Cronos v1.8 moves its Block-STM parallel executor into the Cosmos SDK’s own transaction runner.
The default executor is still “sequential,” and the migration guide shows that one wrong combination of config settings can halt a node.
The first wave of coverage on Cronos v1.8 said its parallel execution engine had moved into the SDK. That is true, but it leaves out what operators and developers actually need to know: whether anything changes when they upgrade.
The answer comes from the code and from Cronos’ own v1.7-to-v1.8 migration guide. Out of the box, almost nothing changes. Parallel execution is still a setting you turn on yourself. It is safe to skip, but the guide warns that a few nearby settings, combined the wrong way, can make a node fall out of consensus.
Here is what the release changes, which settings you need to edit, and where the risks are.
What Actually Moved Into The SDK
In Cronos v1.7, parallel execution came from a separate library. The v1.7.9 code wired Block-STM in through Ethermint’s STMTxExecutor, which relied on the standalone go-block-stm package. That package had been part of Cronos since the chain moved to Cosmos SDK v0.50.
In v1.8.1, tagged on October 26, the setup is different. The node now calls SetBlockSTMTxRunner with txnrunner.NewSTMRunner, imported from [github.com/cosmos/cosmos-sdk/baseapp/txnrunner](https://github.com/cosmos/cosmos-sdk/baseapp/txnrunner), the transaction-runner package inside the Cosmos SDK. The standalone go-block-stm dependency is gone from the module file. The release runs on Cosmos SDK v0.54 (Cronos’ fork), CometBFT v0.39 and go-ethereum v1.16.9.
In practice, the parallel engine is now part of the framework rather than an add-on. That makes it easier to maintain and easier to keep in line with the upstream SDK. It does not make it the default.
Parallel Execution Is Still Opt-In
The node only turns on the parallel engine when one setting has one exact value. In app/app.go, it checks whether block-executor equals the string "block-stm". If it doesn’t, the node uses the SDK’s default runner, which processes transactions one at a time.
The migration guide’s sample config states it plainly: "block-stm" for parallel execution; "sequential" is the default.
So a node operator who upgrades without changing app.toml gets a v1.8 node that still executes blocks one transaction at a time. The parallel engine is there but unused. Nothing breaks, but none of the speed gains the announcement described apply until the setting changes.
The guide’s advice goes further than allowing it. For v1.8 it recommends the same setup for every role: Block-STM on, zero-copy = true and optimistic execution on, for validators, public RPC nodes and archive nodes alike.
The Three Lines That Turn It On
Turning it on takes three lines in the [evm] section of app.toml:
Ini, TOML
[evm]block-executor = "block-stm"block-stm-workers = 0block-stm-pre-estimate = true
block-stm-workers: a value of0doesn’t mean zero workers. The code replaces it with the smaller of Go’sGOMAXPROCSand the machine’s CPU count. RPC operators who run out of CPU for queries are told to “pinblock-stm-workersbelow your core count rather than turning Block-STM off.”block-stm-pre-estimate: this enables a feature Cronos added earlier to improve worst-case performance.
The guide pairs these with memiavl storage settings:
Ini, TOML
[memiavl]enable = truezero-copy = true
There is a subtle reason to keep zero-copy on. Setting it to false has only one benefit, re-enabling the SDK address cache, and Block-STM turns that cache off anyway. The code calls sdk.SetAddrCacheEnabled(false) whenever the parallel engine is active. With Block-STM running, zero-copy = false just “adds a copy to every latest-height read and gains nothing.”
What Breaks If You Copy The Wrong Config
The parallel engine does not cause the breakage. The risk is in the settings around it.
Sequential execution plus disabled optimistic execution. The node only enables the memiavl node cache when block-executor = "sequential" and disable-optimistic-execution = true are both set. With any other combination, the code forces the cache size to zero. The guide warns that on this release, the cache “can halt the node with an app hash mismatch.” CheckTx and latest-height queries can read the live state tree at the same moment a block is writing to it. A stale cached value can then make the next block compute a different app hash from the rest of the network.
The guide is direct: “Sender accounts and balances are read in CheckTx and written in almost every block, so on a busy node the overlap is likely rather than theoretical.”
The default settings avoid this, because optimistic execution is on by default. An operator who disables optimistic execution and also leaves Block-STM off is running the one combination the guide warns against. Anyone who must run that setup is told to set cache-size = 0 explicitly.
Regenerating config files. The guide also says: “Do not regenerate config.toml or app.toml from scratch.” A fresh app.toml sets mempool.max-txs = -1. If an operator also turns on v1.8’s new app-side mempool (type = "app"), the node won’t start and logs: panic: mempool.type=app is incompatible with NoOpMempool.
Changes Dapp Developers Will Feel Without Touching A Node
Developers who never run a node still need to know about several v1.8 changes.
txfee-cap is now enforced.
- v1.7 ignored
json-rpc.txfee-capforeth_sendRawTransaction, and v1.8 enforces it, as geth does. - The check multiplies
maxFeePerGasby the gas limit. - At mainnet’s 375 gwei minimum gas price, the default cap of 1 CRO allows only about 2.6 million gas.
- According to Cronos, about 5% of recent mainnet transactions would exceed it.
- The guide tells RPC operators to set
txfee-cap = 0, or at least100, before upgrading. - Developers using third-party endpoints may see the error
tx fee (... ether) exceeds the configured capif a provider hasn’t made the change.
Contracts and transactions changing at the upgrade height.
- Transactions with large calldata now have a minimum gas cost.
- Interchain Accounts and their precompile are removed.
SendCroToIbcis limited to authorized contracts.MsgTurnBridgeis no longer supported.
Pending-transaction RPCs. Under the new app-side mempool, eth_newPendingTransactionFilter and CometBFT’s /unconfirmed_txs return nothing. The txpool_* methods still work. Bots and wallets that watch pending transactions should check which mempool their RPC provider runs.
A per-transaction gas cap. A cap of 16,777,216 gas is in the code, but it needs a separate governance change before it applies.
None of these changes depend on Block-STM. The parallel engine produces the same results as sequential execution. When transactions touch the same storage, they are re-executed, which makes them slower but does not change the outcome.
What Builders Should Check Before The Upgrade
The v1.8.0 changelog and the migration guide point to a short checklist:
- Stage the binary under the plan name
v1.8exactly. Cosmovisor won’t recognize any other directory name. - Edit
app.toml; don’t regenerate it. Addblock-executor = "block-stm"to turn on parallel execution. - Leave optimistic execution on. Never combine
sequentialwithdisable-optimistic-execution = trueunlesscache-size = 0. - Raise or remove
txfee-capon any node that serves JSON-RPC. - Choose a mempool mode based on whether your users depend on pending-transaction filters.
So “built into the SDK” describes where the engine’s code lives, not what a node does after the upgrade. For any given node, parallel execution is only on if its operator sets block-executor = "block-stm" in app.toml.
Disclosure: This is not trading or investment advice. Always do your research before buying any cryptocurrency or investing in any services.
Follow us on Twitter @themerklehash to stay updated with the latest Crypto, NFT, AI, Cybersecurity, and Metaverse news!

