Comment on page
Round-robin Operator Manager
Round-robin relayer selection
Non-production feature
The Round-Robin Operator Manager may be implemented as a future use case. It has several disadvantages including high gas consumption for relayers (even when idle) and requires additional improvements.
An operator and relayer are equivalent entities in the contracts layer.
The current zkBob implementation assumes that only a single relayer can interact with the pool contract at any given time. An approach is needed to select a single relayer for each time interval within a multiple-relayers configuration.
The simplest way to do this is to select a relayer from a list in a round-robin manner. We implement this method in the RoundRobin contract.
The relayer list is managed by the contract owner which can add or remove the relayer at any time. The contract owner should transfer the operator manager contract to the maintenance state to change the list of approved relayers. In this state the main functionality is disabled and all regular methods are reverted.
Only strictly approved relayers are accepted to avoid malfunction operations. The relayer should be stable and available through public networks and it should have a quality communication link. The relayer should have a maintainer who is available to fix operating issues.
TODO: Develop / describe relayer approval procedure
The operator manager contract lifetime is divided by time slots. Each slot contains a fixed amount of underlying blockchain blocks. It starts with the Genesis slot (the first 1000 blocks after contract deployment). The Genesis slot's purpose is to configure the contract by the owner and prepare relayers for operation. In that block the owner should create an operator list with at least one entry.
Each relayer should periodically inform the operator manager contract that it's ready to operate with the Pool. To do that a relayer should invoke the
claim()contract method. The claim received in the current slot is intended for the next one. If the relayer does not provide a claim for the specified block it will be ignored in the round-robin selection and the next available one will be used. This approach is intended to decrease the probability of unavailable relayers.If there are no claims for the current slot, the last operator that sent a claim will be selected. So it's not necessary for operator to send claims for the each slot in the single relayer configuration.
The current operator index
(within an operator list with the length
) for the each slot with index
is calculated in round-robin manner:
The Genesis slot has the index 0 so the operator selection during that slot is unavailable. The resulting index
can be changed in the interval
inclusively.
Relayer selection scheme
In the picture above you can explore the selection mechanism. After contract deployment the owner adds three approved relayers (named RL-1, RL-2 and RL-3). The first and the second relayer claimed for the slot 1 during Genesis slot, so the first one will be selected as an active operator.
The third relayer becomes available only in slot 2, so his first claim is intended for slot 3. Fortunately the third operator should be selected in slot 3 exactly (according to the round-robin scheme). So it's become an active relayer in slot 3.
During slot 3 the first relayer (RL-1) has failed, and the operator manager contract didn't receive a claim from it. That's why in slot 4 the RL-1 has been ignored and the next claimed operator (RL-2) has been selected.
In the 4 the third relayer (RL-3) has also failed, so only the single claim has been received from RL-2. In slot 5 RL-2 should be selected according to the round-robin scheme.
Let's imagine that RL-2 has decided to stop claim sending after slot 4. This led to an absence any claims for slot 6. In that case the last claimed operator (RL-2) will be selected for the next slot.
Anyone can invoke the following contract methods to retrieve the current relayer properties in the normal state:
operator()will return the current operator blockchain address. In the maintenance state the method call will be reverted. If there are no selected relayers for the current slot, this method returnsaddress(0)[To be discussed: return 0 or revert?]operatorURI()will return the current operator's endpoint URL (string).operatorName()will return the current operator name (string).amIOperator()will return a boolean value indicating that a message sender is an active operator at that time.
If the contract owner intends to make some changes the contracts should be changed to a maintenance state. In this state all regular method calls are reverted and the operator manager accepts the service calls only. After maintenance is complete, the contract should be returned in the regular state to proceed a normal flow. Changing the contract's state is performed by the method
setMaintenance(bool)In the maintenance state the following service methods are available:
addOperator(stringname,addressaddr,stringendpoint)will add a new operator with the specified name, blockchain address and URL endpoint.removeOperator(addressaddr)will remove the operator with the specified address. Please note that the last operator in the list will be relocated to the vacant place.setSlotSize(uint32blocks)will change slot size (specified in blocks). The number of blocks should be greater than 0.
Last modified 1yr ago