Fault recovery support apparatus, fault recovery support method and program
A fault recovery support apparatus according to an embodiment includes: a fault insertion unit configured to insert a fault into a target system; a behavior execution unit configured to execute a behavior related to recovery from the fault for the target system; a first construction unit configured to construct an automaton representing a recovery process from the fault by using observation data acquired from the target system as a result of the behavior by the behavior execution unit; and a second construction unit configured to construct a workflow representing a behavior for separating each fault included in a plurality of faults and a recovery process of each fault by using a plurality of the automatons related to the plurality of faults.
1. A fault recovery support apparatus comprising:
a memory; and
a processor configured to execute:
inserting, by a fault inserter, a plurality of faults into a target system;
executing, by a behavior executer, a behavior related to recovery from each of the plurality of the faults for the target system;
constructing, by a first constructor, a plurality of automatons, each representing a recovery process from each of the plurality of the faults, by clustering observation data acquired from the target system as a result of the behavior by the behavior executer; and
constructing, by a second constructor, a workflow representing a behavior for separating each fault included in a plurality of faults and a recovery process of each fault by using a plurality of the automatons related to the plurality of faults,
wherein the constructing by the first constructor includes
clustering the observation data to allocate discrete numbers to a third behavior sequence formed by combining a first behavior sequence for the target system before the observation data is obtained and a second behavior sequence for the target system when the observation data is obtained, and
constructing an automaton based on an observation table in which values of elements having the first behavior sequence as a row index and the second behavior sequence as a column index are set as the discrete numbers by an L* algorithm,
wherein the constructing by the second constructor includes constructing the workflow in which a type of fault and a state that is able to be a state of the target system are indicated by observation data acquired when a behavior is executed on the target system in a form of a directed graph by using a plurality of the automatons of which initial states are not distinguishable,
wherein the workflow is a directed graph comprising a plurality of nodes and a plurality of directed edges,
wherein each node is defined as a tuple of a label of the observation data and a vector indicating current states of the plurality of automatons corresponding to the plurality of faults;
wherein each directed edge from a first node to a second node represents a behavior executable on the target system and the label of the observation data obtained as a result of executing the behavior based on a current state,
wherein the executing, by the behavior executer, includes transmitting, to the target system, a command corresponding to the behavior represented by the each directed edge in the workflow, and acquire observation data from the target system in response to the command execution,
wherein the executing, by the behavior executer, includes:
assigning the label of the observation data to the observation data using a result of the clustering,
determining a transition to a next node in the workflow based on the assigned label of the observation data, and updating the vector indicating the current states of the plurality of automatons corresponding to the plurality of faults, and
executing, based on the determined transition and the updated vector, a recovery control for the target system so as to recover the target system from a fault state toward a normal state.
2. The fault recovery support apparatus according to claim 1 , wherein
the constructing by the first constructor includes
giving the discrete number to the third behavior sequence by a classifier that has been learned in advance.
3. A fault recovery support method of causing a computer to perform:
inserting a plurality of faults into a target system;
executing a behavior related to recovery from each of the plurality of the faults for the target system;
constructing a plurality of automatons, each representing a recovery process from each of the plurality of the faults, by clustering observation data acquired from the target system as a result of the behavior in the behavior executer; and
constructing a workflow representing a behavior for separating each fault included in a plurality of faults and a recovery process of each fault by using a plurality of the automatons related to the plurality of faults,
wherein the constructing the plurality of automatons includes
clustering the observation data to allocate discrete numbers to a third behavior sequence formed by combining a first behavior sequence for the target system before the observation data is obtained and a second behavior sequence for the target system when the observation data is obtained, and
constructing an automaton based on an observation table in which values of elements having the first behavior sequence as a row index and the second behavior sequence as a column index are set as the discrete numbers by an L* algorithm,
wherein the constructing the workflow includes constructing the workflow in which a type of fault and a state that is able to be a state of the target system are indicated by observation data acquired when a behavior is executed on the target system in a form of a directed graph by using a plurality of the automatons of which initial states are not distinguishable,
wherein the workflow is a directed graph comprising a plurality of nodes and a plurality of directed edges,
wherein each node is defined as a tuple of a label of the observation data and a vector indicating current states of the plurality of automatons corresponding to the plurality of faults;
wherein each directed edge from a first node to a second node represents a behavior executable on the target system and an observation label obtained as a result of executing the behavior based on a current state, and
wherein the executing includes transmitting, to the target system, a command corresponding to the behavior represented by the each directed edge in the workflow, and acquiring observation data from the target system in response to the command execution,
wherein the executing, by the behavior executer, includes:
assigning the label of the observation data to the observation data using a result of the clustering,
determining a transition to a next node in the workflow based on the assigned label of the observation data, and updating the vector indicating the current states of the plurality of automatons corresponding to the plurality of faults, and
executing, based on the determined transition and the updated vector, a recovery control for the target system so as to recover the target system from a fault state toward a normal state.
4. A non-transitory computer-readable recording medium having computer-readable instructions stored thereon, which when executed, cause a computer to function as the fault recovery support apparatus according to claim 1 .