Ensuring determinism during programmatic replay in a virtual machine

A non-transitory computer-readable storage medium storing instructions which, when executed by a processor, cause the processor to:

• intercept, by a replay core during a record time, an event that indicates that a particular class is being loaded by code that is executed by a virtual machine that executes an application for which the replay core records instructions during the record time;
• modify the particular class in response to the replay core intercepting the event, comprising instructions that cause the processor to inject bytecodes into the particular class thereby producing a modified particular class, wherein execution of the injected bytecodes by the processor causes values produced by objects instantiated from the particular class in a replay time of the application executed by the virtual machine to be the same as the values produced by objects instantiated from the particular class during the record time, wherein modifying the particular class the instructions causes the processor to walk a class hierarchy to determine whether the particular class inherits from one or more other classes thereby causing other classes in the class hierarchy to be loaded;
• store the modified particular class in memory before the replay time;
• listen, by the replay core, for invocations of original methods associated with the one or more other classes that load as a result of walking the class hierarchy during the record time;
• replace the invocations of the original methods with invocations of proxy methods that cause data to be stored in a replay stream in memory that (a) identify class loaders that load the other classes, and (b) identify the other classes, wherein the replay stream identifies an order in which the other classes were loaded during the record time;
• invoke, by the proxy methods during the record time, each of the original methods and passing to the original methods the data that identifies the other class for the respective original method;
• intercept, by the replay core during the replay time, an event that indicates that the particular class is being loaded by code that is executed by the virtual machine;
• in response to intercepting the event during the replay time, access the modified particular class from the memory instead of modifying the particular class by injecting bytecodes into the particular class as was done during the record time, therefore the one or more other classes from which the particular class inherits are not caused to be loaded as a result of walking the class hierarchy to determine whether the particular class inherits from one or more other classes;
• in response to intercepting the event during the replay time, access the replay stream that identifies the order in which the other classes were loaded during the record time and, in accordance with the order in which the other classes were loaded during the record time, for each of the other classes:
• i) invoke a proxy method instead of the original method for the other class;
• ii) pass the data to the proxy method that (a) identifies the class loader that loads the other class, and (b) identifies the other class;
• iii) invoke, by the proxy method, the original method and passing to the original method the data that identifies the other class;
• and
• cause the bytecode from the modified particular class to be executed by the virtual machine at the replay time instead of the particular class to ensure deterministic replay.

A computer-implemented method implemented with a replay core equipped with a processor and one or more memory devices having executable instructions stored thereon, which when executed by the processor, perform the method steps, comprising:

• recording, by a replay core, instructions of a program that is executed by a virtual machine during a record time;
• modifying, by the replay core, bytecode of a particular class in response to the particular class being loaded by code that is executed by the virtual machine during the during the record time, comprising the replay core injecting bytecodes into the particular class thereby producing a modified class, the injected bytecodes for ensuring that values produced by instances of the particular class during the record time will also be produced by instances of the particular class during a replay time in which execution of the program by the virtual machine is replayed, comprising the replay core walking a class hierarchy to determine whether the particular class inherits from another class thereby causing other classes in the class hierarchy to be loaded, wherein execution of objects instantiated from the particular class is potentially non-deterministic;
• storing in a persistent storage device, prior to the replay time, the modified class that comprises the modified bytecode;
• storing a replay stream in the persistent storage device, the replay stream comprising information regarding one or more other classes from which the particular class inherits that load during the record time as a result of the replay core walking the class hierarchy, comprising storing:
• 1) identifiers of class loaders used to load the one or more other classes, and 2) an order in which the one or more other classes were loaded during the record time in response to walking the class hierarchy;
• during the replay time, in response to determining that the particular class is being loaded by code that is executed by the virtual machine, the replay core:
• i) accessing the stored modified class from the persistent storage instead of modifying the bytecode of the particular class and without walking the class hierarchy to determine whether the particular class inherits from another class as was done at record time, therefore the one or more other classes from which the particular class inherits are not caused to be loaded during the replay time as a result of walking the class hierarchy;
• and
• ii) based on the information in the replay stream in the persistent storage device regarding the one or more other classes from which the particular class inherits, simulating invocations of methods that were invoked during the record time by the identified class loaders in the order in which the one or more other classes were loaded during the record time, wherein simulating the invocations of the methods causes operations that were performed by the methods during the record time to be performed in the same order during the replay time;
• and
• causing the bytecode from the modified class to be executed at the replay time instead of the particular class to ensure deterministic replay.

A non-transitory computer-readable storage medium storing instructions which, when executed by a processor, cause the processor to:

• record, by a replay core, instructions of a program that is executed by a virtual machine during a record time;
• modify bytecode of a particular class of the program in response to the particular class being loaded during the record time by code that is executed by the virtual machine, the instructions that cause the processor to modify the bytecode cause the processor to inject bytecodes into the particular class thereby producing a modified class, execution of the injected bytecodes by the processor for ensuring that values produced by instances of the particular class during the record time will also be produced by instances of the particular class during a replay time in which execution of the program by the virtual machine is replayed, wherein the instructions to modify the bytecode cause comprising causing the processor to walk a class hierarchy to determine whether the particular class inherits from another class, wherein execution of objects instantiated from the particular class is potentially non-deterministic;
• store a replay stream in a persistent storage device, the replay stream comprising information regarding one or more other classes from which the particular class inherits that load during the record time as a result of the replay core walking the class hierarchy, comprising causing the processor to store:
• 1) identifiers of class loaders used to load the one or more other classes, and 2) an order in which the one or more other classes were loaded during the record time in response to walking the class hierarchy;
• store in the persistent storage device, prior to the replay time, a modified class that comprises the modified bytecode;
• during the replay time, in response to the instructions causing the processor to determine that the particular class is being loaded by code that is executed by the virtual machine:
• i) access the stored modified class from the persistent storage instead of modifying the bytecode of the particular class and without walking the class hierarchy to determine whether the particular class inherits from another class as was done at the record time, therefore the one or more other classes from which the particular class inherits are not caused to be loaded during the replay time as a result of walking the class hierarchy;
• and
• ii) access the information from the persistent storage device regarding the one or more other classes from which the particular class inherits that load during the record time as a result of the replay core walking the class hierarchy;
• and
• iii) based on the information in the replay stream regarding the one or more other classes, simulate invocations of methods that were invoked during the record time by the identified class loaders in the order in which the one or more other classes were loaded during the record time, wherein causing the processor to simulate the invocations of the methods causes operations that were performed by the methods during the record time to be performed in the same order during the replay time;
• and
• cause the bytecode from the modified class to be executed at the replay time instead of the particular class to ensure deterministic replay.