Conversation
feat: add Slack alert for exhausted Walrus uploads
… 2.17.0 Aligns sidecar SDKs with the current Walrus on-chain package so future package upgrades are picked up by a fresh client without breaking API surface. Bumps are minor-version only (no breaking changes observed).
When the Walrus on-chain package is upgraded after the sidecar boots, the cached @mysten/walrus client surfaces MoveAbort EWrongVersion from walrus::system::inner_mut until restart. This change makes that self-recovering instead of redeploy-dependent. Sidecar (TS): - isWalrusPackageVersionMismatch detector anchored on the location fragment "::system::inner_mut" (cross-transport stable) and the symbolic "EWrongVersion" (gRPC/GraphQL). - Upload catch block calls existing refreshWalrusClient() and logs on-chain System.version before vs after refresh via systemObject(). Server (Rust): - classify_sidecar_error: EWrongVersion is now Transient so Apalis retries against the refreshed client instead of marking Dead. - AlertManager gains notify_walrus_package_upgrade_detected backed by WalrusPackageUpgradeDetectedAlert; fired from the wallet-job upload error path when the same pattern is observed. Tests: - 11 new sidecar tests covering JSON-RPC + gRPC error formats, case-insensitivity, and false-positive guards (balance::split, bare tokens, abort code 1 from non-Walrus modules). - 4 new Rust tests on the classifier carve-out, the Walrus-only detector, and the new Slack payload (version diff, missing-data, oversized-error truncation).
User feedback (2026-05-27) flagged confusion around namespace overwrite, hierarchy, and restore() response. Add two dedicated sections sourced from the actual server behavior: - Namespace Semantics: opaque flat strings, exact-equality match, no validation, no hierarchy, and every remember() is append-only (no upsert by namespace). - Restore Semantics: response field meanings (restored / skipped / total), default limit, no pagination cursor, and bounded concurrency model (10 downloads, 3 decrypts) so on-call can reason about latency. Also re-points the existing recall() example to the new object form so the recommended call style is consistent across the doc.
Positional recall(query, limit, namespace) was easy to mis-read as
recall(query, namespace), so both SDKs now accept an object/dataclass
input alongside the existing positional signature. Positional calls
keep working — this is additive, not a breaking change.
TS SDK:
- New RecallParams type exposing { query, limit?, namespace?, maxDistance?, topK? }
- recall() overloaded: recall(params) and recall(query, limitOrOptions?, namespace?)
- JSDoc reframed to call out the object form as preferred
Python SDK:
- New RecallParams dataclass with the same shape
- recall() now accepts either str or RecallParams as the first arg
- Sync MemWalSync.recall() mirrors the overload
- Both new and existing forms exercised by test_client.py (6 recall tests)
Also fixes a default-value drift: Python restore() defaulted to limit=50
while TS and the server default to 10. Bringing Python in line removes
a foot-gun where the same call across SDKs returned different result
counts. Existing callers that pass an explicit limit are unaffected.
Both SDKs' restore() docstrings now spell out response field meanings,
the no-pagination caveat, and the linear-in-limit performance profile.
… fns Next.js server-action modules reject any non-async-function export with a build-time error. WALM-53 user feedback called out that this is easy to miss because the failure mode is a build error pointing at the file rather than at the offending export. Add a README section with the rule and a wrong/right example so the constraint is discoverable next to the running-locally instructions. (Originally drafted as apps/chatbot/CLAUDE.md, but CLAUDE.md is gitignored repo-wide as a per-dev workspace overlay convention, so the note belongs in committed docs instead.)
Phase 1 of the WALM-53 recall() rollout plan: the new object-style
recall({ query, limit, namespace }) overload landed in the previous
commit; this commit adds a formal @deprecated JSDoc tag on the
positional signature so IDEs surface the warning at call sites. The
positional form keeps working — this is signaling intent, not removing
behavior.
Phase 1 doc migration: every recall() snippet in user-facing docs and
examples now uses the recommended object form so the first example a
new user copy-pastes is the unambiguous one. Positional form still
works (and the test suite still exercises it) — this is a doc-only
change with no runtime effect.
Updated:
- top-level README.md
- SKILL.md (Quick Start)
- docs/getting-started/quick-start.md
- docs/sdk/{quick-start,examples}.md, docs/sdk/usage/memwal.md
- docs/python-sdk/{quick-start,usage}.md, docs/python-sdk/usage/memwal.md
- docs/examples/example-apps.md
- docs/relayer/nautilus-tee.md
- packages/python-sdk-memwal/README.md
- packages/python-sdk-memwal/memwal/__init__.py + client.py docstrings
- packages/python-sdk-memwal/examples/{async_remember_demo,interactive_demo}.py
Left on positional intentionally:
- tests/test_client.py — verifies the backwards-compat path still works
- memwal/middleware.py — internal, will follow when callers update
- scripts/test-namespace.ts, test-zero-restore.ts — internal smoke
…comment-references feat: clean up internal comment references
…4-marketing-analytics-for-memwalai-launch-traffic Add GA4 launch analytics
The new wallet-job upload path (jobs.rs:866) already classified Walrus package-version mismatches as Transient + fired the informational alert. The legacy RememberJob handler that still drains older queue rows was calling fail!() unconditionally on UploadBlobError::App, so the same error would mark the row failed and skip the Apalis retry that the sidecar's client refresh was meant to set up. Carve out the same is_walrus_package_version_mismatch detection ahead of fail!() — fire the same alert, log a warn, and return Err WITHOUT writing status='failed'. Apalis then retries against the refreshed client; the row only flips to 'failed' if the retry budget is exhausted by some other transient error. Reported by @ducnmm on PR #199 review.
The package-upgrade alert was described as one-shot but currently fires for every matching upload failure. A real upgrade event triggers EWrongVersion on every queued job concurrently, so the alerter could emit dozens of identical messages back-to-back. Add an in-memory dedup gate inside AlertManager keyed by (sui_network, sidecar_walrus_dep_version) — the natural identity of a distinct upgrade event: - Same key, fired within the window → suppress. - Different network or different dep version → independent event, fires immediately. - After the window expires, the key re-fires (so a second upgrade on the same network surfaces). Window default 600s, env-overridable via WALRUS_PACKAGE_UPGRADE_ALERT_DEDUP_SECS so ops can widen / narrow it without redeploying. Reported by @ducnmm on PR #199 review.
Cast undefined / null through `as unknown as string` so the empty-input test keeps working under any tsconfig strictness, instead of relying on @ts-expect-error directives that go stale if the test file is ever typechecked under different flags. Reported by @ducnmm on PR #199 review.
Code comments should carry stable technical rationale, not Linear-ticket narrative — the ticket ID rots once the system is renamed or the ticket is closed. Reframe the carve-out comment as the invariant it's actually protecting: the sidecar refreshes the cached Walrus client before the error bubbles up, so the row must not flip to status='failed' here because the next Apalis attempt is about to succeed.
Per repo convention, in-source comments and doc text should carry stable technical rationale — not the Linear ticket that introduced the change. Ticket IDs rot once renamed or closed; a "WALM-53 — Response semantics" heading reads like a release note rather than a contract. Strip the WALM-53 prefix from every docstring, JSDoc block, test docstring, and markdown section it was introduced in. No behavior change; the prose already explains why the symbol exists.
…package-version-mismatch-without-redeploy fix(MEM-34): handle Walrus package version mismatch without redeploy
…sks-from-user-feedback fix(WALM-53): address DX risks from 2026-05-27 user feedback
staging <- dev
harrymove-ctrl
approved these changes
May 28, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.