[T4rk1n]

Follow-up to #4095. With shared storage off, a streaming callback over HTTP fell back to one connection per stream. That works on a laptop, then breaks once deployed (browsers cap connections per host), so it gave false confidence. Now it fails loudly, the same way locally and when deployed.

Changes

• _setup_server raises StreamCallbackError when the app has streaming callbacks that run over HTTP and no shared storage. The message names the callbacks and how to fix it: shared_storage=, DASH_SHARED_STORAGE, or WebSocket. It runs next to the background-callback validation, so every request fails with the same error until it's fixed.
• WebSocket streaming callbacks (websocket_callbacks=True or websocket=True) are exempt only on a backend that serves WebSockets (FastAPI, Quart). On Flask those flags still mean HTTP, so the check applies.
• Docs (Dash docstring, .ai/ARCHITECTURE.md) no longer describe the fallback. The doc notes that browsers without SharedWorker still stream WebSocket-app callbacks over HTTP; the server can't detect that.

Tests

• stcb025/026: raises at setup and on every request.
• stcb027: FastAPI WebSocket streams (app-level and per-callback) need no shared storage.
• stcb029: the same flags on Flask still raise.
• stcb028: shared storage enabled sets up fine.
• tests/streaming/conftest.py sets DASH_SHARED_STORAGE=local, so streaming tests run with shared storage unless they pass their own.

[github-actions]

Dash performance benchmarks

✅ all within thresholds

	scenario	metric	p90 (ms)	median	growth	baseline p90	note
✅	callback_chain	chain_ms	433.9	419.8	0.93x	499.1
✅	callback_chain	graph_ms	2.9	2.9	1.0x	2.4
✅	callback_fanout	fanout_ms	82.3	79.8	0.92x	92.5
✅	deep_nesting	render_ms	56.4	54.8	1.18x	56.8
✅	full_children_replace	replace_ms	5505.3	1975.9	18.39x	4697.1
✅	initial_render_large	render_ms	631.4	590.8	1.06x	694.4
✅	initial_render_small	render_ms	119.9	98.7	0.93x	104.0
✅	patch_append_nested	append_ms	126.7	96.4	2.22x	192.3
✅	patch_append_toplevel	append_ms	117.2	88.5	2.14x	140.2
✅	patch_scalar_update_large	update_ms	144.7	134.1	0.91x	202.9
✅	wildcard_all_resolve	wildcard_ms	310.6	298.2	0.93x	313.6
✅	wildcard_all_resolve	graph_ms	1.3	1.3	1.0x	1.3

growth = late-third / early-third per-op time; ~1 is flat, a large value means the per-op cost scales with accumulated state.

machine scale vs baseline: 1.03x - divided out of the baseline ratios so they compare like for like (the absolute warn/fail ceilings are left un-scaled); calibrated on initial_render_small.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud