Sui just fixed a withdrawal bug before main rollout. Operators still have cleanup work.

Sui testnet just moved to protocol version 130. MystenLabs released testnet-v1.76.0 on July 21, 2026. Mainnet is still on protocol version 129, according to the July 15 mainnet-v1.75.2 release.

The real change is not the version number. Protocol 130 fixes a withdrawal-ordering bug in Sui's balance accounting. When deposits and withdrawals happened in the same transaction, unrelated withdrawals from that same account could get blocked inside the same consensus commit.

At the contract level, that is a bookkeeping bug with sharp edges. The old path used running maximum withdrawal amounts. The new path uses per-account net amounts from transaction effects. That sounds small until your bag is the account whose withdrawal fails for no economic reason.

The code does what the announcement says, plus one operational demand. Sui is also adding a new gRPC RPC Store for indexes. Node operators must run a one-time resumable reindexing pass. During startup, HTTP availability can be delayed until live object reindexing finishes.

And that's the catch. The reindex is resumable, so restarts don't force operators back to zero. Historical indexing continues in the background after live object indexing completes. Normal RPC operations should continue once the node serves.

MystenLabs also changed how indexing pipelines work. Pipelines are now grouped into cohorts based on distance from the chain tip. The default boundary is 25,000 checkpoints, according to the release notes. A lagging pipeline should no longer slow down one that is already caught up.

This is a backpressure problem, not a marketing problem. If one slow indexer can block faster ones, the RPC layer inherits failure from its slowest task. Splitting cohorts is the boring fix engineers usually make after production load teaches them where the queue bends.

There is also config cleanup. The admission-queue-bypass-fraction validator field is silently removed in v1.76.0. Operators with that field in config files need to delete it. The field is ignored if present, which limits breakage but still means fleets need a hygiene pass.

Fork recovery remains conditional too. Forks caused by the currently running binary still halt the node. Operators still need intervention or a corrected binary. Automatic recovery only applies after the network certifies the canonical outcome and a new binary is deployed.

So protocol 130 is useful, but not glamorous. Sui is still calibrating balance accounting and RPC infrastructure under real load. That is normal for a maturing chain. It is also exactly why testnet releases matter before mainnet gets the same rules.