 51
commits later) is the latest release, and it changes three things
NemoClaw depends on:

- The chart refuses `service.metricsPort: 0` (NVIDIA/OpenShell#5478),
which NemoClaw sets on a gateway published through `expose.service`; the
chart's exposed contract scenario fails to render with
`service.metricsPort must be an integer between 1 and 65535; to disable
metrics, set metrics.enabled=false`.
- `server.oidc.caConfigMapName` no longer sets a process-wide
`SSL_CERT_FILE` on the gateway; the chart renders
`[openshell.gateway.oidc] ca_bundle`, so the CA applies only to OIDC
discovery and JWKS requests (NVIDIA/OpenShell#3673).
- Middleware selectors with a port fail validation, and the legacy
`EvaluateHttpRequest` and `HttpResponsePreReturn` hooks are deprecated,
to be removed in 0.2.0 (NVIDIA/OpenShell#4359).

Part of #13829: it settles question 2 of [cv's
answers](#13829 (comment)),
keeping `caConfigMap` as a typed field under
`target.kubernetes.authentication`. With v0.1.8 the chart turned that
ConfigMap into the gateway's process-wide `SSL_CERT_FILE`, so a field
named for the issuer's CA also set the trust of the gateway's other
HTTPS clients; with v0.1.9 it affects only the issuer, which is what a
typed authentication field says. The rest of #13829 (pull Secrets,
Docker and Podman ports, the `openshift` target) doesn't depend on this
upgrade, and the OpenShift UID nulls stay until NVIDIA/OpenShell#5483.

## Change

- **Pins:** the Cargo revision of `openshell-core`, `openshell-sdk` and
`openshell-policy`, and of the vendored SDK's `openshell-core` and
`openshell-crypto`; in `versions.json`, the version, revision, Linux CLI
archives, gateway, supervisor and sandbox images, and gateway chart; the
managed-gateway reference, a contract test and the gateway contract
module, which carry the gateway digest. The CLI checksums come from the
release's `openshell-checksums-sha256.txt`; the image and chart digests
come from GHCR's `0.1.9` tags, and GHCR's `0.1.8` tags match the
previous pins. `Cargo.lock` changes only the five OpenShell crates'
revision.
- **Vendored SDK:** upstream changed `src/client.rs` and `src/raw.rs` (a
raw `ServiceDiscovery` client), `src/transport.rs` and
`tests/client_mock.rs`, copied unchanged. The SDK manifest still enables
core telemetry, so the manifest patch stays; `NOTICE.md` records the
refresh. The SDK integration limits in
`internal-docs/design/architecture.md` still hold at v0.1.9 and cite
that revision.
- **Published gateway metrics:** `expose.service` sets the chart's
`metrics.enabled: false` instead of `service.metricsPort: 0`, so the
NodePort Service still publishes only the gRPC port and the gateway
serves no plaintext metrics. This deletes the `metricsPort` workaround
and its comment, which #14313 left for this upgrade. The chart
contract's exposed scenario failed with the v0.1.9 chart before the
change and passes after it.
- **Middleware:** the `docker-dummy` live case's middleware fixture
implements OpenShell's `HttpHooks` service and requires the
`openshell.supervisor-middleware.http-hooks` capability, deciding at
preflight without the body; it answers the legacy `EvaluateHttpRequest`
with `UNIMPLEMENTED`. The case's registration and assertions are
unchanged. A middleware selector with a port is a new case in the SDK's
OpenShell-validation test, which failed against v0.1.8 and passes
against v0.1.9; the selector's schema description and the sandbox
network guide say hosts take no port.
- **Docs:** OpenShell links move to the v0.1.9 revision; each path and
anchor exists there, and the three proxy line ranges in the sandbox
network guide move 71 lines down to the same text. The Kubernetes guide
says the gateway trusts `caConfigMap`'s CA only for the issuer.
- **Release notes** list the new pin and the v0.1.9 changes users see,
with two breaking changes: a middleware selector with a port fails plan,
and a DNS-glob endpoint no longer admits a private IP literal on its own
(NVIDIA/OpenShell#5522).

## Upgrade notes checked

- **Supervisor shutdown** (NVIDIA/OpenShell#5342): a dedicated
supervisor now stops within 8 s and Docker waits 10 s, so stops no
longer end in `SIGKILL`, and a connection that drops during shutdown is
no longer recovered. That can shorten the 13 s between stop and start in
#13953's failure, but the failure itself, a restarted supervisor finding
its sandbox session terminal, isn't addressed, NVIDIA/OpenShell#4910 is
still open, and #13953 didn't reproduce in 260 runs; #13953 stays open.
- **OpenShift UIDs:** NVIDIA/OpenShell#5483 is still open and not in
v0.1.9, so `openshift` keeps nulling the chart's UIDs (#13828).
- **OCSF** (NVIDIA/OpenShell#5606): NemoClaw reads no OCSF records; the
release notes tell log consumers how to select denials now.
- **Go SDK move, mutation admission limits, startup windows, service
discovery:** NemoClaw uses the Rust SDK and OpenShell's defaults;
authored gateway configuration can set the new keys.

Signed-off-by: Carlos Villela <cvillela@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

test:e2e Requires end-to-end coverage

Projects

None yet

Development

Successfully merging this pull request may close these issues.

feat(supervisor): populate RequestContext.originating_process for supervisor middleware

2 participants