Method and apparatus for ensuring data integrity in a storage cluster with the use of NVDIMM
An information handling system includes a persistent storage and a memory controller. The persistent storage includes a volatile memory and a non-volatile memory. The memory controller stores data and metadata for a data file within the volatile memory, and the data file is synchronized within other information handling systems of a storage cluster. The memory controller updates the metadata in response to a change in the data of the data file, stores the data and the metadata for the data file within the non-volatile memory prior to a power loss of the information handling system, and synchronizes the data and the metadata of the data file with current data and current metadata for the data file found in the other information handling systems in response to the information handling system being back online. The data is synchronized with the current metadata based on a transform for the data file being received from the other information handling systems.
1. An information handling system comprising:
a persistent storage including a volatile memory and a non-volatile memory; and
a memory controller in communication with the persistent storage, the memory controller to store data and metadata for a data file within the volatile memory, wherein the data file is synchronized with corresponding data files stored within other information handling systems of a storage cluster, to update the metadata in response to a change in the data of the data file, to store the data and the metadata for the data file within the non-volatile memory prior to a power loss of the information handling system, and in response to the information handling system being back online, the memory controller to:
transfer the data and the metadata for the data file from the non-volatile memory to the volatile memory;
send the metadata including a latest timestamp for the data file to the other information handling systems of the storage cluster;
receive a first transform calculated for changes to the data of the data file at a first timestamp, wherein the first timestamp is later than the latest timestamp;
receive a second transform calculated for changes to the data of the data file at a first second timestamp, wherein the second timestamp is later than the first timestamp;
determine, during a first pass, whether first metadata for the data file at the first timestamp, is the same as the metadata for the data file currently stored in the volatile memory, wherein no update transform is performed based on the first metadata being the same as the metadata;
perform, during a second pass, a transform on the data file to update the data file to correct data at the second timestamp;
store the updated data file in the volatile memory; and
update the metadata for the data file to add a first timestamp entry with the first timestamp and with the changes to the data file at the first timestamp, and to add a second timestamp entry with the second timestamp and with the changes to the data file at the second timestamp.
2. The information handling system of claim 1 , further comprising:
a processor to communicate with the other information handling systems within the storage cluster, and to provide a heartbeat to the other information handling systems.
3. The information handling system of claim 2 , wherein the sending of the metadata including the latest timestamp for the data file to the other information handing systems of the storage cluster includes, the processor to broadcast a state of the data and metadata for the data file to the other information handling systems in the storage cluster in response to the information handling system being back online within the storage cluster.
4. The information handling system of claim 3 , wherein the processor to receive the first and second transforms for the data file from the other information handling systems in response to the broadcast.
5. The information handling system of claim 1 , wherein the memory controller to further store a current timestamp and a change of the data of the data file in response to the data file being updated.
6. The information handling system of claim 5 , wherein the change of the data for the current timestamp is based on the data of a previous timestamp.
7. A method comprising:
detecting that a data file is updated on a first of a plurality of nodes of a storage cluster;
updating data and metadata for the data file on each node of the storage cluster;
detecting a failure of the first node of the storage cluster;
storing the data and the metadata for the data file in a non-volatile memory of the first node before the first node is offline;
updating the data and the metadata for the data file on another node remaining in the storage cluster in response to a change of the data file after the first node is offline;
detecting that the first node is back online;
in response to the first node being back online:
transferring the data and the metadata for the data file from the non-volatile memory to a volatile memory;
sending the metadata including a latest timestamp for the data file to the other nodes of the storage cluster;
receiving a first transform calculated for changes to the data of the data file at a first timestamp, wherein the first timestamp is later than the latest timestamp;
receiving a second transform calculated for changes to the data of the data file at a second timestamp, wherein the second timestamp is later than the first timestamp;
determining, during a first pass, whether first metadata for the data file at the first timestamp is the same as the metadata for the data file currently stored in the volatile memory, wherein no update transform is performed based on the first metadata being the same as the metadata,
performing, during a second pass, a transform on the data file to update the data file to correct data at the second timestamp;
storing the updated data file in the volatile memory; and
updating the metadata for the data file to add a first timestamp entry with the first timestamp and with the changes to the data file at the first timestamp, and to add a second timestamp entry with the second timestamp and with the changes to the data file at the second timestamp.
8. The method of claim 7 , wherein the data and the metadata are utilized by each node of the storage cluster to verify data integrity within the storage cluster.
9. The method of claim 8 , wherein the sending of the metadata including the latest timestamp for the data file to the other nodes of the storage cluster further comprises:
broadcasting a state of the data and metadata for the data file within the first node to the other nodes in the storage cluster in response to the first node being back online within the storage cluster.
10. The method of claim 8 , further comprising:
determining that the data and metadata for the data file within the first node is stale based on the state broadcasted; and
providing the first and second transforms for the data file to the first node in response to determining that the data and the metadata are stale.
11. The method of claim 7 , further comprising:
storing a current timestamp and a change of the data in the volatile memory in response to the data file being updated.
12. The method of claim 11 , wherein the change of the data for the current timestamp is based on the data of a previous timestamp.
13. A storage cluster comprising:
a first node including:
a first persistent storage including a first volatile memory and a first non-volatile memory; and
a first memory controller in communication with the first persistent storage, the first memory controller to store data and metadata for a data file within the first volatile memory, to update the metadata in response to a change in the data of the data file; and
a second node including:
a second persistent storage including a second volatile memory and a second non-volatile memory; and
a second memory controller in communication with the second persistent storage, the second memory controller to:
store data and metadata for the data file within the second volatile memory, wherein the data and metadata for the data file within the second volatile memory are synchronized within the data and metadata for the data file within the first volatile memory;
update the metadata in response to a change in the data of the data file;
store the data and the metadata for the data file within the second non-volatile memory prior to a power loss of the information handling system; and
in response to the second node being back online, the second memory controller to:
transfer the data and the metadata for the data file from the second non-volatile memory to the second volatile memory;
send the metadata including a latest timestamp for the data file to the first node;
receive a first transform calculated for changes to the data of the data file at a first timestamp, wherein the first timestamp is later than the latest timestamp;
receive a second transform calculated for changes to the data of the data file at a second timestamp, wherein the second timestamp is later than the first timestamp;
determine, during a first pass, whether first metadata for the data file at the first timestamp is the same as the metadata for the data file currently stored in the second volatile memory, wherein no update transform is performed based on the first metadata being the same as the metadata;
perform, during a second pass, a transform on the data file to update the data file to correct data at the second timestamp;
store the updated data file in the volatile memory; and
update the metadata for the data file to add a first timestamp entry with the first timestamp and with the changes to the data file at the first timestamp, and to add a second timestamp entry with the second timestamp and with the changes to the data file at the second timestamp.
14. The storage cluster of claim 13 , wherein the first node further comprises a first processor to communicate with the second node, and to provide a heartbeat to the second node, and wherein the second node further comprises a second processor to communicate with the first node, and to provide a heartbeat to the first node.
15. The storage cluster of claim 14 , wherein the sending of the metadata including the latest timestamp for the data file to the first node includes, the second processor to broadcast a state of the data and metadata for the data file to the first node in response to the second node being back online within the storage cluster.
16. The storage cluster of claim 15 , wherein the second processor to receive the first and second transforms for the data file from the first node in response to the broadcast.
17. The storage cluster of claim 13 , wherein the second memory controller to further store a current timestamp and a change of the data of the data file in response to the data file being updated.
18. The storage cluster of claim 17 , wherein the change of the data for the current timestamp is based on the data of a previous timestamp.