Mainnet writes fail on both prod relayers: Enoki dry_run_failed (MoveAbort balance::destroy_zero in sponsored upload)

[dun999]

Summary

All mainnet memory writes through the MCP server currently fail at the Walrus upload step with an Enoki sponsorship error. Reads (memwal_recall) and auth work fine. Reproduced consistently on both production relayers.

Error (verbatim)

Tool error: remember job failed: walrus upload failed: Internal Error: walrus upload failed:
Enoki API error (400): {"errors":[{"code":"dry_run_failed","message":"Dry run failed, could
not automatically determine a budget: MoveAbort(MoveLocation { module: ModuleId { address:
0000000000000000000000000000000000000000000000000000000000000002, name: Identifier(\"balance\") },
function: 9, instruction: 8, function_name: Some(\"destroy_zero\") }, 0) in command 9"}]}

Repro

1. @mysten-incubation/memwal-mcp v0.0.5, fresh login via memwal_login (delegate key registered on-chain, credentials saved).
2. Call memwal_remember with any short text and namespace, e.g. {"text": "[FACT] smoketest :: ...", "namespace": "gardener-smoketest"}.
3. Fails identically every time. Tried 2026-07-02 between ~11:25 and ~11:40 UTC:
   • https://relayer.memory.walrus.xyz: fails
   • https://relayer.memwal.ai: fails with the same error
4. memwal_recall on the same account/namespace works (returns "No matching memories found." cleanly), so relayer auth and the read path are healthy.

Account (mainnet): 0x057359bbb9bc3b6eed2b1c04d51bcbee421b9c802e90543b0ca65844c006d7cb. The object is active with 3 registered delegate keys, verified via sui_getObject.

Analysis

The abort is in the Sui framework's balance::destroy_zero (abort code 0 = ENonZero) at command 9 of the sponsored PTB that Enoki dry-runs. That means the transaction being constructed destroys a coin/balance that still has a non-zero value, i.e. the WAL payment split inside the sponsored upload flow leaves a non-zero remainder where the code expects exact change. Nothing client-side affects this (payload size, namespace, relayer choice, fresh login), which points at the sponsor-side coin selection / storage-cost calculation, possibly triggered by a recent storage price or epoch change.

Impact

Every MCP user is currently unable to write memories to mainnet (Session 5 proof runs included). Reads unaffected.

Environment

• @mysten-incubation/memwal-mcp 0.0.5 (stdio), --prod
• Node.js 24.14.0, Windows 11

[github-actions]

Thank you for opening this issue, a team member will review it shortly.

[ducnmm]

Root-caused — not an Enoki issue. Fixed and deployed; mainnet writes work again.

The Walrus register PTB pre-funds an exact WAL payment from the client's cached storage price, then asserts zero change via coin::destroy_zero. When the on-chain price drifts down between the cached read and execution, the contract deducts less WAL, the leftover trips destroy_zero (ENonZero), and Enoki just surfaces it in its dry-run. Confirmed: on-chain storage_price moved 71464→70922 during the incident, and a fresh relayer served writes then flipped to 100% destroy_zero at the next price move.

Fix: detect the abort → refresh the client and retry (classified transient, not a permanent MoveAbort) + shorten the cached-price max-age. Hot-fixed on the prod relayer; the same change is up for review in #352.

Also worth hardening upstream: @mysten/walrus's exact-change + destroy_zero is racy against a moving on-chain price.

@dun999 please retry — should work now.

[DrVelvetFog]

Nice root-cause + fast turnaround. The upstream hardening you flagged is now filed as MystenLabs/ts-sdks#1127 — confirmed the same exact-change + destroy_zero pattern sits in all three @mysten/walrus payment paths (createStorage, registerBlob, extendBlob), the price cache only invalidates on RetryableWalrusClientError (which this abort isn't), and 1.2.3 is unchanged. Proposed merge-the-remainder-back as the race-free fix, with your #352 evidence linked, and offered the PR.

[dun999]

Retried after the hotfix (14:39 UTC): memwal_remember and memwal_recall both succeed on mainnet again. Example write from the retest: https://walruscan.com/mainnet/blob/1aVoLN5nJIipBNKVdhgk0f1r54Yyphh_5Vc9dxfL_pw. Thanks for the fast turnaround and the detailed root cause, the cached-price explanation matches everything we observed.