[http-client-python] Add automatic SSE reconnection

[Libba Lawrence (l0lawrence)]

Structured SSE streams currently stop at unexpected EOF and require callers to manually resume with last_event_id through an internal workaround. This adds automatic reconnection so generated clients follow the SSE processing model without requiring application-owned reconnect loops.

The generated stream runtime now:

• reconnects after unexpected EOF using the default three-second delay or the latest valid retry: value --- Do we want a 3 second default or would we prefer no time delay, just immediately try to reconnect?
• preserves and sends the latest event ID through Last-Event-ID
• stops reconnecting after terminal events and HTTP 204 responses (per the spec)
• routes reconnect failures through the generated operation's normal error mapping
• supports both sync and async streams

Closes #11761

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@typespec/http-client-python@11904

commit: bacc909

[github-actions]

All changed packages have been documented.

• ✅ @typespec/http-client-python

Show changes

@typespec/http-client-python - feature ✏️

Automatically reconnect SSE streams after an unexpected disconnect, honoring server-provided retry delays and resuming with the latest event ID.,> ,> python,> with client.events.stream() as stream:,> for event in stream:,> print(event),>

[azure-sdk-automation]

You can try these changes here

🛝 Playground	🌐 Website	🛝 VSCode Extension

[Copilot]

🟡 Changes recommended

The reconnect implementation conflicts with the stated azure-core requirement, and event ID header handling has an unresolved bug.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Adds automatic sync and async SSE reconnection for generated Python clients, including retry delays, event ID resumption, terminal-event handling, and HTTP 204 stopping.

Changes:

• Extends streaming runtimes with reconnection behavior.
• Generates reconnect callbacks through the request and error-mapping pipeline.
• Adds sync/async tests and a feature changelog.

File summaries

File	Summary
packages/http-client-python/tests/unit/test_sse_streaming.py	Tests generated reconnect behavior.
packages/http-client-python/tests/mock_api/shared/test_streaming_sse.py	Adds synchronous SSE reconnection coverage.
packages/http-client-python/tests/mock_api/shared/asynctests/test_streaming_sse_async.py	Adds asynchronous SSE reconnection coverage.
packages/http-client-python/generator/pygen/codegen/templates/streaming_base.py.jinja2	Adds reconnect runtime behavior. Critical (3 votes): Resolve the requirement to use an azure-core reconnect primitive instead of a generator-vendored loop.
packages/http-client-python/generator/pygen/codegen/serializers/builder_serializer.py	Generates reconnect callbacks. Moderate (2 votes): Omit Last-Event-ID when the event ID buffer is empty. Nit (1 vote each): Add redirect coverage and end-to-end reconnect error-mapping coverage.
.chronus/changes/http-client-python-sse-reconnection-2026-09-08.md	Records the feature release change.

Review details

Suppressed comments (3)

packages/http-client-python/generator/pygen/codegen/serializers/builder_serializer.py:1405

• The normal operation path deserializes declared custom error responses and attaches them as HttpResponseError.model after map_error (the logic at builder_serializer.py:1170-1274). This reconnect path only calls map_error and then raises a bare HttpResponseError, so an unmapped custom or ranged error status during reconnect loses the generated error model and does not follow the operation's normal error behavior.

                "        map_error(status_code=_reconnect_response.status_code, "
                "response=_reconnect_response, error_map=error_map)"
            )
            retval.append("        raise HttpResponseError(response=_reconnect_response)")

packages/http-client-python/generator/pygen/codegen/serializers/builder_serializer.py:1394

• The linked requirement also calls out preserving 301/307 redirect handling, but the new sync/async integration tests only exercise 200/204 responses and do not drive a redirect during reconnect. Add a mock-server case for both modes so the generated send_request path and redirect policy are verified rather than only asserted as generated text.

                f"    _reconnect_response = {self._call_method}self._client.send_request("
                "_request, stream=True, **kwargs)"

packages/http-client-python/generator/pygen/codegen/serializers/builder_serializer.py:1403

• The new reconnect-failure branch is only checked as generated text; neither sync nor async tests execute a generated SSE operation that returns a non-success reconnect response and verifies the error mapping. Add an end-to-end generated-client case for this branch, including a custom error response, so this behavior is protected.

            retval.append(f"    if _reconnect_response.status_code not in {reconnect_status_codes!r}:")
            retval.append(f"        {self._call_method}_reconnect_response.read()")
            retval.append(
                "        map_error(status_code=_reconnect_response.status_code, "
                "response=_reconnect_response, error_map=error_map)"

• Files reviewed: 6/6 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

🟡 Changes recommended

Two moderate findings remain unresolved regarding error-model preservation and use of the required azure-core streaming implementation.

Get a fresh assessment by requesting another Copilot review.

Review details

Suppressed comments (1)

packages/http-client-python/generator/pygen/codegen/templates/streaming_base.py.jinja2:709

• The linked #11761 requirements explicitly call for using the azure-core streaming/transport implementation rather than a generator-vendored reconnect loop. This code puts the state machine in streaming_base.py.jinja2, which is copied into every generated client, so the generated runtime will fork from core behavior and miss future core fixes; integrate the core implementation (or update the stated requirement/dependency) instead.

        terminal = False
        try:
            while True:

• Files reviewed: 7/7 changed files
• Comments generated: 1
• Review effort level: Lite

[Copilot]

🔵 Needs a closer look

The implementation embeds the reconnect loop in generated code and loses declared error models for reconnect failures.

Review details

Suppressed comments (2)

packages/http-client-python/generator/pygen/codegen/templates/streaming_base.py.jinja2:713

• The linked issue explicitly defers this behavior until the azure-core streaming/transport layer provides a reconnect primitive and says not to add a generator-vendored reconnect loop. This new while True loop in the generated streaming runtime (together with the generated callback below) reintroduces that deferred architecture, so please integrate the azure-core primitive instead of embedding reconnection policy in every generated client.

            while True:
                if getattr(self._response, "status_code", None) == 204:
                    break

packages/http-client-python/generator/pygen/codegen/serializers/builder_serializer.py:1404

• This does not follow the operation's normal error path for declared error models: handle_error_response deserializes the reconnect error body and attaches model=error to HttpResponseError, whereas this branch only calls map_error and then raises without a model. A reconnect that returns a custom 4xx/5xx therefore loses the generated error payload; reuse the same error deserialization/mapping path for the reconnect response.

            retval.append(
                "        map_error(status_code=_reconnect_response.status_code, "
                "response=_reconnect_response, error_map=error_map)"
            )
            retval.append("        raise HttpResponseError(response=_reconnect_response)")

• Files reviewed: 7/7 changed files
• Comments generated: 0 new
• Review effort level: Lite

[Copilot]

Copilot review overview

🔵 Needs a closer look

Reconnect failures lose operation-specific error mapping, and the implementation does not use the azure-core reconnect primitive required by the linked issue.

Review effort: Lite
Findings: 1

Open (1)

• For a reconnect response whose status is not one of the standard error_map entries, map_error…

[github-actions]

Python emitter diff

Baseline gh:bcd8ff10ca996d912eebff01b25a777fbf5bd51c vs this PR.

Diff summary: 24 file(s), +1476 / -244

Rendered diff: inline on the run summary, or the emitter-diff-html artifact.

Informational check (eng/emitter-diff); does not block the PR.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Moderate unresolved issues remain in the reconnect runtime and event ID handling.

Review effort: Lite
Findings: None

Resolved since last review (1)

• For a reconnect response whose status is not one of the standard error_map entries, map_error…

Previously missed (1)

In code that hasn't changed since last review

Persist SSE IDs parsed from incomplete event blocks

packages/​http-client-python/​generator/​pygen/​codegen/​templates/​streaming_base.py.jinja2:491

A valid SSE id: field updates the last-event-ID buffer when it is parsed, even if the current event block is incomplete. Deferring _last_id until _dispatch() means an EOF after id: second reconnects with a stale ID, so update the committed buffer in this branch as well as the pending value, and adjust the new incomplete-block test accordingly.

[iscai-msft]

yay amazing

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved findings include incorrect incomplete-event ID handling and missing or inconsistent reconnect integration and coverage.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1 · 2

Open (3)

• Update last-event-ID when parsing id fields · New
• Use azure-core reconnect primitive instead of vendored loop · New
• Add end-to-end tests for reconnect error deserialization · New