The Sui Foundation published a post-mortem at 5:38 AM EDT on June 1, 2026, tracing three separate mainnet outages across May 28–29 to an edge-case flaw introduced in the v1.72 release — and explaining why each attempted fix left the next failure waiting.

The chain was down for a combined total of roughly 16 hours across two days. No user funds were at risk, and the network did not revert any committed transactions when it resumed after each halt.

The bug

Sui v1.72 introduced address balances, a new way for users to pay gas without using coin objects. The root cause involved an edge case in how the runtime handles "gas smashing" — the process of combining input coins before charging gas — when a transaction using address balances is canceled for insufficient funds.

When such a transaction is canceled with an InsufficientFundsForWithdraw error, gas smashing still attempts to charge the same address balance. The resulting negative delta, applied to a zero balance, caused validators to crash. It was an underflow: a transaction the network had already rejected was still being charged.

The first outage

The first halt began at approximately 7 AM PT on Thursday, May 28. The core team identified the bug and proposed a fix: skip gas smashing when a transaction is canceled with InsufficientFundsForWithdraw. The fix was proposed around noon PT, enough validators adopted it, and the network recovered by approximately 1:30 PM PT — roughly 6.5 hours of downtime.

The team's own post-mortem acknowledges the fix was interim. Changing gas logic in a live network carries compounding risks: interactions between address balances and coin objects are intricate, nodes must not fork while replaying old transactions with updated logic, and Sui's conservation checks ensure no transaction can create or destroy SUI. The team chose speed over completeness, accepting the known risk of a low-probability second halt to restore service while a more robust fix was developed.

The second outage

That risk materialized the next morning. The second halt began at approximately 5 AM PT on Friday, May 29.

The interim fix only handled the case where InsufficientFundsForWithdraw was the explicit cancellation reason. The post-mortem describes the scenario that bypassed it: a transaction using address balances, waiting on a hot shared object, could be canceled first for priority reasons and then also for insufficient funds. In that case the insufficient-funds error is masked by the earlier cancellation reason, the fix never fires, and the validator crashes with the identical underflow.

The robust fix was already nearly complete when Friday's outage hit. The team finalized it, proposed it to validators by approximately 8 AM PT, and the network came back by approximately 8:30 AM PT — roughly 3.5 hours of downtime.

The third outage

The network ran normally from approximately 8:30 AM PT until approximately 1:30 PM PT on Friday, when the scheduled epoch transition failed and the chain halted a third time.

The cause was unrelated to gas logic. At each epoch start, Sui validators run a distributed key generation (DKG) protocol that seeds the random beacon used by transactions requiring on-chain randomness. DKG requires a higher participation threshold than normal consensus. When validators restarted to adopt the Friday morning fix, participation wasn't high enough, and DKG correctly disabled itself for the epoch.

The latent bug: that failure verdict was never written to disk. Each subsequent validator restart came back up with no memory of DKG having failed. Randomness-dependent transactions entered a state where they could neither execute nor be canceled. The queue stalled, and so did the chain.

The team deployed a targeted fix and the network recovered by approximately 7:20 PM PT on Friday — roughly six hours of downtime for the third incident.

On-chain impact

Sui's DeFi TVL stood at $548.7 million on May 28, the day the first outage began. By June 1, it had declined to $513.6 million — a drop of approximately $35 million, or roughly 6%, across the outage window. The decline tracks a broader pullback across the period but the timing is consistent with capital moving off a chain that was intermittently unavailable.

What it means

The Sui Foundation's post-mortem is unusually candid about the interim-fix decision. The team knew the first fix had a known shortcoming and deployed it anyway, reasoning that returning the network to operation outweighed the residual risk. That risk materialized within hours.

The third outage compounded the problem. A latent bug in state persistence — unconnected to the original gas issue — surfaced precisely because the network had been restarted multiple times in quick succession. Complex systems under stress expose failure modes that quiet operation never reaches.

This is Sui's second full mainnet outage of 2026, a recurring reliability narrative for a chain that has positioned itself for institutional and DeFi adoption. The three fixes are now deployed and the network is operating normally. The questions that follow — about the maturity of the upgrade process, the depth of pre-deployment testing for state interactions — are ones the Sui team will be answering for some time.

Primary source: Sui Foundation post-mortem, published June 1, 2026 — blog.sui.io/sui-mainnet-halts-resolved-after-major-upgrade