[T4rk1n]

Shared storage and the streaming downlink it powers (#3930, #3931) were on by default for every app. To keep upgrades from changing how existing deployments run, both are now opt-in, and msgspec is no longer a hard dependency.

Changes

• Dash() has no shared storage unless shared_storage= is passed or DASH_SHARED_STORAGE is set. An unset, empty or none value disables it, and local turns it on. Streaming callbacks still work without it, each on its own connection, as before.
• msgspec moves out of the install requirements:
  • The new dash[shared_storage] extra brings it for the persistent LocalSharedStorage modes (persist, persist-reset).
  • dash[diskcache] and dash[redis] each include it, so those backends work from their own install.
• Without msgspec, in-memory LocalSharedStorage still works across workers, and its socket transport falls back to stdlib JSON.
• Every backend that writes encoded bytes somewhere lasting (persistent modes, Diskcache, Redis) raises an ImportError naming the extra to install, so the stored format never depends on what happens to be installed.
• A stream downlink request on an app without shared storage now gets a 403 instead of a 500.

Tests

• tests/shared_storage/test_no_msgspec.py runs with msgspec blocked. It checks that memory mode works between two processes over JSON, and that each lasting backend raises the right extra's error.
• test_stwg004 runs 4 gunicorn workers with msgspec blocked and a browser streaming across them.
• test_downlink_without_shared_storage_is_rejected covers the 403 on Flask, FastAPI and Quart.
• Tests that relied on the old default now pass LocalSharedStorage explicitly.

[github-actions]

Dash performance benchmarks

✅ all within thresholds

	scenario	metric	p90 (ms)	median	growth	baseline p90	note
✅	callback_chain	chain_ms	278.5	258.2	0.9x	499.1
✅	callback_chain	graph_ms	2.2	2.2	1.0x	2.4
✅	callback_fanout	fanout_ms	50.2	47.3	0.87x	92.5
✅	deep_nesting	render_ms	38.3	35.7	1.25x	56.8
✅	full_children_replace	replace_ms	3121.2	1118.3	18.05x	4697.1
✅	initial_render_large	render_ms	396.9	378.5	1.02x	694.4
✅	initial_render_small	render_ms	65.0	59.2	2.63x	104.0
✅	patch_append_nested	append_ms	113.2	73.2	2.51x	192.3
✅	patch_append_toplevel	append_ms	100.4	75.7	2.69x	140.2
✅	patch_scalar_update_large	update_ms	129.4	123.7	1.07x	202.9
✅	wildcard_all_resolve	wildcard_ms	177.4	169.5	0.94x	313.6
✅	wildcard_all_resolve	graph_ms	0.7	0.7	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: 0.62x - 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