Testnet Retro - A Case of Divergent Mempools | Aztec Blog

The major incident on May 29th, 2025 involved a dramatic slowdown in block production on the Aztec testnet. As the network rapidly scaled to more than 19,000 nodes, it encountered unforeseen hurdles in achieving efficient transaction propagation and consensus. This event shed light on the growing pains of scaling zero-knowledge rollup ecosystems and offered a valuable case study in protocol resilience amid surging adoption.

At the heart of the problem were “divergent mempools”—a situation caused by many nodes running with default, low-capacity mempool settings and lacking mechanisms like priority fees to organize and prioritize transactions. This led to arbitrary evictions of older transactions from the mempools, which in turn meant that validator nodes didn’t have the full transaction set necessary to verify incoming block proposals. The end result: stalled block production and network-wide disruption.