fix: increased cache robustness for wire level cell resolution by dmadisetti · Pull Request #10109 · marimo-team/marimo

dmadisetti · 2026-07-08T19:33:33Z

📝 Summary

This PR resolves 3 issues with the cell level caching introduced.

@mo.persistent_cache functions were not correctly resolved and their subsequent calling would break if dependent on a unhashable stub
The stubbing mechanism frequently failed because unhashable stubs could not be resolved to compute downstream hashes. This PR introduces a change to include the runtime stub into the saved format, unlocking downstream hash computation.
The kernel can exit and finish before the background cache is fully done processing. Kernel hooks are already known surface area, this allows for hooks to be called after the kernel has run. This in turn is used to flush the cache to ensure export consistency.

vercel · 2026-07-08T19:33:38Z

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
marimo-docs	Ready	Preview, Comment	Jul 8, 2026 8:28pm

Bundle a session's per-cell and per-function caches into the output of marimo export html-wasm --execute, so a dependency-free browser (Pyodide) can skip recomputation. Stacked on the cache-robustness PR (#10109), which provides the graceful-degrade / hashing-parity restore machinery this relies on to consume the bundled caches without the original deps. This pull request was authored by a coding agent.

Copilot

Pull request overview

This PR bumps the persistent cache schema to v5 and hardens cell-level caching in environments where optional dependencies (e.g. torch in WASM) are unavailable, while also ensuring background cache writes are flushed during kernel teardown for export/durability consistency.

Changes:

Improve stub/restore behavior so unmaterializable defs (missing optional deps) still allow downstream cache-key reproduction via persisted content digests, and cached wrappers (@mo.cache / @mo.persistent_cache) can be reconstituted as working wrappers on restore.
Avoid fetching blobs that can’t be deserialized in the current environment (root module absent), substituting an UnhashableStub that carries the persisted digest.
Add teardown-time flushing for lazy loader background writes and adjust invalidation semantics to mark manifests stale (without destroying recoverable cache entries).

Reviewed changes

Copilot reviewed 18 out of 18 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
tests/_save/stubs/test_module_stub.py	Adds coverage for ModuleStub version fallback to root package version and empty-version behavior.
tests/_save/loaders/test_loader.py	Adds coverage ensuring absent-module blobs are not fetched and content digests propagate into stubs.
tests/_runtime/test_cached_stage.py	Updates invalidation expectations (stale-only) and adds scenario coverage for serializable siblings not forcing live reruns.
marimo/_session/managers/kernel.py	Attempts cleaner shutdown and gives the kernel time to flush pending work before termination.
marimo/_save/stubs/module_stub.py	Captures root package `__version__` for submodule aliases to preserve version-pinned hashes on restore.
marimo/_save/stubs/lazy_stub.py	Extends schema (function tuples) and enriches `UnhashableStub` with `content_hash` for downstream hashing.
marimo/_save/stubs/function_stub.py	Adds cached-wrapper support (`is_cached`) and dump/restore helpers for schema persistence.
marimo/_save/loaders/loader.py	Introduces a `flush()` hook for loaders to drain async writes.
marimo/_save/loaders/lazy.py	Adds module-availability gating for blob fetch, persists content digests, and wires in global flush for active loaders.
marimo/_save/loaders/init.py	Exposes `flush_active_caches` from loaders package.
marimo/_save/hash.py	Teaches the hasher to replay persisted content digests from restored stubs and to handle marimo-owned stubs without flipping cache mode.
marimo/_save/cache.py	Bumps cache version to 5; improves restore degradation behavior and adds cached-wrapper stub capture/restore logic.
marimo/_runtime/runtime.py	Adds callback teardown phase while runtime context is still alive.
marimo/_runtime/kernel_lifecycle.py	Invokes callback teardown before tearing down runtime context.
marimo/_runtime/executor/lifecycles/cached.py	Improves logging/UI def detection and changes invalidation to stale-only (no clear).
marimo/_runtime/callbacks/protocol.py	Adds `SupportsTeardown` runtime-checkable protocol.
marimo/_runtime/callbacks/cache.py	Implements teardown flush of active caches for durability on kernel teardown.
marimo/_runtime/callbacks/init.py	Re-exports `SupportsTeardown`.

cubic-dev-ai

All reported issues were addressed

_{Reply with feedback, questions, or to request a fix.

Re-trigger cubic}

Bundle a session's per-cell and per-function caches into the output of marimo export html-wasm --execute, so a dependency-free browser (Pyodide) can skip recomputation. Stacked on the cache-robustness PR (#10109), which provides the graceful-degrade / hashing-parity restore machinery this relies on to consume the bundled caches without the original deps. This pull request was authored by a coding agent.

Copilot

Pull request overview

Copilot reviewed 18 out of 18 changed files in this pull request and generated 2 comments.

dmadisetti · 2026-07-08T22:13:01Z

+    def __init__(self, function: Any, is_cached: bool = False) -> None:
+        self.is_cached = is_cached
+        if is_cached:
+            # Capture the wrapped body's source (incl. the decorator line).
+            function = function.__wrapped__


I don't think anyone has FunctionStub pickles yet, so this is fine. Technically breaking but I expect 0 impact

mscolnick · 2026-07-08T22:21:43Z

+            if var not in variable_hashes:
+                # NB. persist a digest for data primitives so a consumer
+                # reproduces its key without the value.
+                digest = _maybe_content_digest(obj)


should this happen in a thread? is this expensive to do on the main thread and in a for loop

Good point. will be care for when this runs through WASM. Doing this as a follow up to unblock my other branch in the meantime

mscolnick

looks great! one small comment on perf for the lazy store

github-actions · 2026-07-08T23:13:05Z

🚀 Development release published. You may be able to view the changes at https://marimo.app?v=0.23.14-dev37

Bundle a session's per-cell and per-function caches into the output of marimo export html-wasm --execute, so a dependency-free browser (Pyodide) can skip recomputation. Stacked on the cache-robustness PR (#10109), which provides the graceful-degrade / hashing-parity restore machinery this relies on to consume the bundled caches without the original deps. This pull request was authored by a coding agent.

Copilot AI review requested due to automatic review settings July 8, 2026 19:33

Copilot started reviewing on behalf of dmadisetti July 8, 2026 19:34 View session

dmadisetti added the bug Something isn't working label Jul 8, 2026

dmadisetti force-pushed the dm/cache-robustness branch from 6450f9d to 32b44cb Compare July 8, 2026 19:42

vercel Bot deployed to Preview July 8, 2026 19:44 View deployment

Copilot AI reviewed Jul 8, 2026

View reviewed changes

cubic-dev-ai Bot reviewed Jul 8, 2026

View reviewed changes

Comment thread marimo/_save/stubs/function_stub.py

Comment thread marimo/_save/cache.py

Comment thread marimo/_save/hash.py

Comment thread marimo/_save/stubs/lazy_stub.py

dmadisetti requested review from kirangadhave and mscolnick July 8, 2026 20:17

fix: increased cache robustness for wire level cell resolution

72f09a2

dmadisetti force-pushed the dm/cache-robustness branch from 32b44cb to 72f09a2 Compare July 8, 2026 20:27

vercel Bot deployed to Preview July 8, 2026 20:28 View deployment

mscolnick requested a review from Copilot July 8, 2026 22:01

Copilot started reviewing on behalf of mscolnick July 8, 2026 22:02 View session

Copilot AI reviewed Jul 8, 2026

View reviewed changes

dmadisetti changed the title ~~fix(cache v5): increased cache robustness for wire level cell resolution~~ fix: increased cache robustness for wire level cell resolution Jul 8, 2026

mscolnick reviewed Jul 8, 2026

View reviewed changes

mscolnick approved these changes Jul 8, 2026

View reviewed changes

dmadisetti merged commit b8b4ebb into main Jul 8, 2026
45 checks passed

dmadisetti deleted the dm/cache-robustness branch July 8, 2026 23:08

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: increased cache robustness for wire level cell resolution#10109

fix: increased cache robustness for wire level cell resolution#10109
dmadisetti merged 1 commit into
mainfrom
dm/cache-robustness

dmadisetti commented Jul 8, 2026 •

edited

Loading

Uh oh!

vercel Bot commented Jul 8, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

cubic-dev-ai Bot left a comment •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

dmadisetti Jul 8, 2026

Uh oh!

Uh oh!

mscolnick Jul 8, 2026

Uh oh!

dmadisetti Jul 8, 2026

Uh oh!

mscolnick left a comment

Uh oh!

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

dmadisetti commented Jul 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

📝 Summary

Uh oh!

vercel Bot commented Jul 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

dmadisetti Jul 8, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mscolnick Jul 8, 2026

Choose a reason for hiding this comment

Uh oh!

dmadisetti Jul 8, 2026

Choose a reason for hiding this comment

Uh oh!

mscolnick left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions Bot commented Jul 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

dmadisetti commented Jul 8, 2026 •

edited

Loading

vercel Bot commented Jul 8, 2026 •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading