Using log objects in object storage for durability of file objects in volatile memory

• US 10,430,103 B1
• Filed: 03/20/2017
• Issued: 10/01/2019
• Est. Priority Date: 03/20/2017
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Jamail, John, Mesard, Wayne, Wharton, Ian Marshal
• Primary Examiner(s): Dudek, Jr., Edward J
• Assistant Examiner(s): Li, Sidney
• Application Number:
  US15/464,305
• Time in Patent Office: 925 Days
• Field of Search: 711165
• CPC Class Codes: G06F 3/061 Improving I/O performance
• CPC Class Codes: G06F 3/0619 in relation to data integrity, e.g. data losses, bit errors
• CPC Class Codes: G06F 3/0643 Management of files
• CPC Class Codes: G06F 3/0652 Erasing, e.g. deleting, data cleaning, moving of data to a wastebasket
• CPC Class Codes: G06F 3/0656 Data buffering arrangements
• CPC Class Codes: G06F 3/0667 at data level, e.g. file, record or object virtualisation
• CPC Class Codes: G06F 3/067 Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS]

Abstract

A file system interface receives, from one or more clients, file system operations over a network. In response to receiving a file system operation for a file, a file-object manager calls an object storage interface to store, via an application-level network protocol, a log object indicating the file system operation into a file system log in the object storage system and creates or updates a file object in a file object construction buffer in volatile system memory. To evict the file object, the file-object manager calls the object storage interface to store the file object into a file object store in the object storage system. The file-object manager does not have access to local persistent storage. After power failure, the file-object manager calls the object storage interface to retrieve log objects from the file system log and replays file system operations to restore file objects in the volatile system memory.

Claims

1. A system, comprising:

an object storage system configured to store data objects via an application-level network protocol; and

a file gateway implemented by one or more computing devices comprising one or more processors and volatile system memory, wherein the file gateway comprises;

a file system interface configured to receive, from one or more clients over a network, file system operations;

an object storage interface configured to store data objects to the object storage system via the application-level network protocol; and

a file-object manager configured to;

in response to receiving by the file system interface a file system operation for a file;

call the object storage interface to store a log object into a file system log in the object storage system, wherein the log object indicates the file system operation for the file; and

create or update, based on the file system operation for the file, a file object in a file object construction buffer in the volatile system memory, wherein the file object corresponds to the file, and wherein the file-object manager does not have access to local persistent storage; and

in response to a determination to evict the file object, call the object storage interface to store the file object into a file object store in the object storage system.

2. The system as recited in claim 1, wherein the file-object manager is configured to:

in response to a reboot or power failure that results in loss of content stored in the volatile system memory;

call the object storage interface to retrieve one or more log objects from the file system log of the object storage system; and

replay one or more file system operations indicated by the one or more log objects to restore one or more file objects in the volatile system memory.

3. The system as recited in claim 1, wherein, for a file system write received from a client, the file-object manager is configured to:

upon creating or updating a corresponding file object in the file object construction buffer and receiving an acknowledgment from the object storage interface that a corresponding log object is stored in the file system log of the object storage system, instruct the file system interface to acknowledge to the client that the file system write is complete.

4. The system as recited in claim 1, wherein to determine to evict the file object, the file-object manager is configured to determine that one or more eviction criteria for the particular file object are met, and wherein the one or more eviction criteria comprises at least one of:

available storage space for objects in the file object construction buffer falling below a threshold value, or

the file object being the least recently used file object or being in a group of least recently used file objects in the file object construction buffer, or

receiving an indication to commit the file.

5. The system as recited in claim 1, wherein the file system interface is configured to:

provide a network file system mount point for the one or more clients to access over the network, wherein file objects that correspond to respective files for the network file system exist as sequential data structures in a continuous address range of the volatile system memory.

6. A method, comprising:

performing, by one or more computing devices;

receiving, via a file system interface, file system operations from one or more clients over a network; and

in response to receiving, via the file system interface, a file system operation for a file;

calling, by a file-object manager, an object storage interface to store a log object into a file system log in an object storage system, wherein the log object indicates the file system operation for the file; and

creating or updating, based on the file system operation for the file, a file object in a file object construction buffer in a volatile system memory, wherein the file object corresponds to the file.

7. The method as recited in claim 6, wherein the file-object manager does not have access to local persistent storage and the object storage interface stores the log object to the file system log in the object storage system via an application-level network protocol.

8. The method as recited in claim 6, further comprising:

in response to a reboot or power failure that results in loss of content stored in the volatile system memory;

calling, by a file-object manager, the object storage interface to retrieve, via an application-level network protocol, log objects from the file system log of the object storage system; and

replaying, by a file-object manager, file system operations indicated by the log objects in a same sequential order that the file system operations were previously applied to one or more file objects in the volatile system memory to restore the one or more file objects in the volatile system memory.

9. The method as recited in claim 6, further comprising:

for a file system read received from a client;

determining, by the file-object manager, whether a particular file object corresponding to the file system read exists in the file object construction buffer in the volatile system memory; and

in response to determining, by the file-object manager, that the particular file object corresponding to the file system does not exist in the file object construction buffer in the volatile system memory;

calling the object storage interface to retrieve, via an application-level network protocol, the particular file object from the file object store in the object storage system; and

loading the particular file object into the file object construction buffer in the volatile system memory.

10. The method as recited in claim 6, wherein a minimum required size of the log object is smaller than a minimum required size of the file object.

11. The method as recited in claim 10, wherein the minimum required size of the file object corresponds to a minimum required size of a given file object for storage in an object storage service of a service provider network.

12. The method as recited in claim 6, further comprising:

determining, by the file-object manager, that one or more eviction criteria are met to evict the file object; and

in response to determining that the one or more eviction criteria are met, calling, by the file-object manager, the object storage interface to store, via an application-level network protocol, the file object into a file object store in the object storage system.

13. The method as recited in claim 6, further comprising:

for a file system write received from a client;

in response to determining, by the file-object manager, that the file system write is a durable write;

upon creating or updating a corresponding file object in the file object construction buffer and receiving an acknowledgment from the object storage interface that a corresponding log object is stored in the file system log of the object storage system, instructing, by the file-object manager, the file system interface to acknowledge to the client that the file system write is complete.

14. A non-transitory computer-readable storage medium storing program instructions that, when executed by a computing device, cause the computing device to implement:

receiving, via a file system interface, file system operations from one or more clients over a network; and

in response to receiving, via the file system interface, a file system operation for a file;

calling, by a file-object manager, an object storage interface to store a log object into a file system log in an object storage system, wherein the log object indicates the file system operation for the file; and

creating or updating, based on the file system operation for the file, a file object in a file object construction buffer in a volatile system memory, wherein the file object corresponds to the file.

15. The non-transitory, computer-readable storage medium of claim 14, wherein the file-object manager does not have access to local persistent storage and the object storage interface stores the log object to the file system log in the object storage system via an application-level network protocol.

16. The non-transitory, computer-readable storage medium of claim 14, wherein the program instructions cause the computing device to implement:

in response to a reboot or power failure that results in loss of content stored in the volatile system memory;

calling, by a file-object manager, the object storage interface to retrieve, via an application-level network protocol, log objects from the file system log of the object storage system; and

replaying, by a file-object manager, file system operations indicated by the log objects in a same sequential order that the file system operations were previously applied to one or more file objects in the volatile system memory to restore the one or more file objects in the volatile system memory.

17. The non-transitory, computer-readable storage medium of claim 14, wherein the program instructions cause the computing device to implement:

determining, by the file-object manager, that one or more eviction criteria are met to evict the file object; and

in response to determining that the one or more eviction criteria are met, calling, by the file-object manager, the object storage interface to store the file object into a file object store in the object storage system.

18. The non-transitory, computer-readable storage medium of claim 17, wherein the one or more eviction criteria comprises at least one of:

available storage space for objects in the file object construction buffer falling below a threshold value, or

the file object being the least recently used file object or being in a group of least recently used file objects in the file object construction buffer, or

receiving an indication to commit the file.

19. The non-transitory, computer-readable storage medium of claim 14, wherein the program instructions cause the computing device to implement:

for a file system write received from a client;

upon creating or updating a corresponding file object in the file object construction buffer and receiving an acknowledgment from the object storage interface that a corresponding log object is stored in the file system log of the object storage system, instructing, by the file-object manager, the file system interface to acknowledge to the client that the file system write is complete.

20. The non-transitory, computer-readable storage medium of claim 19, wherein the file system write is indicated as a durable write.