Multi-node replication systems, devices and methods
View Patent ↗Replication techniques are presented. According to an embodiment of a method, a node of a replicated storage network is assigned to be an owner of a data block to issue write memory block commands. The network includes at least two nodes including the node assigned to be the owner. If a read memory block command is received to read the data block, a read_lock is issued for the data block, the data block is read, and the read_lock for the data block is released. If a write memory block command is received to write new data to the data block, a write_lock is issued for the data block, the data block is written and a version associated with the data block is incremented, and the write_lock for the data block is released.
1. A method of replicating data in a storage network having at least two nodes, comprising:
assigning nodes of the network to be owners of data blocks to issue write memory block commands, each data block being owned by one node;
performing a local write of a first data block to an owner node for the first data block, including performing a first process at the owner node and performing a second process at the non-owner nodes in the network:
wherein the first process at the owner node includes:
issuing a write_lock request for the first data block to all nodes in the network;
copying old data for the first data block to a memory location, and issuing a write command to non-owner nodes in the network to write new data for the first data block;
waiting for the acknowledgment from each non-owner node indicating that the non-owner node has written the new data for the first data block;
writing the new data for the first data block to the owner node, and incrementing a version associated with the first data block;
issuing a write_unlock request for the first data block to all nodes on the network; and
performing a write_unlock for the first data block on the owner node; and
wherein the second process at non-owner nodes includes:
receiving the write_lock request for the first data block from the owner node for the first data block;
performing a write_lock for the first data block, and copying old data for the first data block into memory;
writing the new data for the first data block, and incrementing a version number associated with the first data block;
sending the acknowledgment to the owner node; and
performing a write_unlock for the first data block.
2. The method of claim 1 , wherein the first process at the owner node further comprises:
determining if a network error occurred prior to writing the new data for the first data block to the owner node;
writing the new data for the first data block to the owner node if a network error did not occur; and
if a network error occurred, performing corrective action.
3. The method of claim 2 , wherein, if a network error occurred prior to writing the new data for the first data block to the owner node, the first process at the owner node further comprises:
determining if the owner node is active or passive after performing corrective action;
aborting the local write of the first data block to the owner node if the owner node is passive after performing corrective action; and
writing the new data for the first data block to the owner node if the owner node is active after performing the corrective action.
4. The method of claim 3 , further comprising making the owner node passive after a network error if the owner node is part of a cluster of nodes that remain in communication with each other that is less than half of the nodes of a replication network before the network error.
5. The method of claim 3 , wherein determining if the owner node is active or passive includes an administrator determining if the owner node is active or passive.
6. The method of claim 1 , wherein the first process at the owner node further comprises:
determining if a network error occurred after writing the new data for the first data block to the owner node;
performing the write_unlock for the first data block on the owner node if a network error did not occur; and
if a network error occurred, performing corrective action.
7. The method of claim 6 , wherein, if a network error occurred after writing the new data for the first data block to the owner node, the first process at the owner node further comprises determining if the owner node is active or passive after performing corrective action;
if the owner node is passive after performing corrective action, writing saved old data back to the first data block and decrementing the version number associated with the first data block; and
if the owner node is active after performing the corrective action, performing the write_unlock for the first data block on the owner node.
8. The method of claim 1 , wherein the second process at the non-owner nodes further comprises:
determining if a network error occurred after writing the new data for the first data block; and
if a network error occurred, performing corrective action.
9. The method of claim 8 , wherein, if a network error occurred after writing the new data for the first data block, the second process at the non-owner nodes further comprises:
determining if the owner node is active or passive after performing corrective action; and
if the owner node is passive after performing corrective action, writing saved old data back to the first data block and decrementing the version number associated with the first data block.
10. The method of claim 1 , further comprising performing a third process to locally write a second data block to one of the non-owner nodes of the second data block, including:
issuing a write_lock for the second data block and copying old data for the second data block into local memory;
sending new data for the second data block to the owner node of the second data block;
if write request is received from the owner node for the second data block, writing the new data for the second data block, and incrementing a version number associated with the second data block, and performing a write_unlock for the second data block.
11. The method of claim 10 , wherein the third process further comprises determining if the owner node sent competing data to write the second data block before the write request is received from the owner node, and writing the competing data to the second data block if the owner node sent competing data before the write request is received from the owner node.
12. The method of claim 10 , wherein the third process further comprises:
determining if a network error occurred after writing the new data for the second data block; and
if a network error occurred, performing corrective action.
13. The method of claim 12 , wherein, if a network error occurred after writing the new data for the second data block, the third process further comprises:
determining if the one of the non-owner nodes is active or passive after performing corrective action;
if the one of the non-owner nodes is passive after performing corrective action, writing saved old data back to the first data block and decrementing the version number associated with the first data block; and
if the one of the non-owner nodes is active after performing corrective action, resending the second data block to a new owner of the second data block.
14. The method of claim 12 , wherein, if a network error occurred after writing the new data for the second data block, the third process further comprises performing a write unlock if a network failure did not occur.
15. A multi-node replicated storage network, comprising:
at least two network nodes, each network node including:
a memory storage device, including replicated storage having a plurality of data blocks stored therein and a version associated with each of the data blocks;
a network link interface; and
a replication manager to communicate with other network nodes using the network link interface, and the replication manager to communicate with the memory storage device;
wherein the replication managers for the at least two network nodes are adapted to:
determine node ownership for data blocks to issue write memory block commands, each data block being owned by one node; and
write a first data block using a first process at an owner node for the first data block and using a second process at non-owner nodes,
wherein the first process at the owner node for the first data block includes:
issuing a write_lock request for the first data block to all nodes in the network;
copying old data for the first data block to memory, and issuing a write command to non-owner nodes in the network to write new data for the first data block;
waiting for the acknowledgment from each non-owner node indicating that the non-owner node has written the new data for the first data block;
writing the new data for the first data block to the owner node, and incrementing a version associated with the first data block;
issuing a write_unlock request for the first data block to all nodes on the network; and
performing a write_unlock for the first data block on the owner node.
16. The network of claim 15 , wherein the second process at non-owner nodes for the first data block includes:
receiving the write_lock request for the first data block from the owner node for the first data block;
performing a write_lock for the first data block, and copying old data for the first data block into memory;
writing the new data for the first data block, and incrementing a version number associated with the first data block;
sending an acknowledgment to the owner node; and
performing a write_unlock for the first data block.
17. The network of claim 16 , wherein:
the first process at the owner node includes:
determining if a network error occurred prior to writing the new data for the first data block to the owner node;
writing the new data for the first data block to the owner node if a network error did not occur; and
if a network error occurred, performing corrective action; and
the second process at the non-owner nodes includes:
determining if a network error occurred after writing the new data for the first data block; and
if a network error occurred, performing corrective action.
18. A multi-node replicated storage network, comprising:
at least two network nodes, each network node including:
a memory storage device, including replicated storage having a plurality of data blocks stored therein and a version associated with each of the data blocks;
a network link interface; and
a replication manager to communicate with other network nodes using the network link interface, and the replication manager to communicate with the memory storage device;
wherein the replication managers for the at least two network nodes are adapted to:
determine node ownership for data blocks to issue write memory block commands, each data block being owned by one node; and
write a first data block using a first process at an owner node for the first data block and using a second process at non-owner nodes,
wherein the replication managers for the at least two network nodes are adapted to locally write a second data block to non-owner nodes of the second data block using a third process, including:
issuing a write_lock for the second data block and copying old data for the second data block into local memory;
sending new data for the second data block to the owner node of the second data block; and
if write request is received from the owner node for the second data block, writing the new data for the second data block, and incrementing a version number associated with the second data block, and performing a write_unlock for the second data block.