fix(protocol): single protocol path — EventMeshFrame across HTTP ingress, filter chain and TCP egress (#5299)

[qqeasonchen]

Issue #5299 Sub-PR A: route UniHttpServer ingress through FrameAdaptor SPI

Summary

The HTTP ingress path now converts structured CloudEvents JSON bytes to
EventMeshFrame at the boundary via the FrameAdaptor SPI
(CloudEventsFrameAdaptor). UniIngressService no longer needs to import
io.cloudevents.CloudEvent on the HTTP code path. The runtime primary
path is now Frame-in / Frame-out; the CloudEvent-typed ingress methods
(publish / publishBatch / request / reply / publishLite /
pollLite) are preserved unchanged for binary compatibility with the TCP
bridge and out-of-tree callers.

New public methods on UniIngressService

New method	Purpose
publishBatchFrames(topic, List<EventMeshFrame>)	Primary batch ingress; HTTP path uses one FrameAdaptor.toFrame per array element, then this method
publishLiteFrame(parent, lite, EventMeshFrame)	Primary lite ingress; HTTP /events/lite/publish
pollLiteFrames(parent, lite, max, timeoutMs) → List<EventMeshFrame>	Primary lite poll; HTTP /events/lite/poll response body serialized back to CloudEvents JSON by the egress adapter
requestFrame(topic, EventMeshFrame, timeout) → EventMeshFrame	Primary request-reply (§17)
replyFrame(correlationId, EventMeshFrame)	Primary reply path

The CloudEvent-typed overloads are kept for binary compatibility with the
TCP bridge and out-of-tree callers. They will be deprecated in sub-PR B
once the TCP path also migrates to the MeshMessageFrameAdaptor.

UniHttpServer changes (8 ingress call sites)

Endpoint	Before	After
POST /events/publish	EventFormatProvider.deserialize(body)	FrameAdaptors.get("cloudevents").toFrame(new ByteTransport(body))
POST /events/publishBatch	same per element, then publishBatch(List<CE>)	same per element, then publishBatchFrames(List<Frame>)
POST /events/lite/publish	EventFormatProvider.deserialize(body)	FrameAdaptor.toFrame(...) + publishLiteFrame(...)
GET /events/lite/poll	EventFormatProvider.serialize(ce) per event	FrameAdaptor.toCloudEventsJson(frame) per event
POST /request (blocking req-reply)	EventFormatProvider.deserialize(body) + request(CE)	FrameAdaptor.toFrame(...) + requestFrame(...); response body via FrameAdaptor.toCloudEventsJson(reply)
POST /events/reply	EventFormatProvider.deserialize(body) + reply(CE)	FrameAdaptor.toFrame(...) + replyFrame(...)
GET /events/poll (egress)	FrameAdaptors.toCloudEventsJson(be.getEvent())	unchanged (already FrameAdaptor)

UniHttpServer no longer imports io.cloudevents.*. The 5 imports
FrameAdaptor / ByteTransport / EventMeshFrame are added.

Security stub — temporary bridge (sub-PR B scope)

The filterChain.check(stubEvent, ctx) ACL check inside UniHttpServer.publish
still consumes a CloudEvent. A TODO(#5299 sub-PR B) comment marks the spot.
The next sub-PR will:

1. Add FilterChain.check(EventMeshFrame, FilterContext) overload
2. Make UniHttpServer pass the Frame directly to the filter chain
3. Update CloudEventFilter / AclFilter / SignatureVerifierFilter /
   TokenAuthFilter to look up extensions from Frame attributes

Until then, the HTTP path does a one-shot frame.toCloudEvent() for the
ACL call only. This is a runtime hot-path minor cost (one allocation per
request) that the next sub-PR removes.

Acceptance against #5299

• Runtime code paths (UniHttpServer) reference only EventMeshFrame,
  not a protocol-specific envelope type
• Each non-primary protocol has an explicit ingress/egress adapter
  (CE is already primary; MeshMessage and A2A follow in sub-PR C
  and existing sub-PR D)
• Ingress / egress goes through FrameAdaptor SPI — no direct
  EventFormatProvider calls in UniHttpServer
• core SDK package has no Netty / gRPC / OpenMessaging — out of
  scope (issue marks this as a follow-up)
• Documentation marks each protocol path as primary / beta / legacy —
  done in sub-PR D

Verification

• ./gradlew.bat :eventmesh-runtime:compileJava --offline → BUILD SUCCESSFUL,
  0 errors, no new warnings (45 deprecation warnings all pre-existing)
• ./gradlew.bat :eventmesh-runtime:checkstyleMain :eventmesh-runtime:checkstyleTest --offline
  → BUILD SUCCESSFUL, 0 violations
• ./gradlew.bat :eventmesh-runtime:test --offline --tests "*UniIngressServiceTest*" --tests "*UniHttpServer*Test*" --tests "*SecurityWiringTest*" --tests "*SseStreamTest*" → BUILD SUCCESSFUL, 17 tests, 0 failures, 0 ignored

Files changed

• eventmesh-runtime/src/main/java/org/apache/eventmesh/runtime/http/UniHttpServer.java
  (61 +/29 -)
• eventmesh-runtime/src/main/java/org/apache/eventmesh/runtime/ingress/UniIngressService.java
  (67 +/2 -)

Follow-up

• Sub-PR B: Filter chain migration (CloudEventFilter / AclFilter /
  SignatureVerifierFilter / TokenAuthFilter) to accept EventMeshFrame
• Sub-PR C: TCP path migration (UniTcpServer + MeshMessagePackageRouter)
  to use MeshMessageFrameAdaptor
• Sub-PR D: Documentation — §13.10 acceptance matrix, primary/beta/legacy
  protocol labels, sdk.core/sdk.streaming/sdk.a2a/sdk.legacy discussion
• Once [Architecture Review][P0] Single protocol path (HTTP + CloudEvents) and ingress/egress adapters #5299 merges, unblocks [Architecture Review][P1] Standardize Storage SPI capabilities + TCK #5303 (Storage SPI capabilities + TCK)

[qqeasonchen]

Force-pushed 69e808f to fix an autocrlf/LF mismatch in the first commit (cf0d068). The diff is now clean: UniHttpServer.java +43/-28 and UniIngressService.java +988/-903 (was previously 1287+/- whole-file rewrite due to CRLF blobs). No code changes between the two commits — only line-ending normalization.

[qqeasonchen]

Force-pushed f17c3c7 to fold Sub-PR B (filter chain migration) into this PR as requested. The PR now spans Sub-PRs A+B in one commit:

Sub-PR A (commit 69e808f): route UniHttpServer 8 ingress endpoints through FrameAdaptor SPI → EventMeshFrame

Sub-PR B (commit f17c3c7): migrate ingress security pipeline (TokenAuthFilter / AclFilter / SignatureVerifierFilter / FilterChain) to operate on EventMeshFrame — auth/acl/signature now read directly from frame.attributes()

Diff: 8 files, +1247 / -958. The CloudEvent-based filter methods remain as @deprecated bridges so existing custom filters keep compiling; they will be removed in sub-PR C after the TCP path migrates.

CI is being re-triggered on the new head.

[qqeasonchen]

Folded the remaining #5299 sub-PRs into this PR (4cf9c8c), as requested. The PR now covers A + B + C + D.

Sub-PR A (69e808f) — UniHttpServer's 8 ingress endpoints go through FrameAdaptors.get("cloudevents").toFrame(...); UniIngressService gains publishBatchFrames / publishLiteFrame / pollLiteFrames / requestFrame / replyFrame.

Sub-PR B (f17c3c7) — ingress security pipeline (IngressFilter / FilterChain / AclFilter / TokenAuthFilter / SignatureVerifierFilter) operates on EventMeshFrame; tenant / signature / token read straight from frame.attributes().

Sub-PR C (4cf9c8c) — TCP egress path:

• TcpFrameCodec.encodePush / TcpPushChannel.deliver now take an EventMeshFrame, dropping the frame → CloudEvent → wire round trip
• deleted the dead CloudEvent-era egress SPI: CloudEventToPackageBody, MeshEventToPackageBody
• UniTcpServer drops the now-unused bodyMapper constructor parameter
• TCP ingress was already frame-native upstream (MeshMessagePackageRouter → toFrameSilent, TcpRequest holds a frame, NettyTcpPushChannel → fromFrameSilent), so no change was needed there.

Sub-PR D (4cf9c8c) — docs/eventmesh-uni-architecture-redesign.md §19.6: protocol status labels (primary / beta / legacy), per-path acceptance matrix, per-sub-PR status.

This commit also fixes two defects the previous head introduced:

• IngressFilter was missing import io.cloudevents.CloudEvent; (broke compileJava)
• EventMeshFrame imports landed in the wrong checkstyle ImportOrder group (org.apache.eventmesh is group 0; maxWarnings=0 so it failed CI)

Verified locally on JDK 21: compileJava + compileTestJava + checkstyleMain + checkstyleTest all green, 25 targeted tests pass (0 skipped, 0 failed).