Distributed transaction log

A system comprising:

• one or more computers and one or more storage devices on which are stored data that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• receiving an input/output (I/O) operation request from a virtual machine at a hypervisor, the I/O operation request for a corresponding virtual disk of the virtual machine, wherein the virtual disk is associated with a distributed storage system;
• determining particular resource component objects of the virtual disk that are subject to the I/O operation, wherein each resource component object is associated with particular physical storage resources of the distributed storage system;
• performing a plurality of distributed transactions on the particular resource component objects to carry out the I/O operation;
• logging each transaction in a journal for each respective resource component object, wherein each entry includes a sequence identifier;
• and
• reconciling the entries on different journals such that lost transactions from an offline resource component object are derived from comparing the content of the journals using the respective sequence identifiers.

A non-transitory computer-readable storage medium storing instructions, which, when executed on a processor, performs operations comprising:

• receiving an input/output (I/O) operation request from a virtual machine at a hypervisor, the I/O operation request for a corresponding virtual disk of the virtual machine, wherein the virtual disk is associated with a distributed storage system;
• determining particular resource component objects of the virtual disk that are subject to the I/O operation, wherein each resource component object is associated with particular physical storage resources of the distributed storage system;
• performing a plurality of distributed transactions on the particular resource component objects to carry out the I/O operation;
• logging each transaction in a journal for each respective resource component object, wherein each entry includes a sequence identifier;
• and
• reconciling the entries on different journals such that lost transactions from an offline resource component object are derived from comparing the content of the journals using the respective sequence identifiers.

A system comprising:

• one or more computers and one or more storage devices on which are stored data that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• preventing previously offline resource component objects in a distributed system from servicing operations if the resource component objects are stale, the system performing transactions without the resource component objects, the resource component objects to be resynchronized from other online resource component objects, comprising:
• when a resource component object goes offline and the system makes progress without it, persisting a record of the fact that it is offline on a majority of all online resource component objects, said majority being a majority of the total number of resource components;
• maintaining information of which resource component objects are stale in the running system and using this information in deciding which resource component objects must participate in transactions;
• upon regaining connectivity to the previously offline resource component objects, disallowing the stale resource component object from servicing operations until a resynchronization method is complete, wherein the resynchronization comprises obtaining transaction logs from one or more other online resource component objects, each transaction log including entries identifying actions performed on each of the one or more online resource component objects for each distributed transaction of the distributed system, and updating the resource component object based on the transaction logs;
• and
• upon completion of the resynchronization, persisting on a majority of nodes the fact that the resource component object is no longer stale before allowing it to participate in transactions.