Smart contracts are self-executing contracts de ning rules for negotiating, verifying the ful llment of rules and executing the agreement using formal code. They run on top of a blockchain. Errors in smart contracts are costly and are mostly found too late after execution, which is too late for xing. To improve the validation of executed smart contracts, this paper suggests a process mining based approach. For this, we present an approach for the extraction of meaningful event logs from a blockchain. The event log can be imported in any process mining tool and validation and veri cation techniques can be used allowing to diagnose (non)conformity in smart contracts by means of common quality measures and with low latency after smart contract execution.
Blockchain is a decentralized, peer to peer network, which makes use of
cryptography to securely host applications and store data, \where non-trusting members
can interact with each other without a trusted intermediary in a veri able
manner" [
execute conditions contract are met query blockchain contract executed optional query results error analysis by means of verification / validation
Solidity). Subsequently, the smart contract is deployed in the blockchain and becomes part of the blockchain. When all conditions are met, the smart contract executes itself. Veri cation by means of formal methods or validation by means of simulation might take place in di erence phases (i.e., at compile time and runtime). To analyze smart contracts after its execution, the optional activity \query blockchain" (see Figure 1) can be appended to the process referring to querying the blockchain. We motivate the need for smart contract validation by the following example. Figure 2 shows the vehicle manufacture example of a smart contract1 within the smart contract process of Figure 1. This smart contract consists of Participants with the attributes <Person, Regulator, Manufacturer>, Assets with the attributes <Order, Vehicle> and Transactions with the attributes <PlaceOrder, UpdateOrderStatus, SetupDemo>. Assume that someone aims to analyze the smart contract and query the blockchain with placeOrder, updateOrderStatus and setupDemo (see activity \query blockchain" in Figure 2). Further, assume that the value of Vehicle is changed from SCRAPPED to ACTIVE or Order is changed from VEHICLEORDER ASSIGNED to DELIVERED although no prior assignment of a vehicle owner took place. The queries would only return the current value of attributes and could not detect any errors. Although such semantic errors can pass compilation and execution phases, they have to be prevented since they lead to incorrect or unexpected behavior. 2
Related works on smart contract analysis can be distinguished by the time point
of analysis. Approaches related to compile time aim to identify syntax or
typechecking errors, see for reference [
This next section suggests our process mining-based approach. Process mining takes event logs, records of the sequence of steps, and discovers a de-facto model of the processes. Mandatory attributes of an event log are a caseID, activity and time-stamp. Although, transactions of a blockchain are logged, the logs are of no avail for process mining due to their transactional data structure that cannot be directly mapped to an event log. We implemented the extraction of event logs from a blockchain for Hyperledger Fabric and Composer. We omitted to continue working with Ethereum due to several reasons. Currently, Ethereum does not target business smart contracts and thus the interest is low to nd semantic errors. Also, too many empty blocks are mined from Ethereum, which unnecessarily expand and complicate the event log. To extract event log activities our approach relies on the queries queryChain and getProcessLog. The query queryChain requests from a client to register on the blockchain and returns ledger blocks as JSON objects. Note, that a direct extraction of events from the smart contract is not possible since a smart contract requires a prior execution before events can be extracted. The extracted events, thus, refer to \block writes" in the blockchain. Block writes are the part of a block that contains a list of unique IDs and the associated key value pairs that are written by a transaction2. The getProcessLog query returns block writes through iteration and transcribing each block as a trace of an event log. This means that the block writes contain the complete new record, not just the value change of attributes. Assume the attribute color of the object car1=ftype:pickup, color:redg is changed from red to blue, then the new block write is car1=ftype:pickup, color:blueg. To identify value changes, we compare the block writes. For this purpose, the last state before the change (=identical ID) is stored and compared with the current state. To ease the comparison, the possibly nested objects are transformed into a at data structure. If a value is changed or new attributes are added, an event is transcribed. The event name consists of three attributes ranging from 1=abstract to 3=concrete, which might be exemplary eventLevel1=\asset updated", eventLevel2=\color changed" and eventLevel3=\color changed from red to blue". Besides the eventLevel and the eventName attributes, an event2 https://hyperledger-fabric.readthedocs.io/en/release-1.3/readwrite.html De nition is de ned, which aggregates simple events to complex events. Assume an eventLevel=\eventLevel3", eventName=\This is now a red pickup" and eventDe nition=[color=. >red] && [make=. > pickup]. The evaluation of the regular expression to true, is considered a complex event. The process model extracted from an exemplary smart contract is shown in Figure 3. The source code of our project can be downloaded from https://github.com/FrankDuchmann/ hf-event-extraction.
This paper presented an approach for event log extraction from Hyperledger
Fabric and Composer. The event logs can be processed by any process mining
tool. We consider an event as the result of a transaction, which goes beyond the
simpli ed solution of [