Fleet node management system

• US 10,657,119 B1
• Filed: 12/18/2015
• Issued: 05/19/2020
• Est. Priority Date: 12/18/2015
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Acheson, Alazel, Carroll, David, Stephens, Christopher Andrew, Markle, Seth William, Herman, Jonathan, Dharamshi, Mihir, Sampath, Rajesh, Ramakrishna, Shreeranga Puttur, Sajwani, Rameez
• Primary Examiner(s): Lin, Shew Fen
• Application Number:
  US14/975,353
• Time in Patent Office: 1,614 Days
• CPC Class Codes: G06F 16/23 Updating
• CPC Class Codes: G06F 16/2329 using versioning
• CPC Class Codes: G06F 2209/5015 Service provider selection
• CPC Class Codes: G06F 9/505 considering the load

Abstract

A fleet node management system may include a metadata store, a plurality of fleet nodes, and one or more metadata mutation devices. The metadata store may be configured to store dynamic metadata. The plurality of fleet nodes may be configured to determine, based on a gossip protocol, whether to continue performance of a function that uses a local version of the metadata. The one or more metadata mutation devices may be configured to determine, based on a global state of the fleet nodes, whether to modify the dynamic metadata for the fleet nodes.

Claims

1. A method, comprising:

performing, by a fleet node implemented by one or more computing devices and configured to perform one or more functions and store a local version of metadata that is used to perform the one or more functions;

requesting, within a gossip group time threshold, metadata version information from a quantity of multiple fleet nodes of a plurality of fleet nodes, wherein the quantity is at least equal to a gossip group size threshold;

in response to receiving a metadata version response from at least one of the multiple fleet nodes within the gossip group time threshold, determining whether the local version is stale based at least in part on whether the metadata version response indicates a version of the metadata that is more recent than the local version by at least a staleness threshold;

determining whether the local version is indeterminate based at least in part on whether at least the gossip group size threshold of metadata version responses were received from the multiple fleet nodes within the gossip group time threshold; and

halting the fleet node from further performance of the one or more functions if the local version is stale, indeterminate, or both;

or

continuing further performance of the one or more functions by the fleet node if the local version is not stale and is not indeterminate.

2. The method of claim 1, wherein halting further performance of the one or more functions comprises halting, by the fleet node, one or more heartbeat notifications that indicate that the fleet node is available to perform the one or more functions.

3. The method of claim 2, wherein continuing performance of the one or more functions comprises:

sending one or more heartbeat notifications that indicate that the fleet node is available to perform the one or more functions; and

in response to the one or more heartbeat notifications, receiving one or more requests to perform at least one of the one or more functions.

4. The method of claim 3, further comprising:

subsequent to determining that the local version is stale, receiving, by the fleet node, an update to the local version; and

based at least in part on receiving the update to the local version, resuming, by the fleet node, performance of the one or more functions.

5. The method of claim 4, wherein halting further performance of the one or more functions if the local version is stale comprises:

halting further performance of one or more functions corresponding to one or more portions of the local version of the local version that are stale; and

continuing further performance of one or more functions corresponding to one or more other portions of the local version that are not stale and not indeterminate.

6. The method of claim 5, further comprising, in response to determining, based on a metadata version response from a different fleet node, that the local version is stale, requesting, by the fleet node, one or more metadata updates from the different fleet node.

7. The method of claim 5, further comprising receiving, at the fleet node, updates to the local version of the metadata broadcasted from a metadata store.

8. The method of claim 5, further comprising receiving, at the fleet node, updates to the local version of the metadata in response to one or more metadata requests, from the fleet node, to a metadata store.

9. A system, comprising:

a particular fleet node implemented by one or more computing devices and configured to perform one or more functions and store respective local versions of dynamic metadata that is used to perform the one or more functions, and configured to;

request, within a gossip group time threshold, metadata version information from a quantity of multiple fleet nodes of a plurality of fleet nodes, wherein the quantity is at least equal to a gossip group size threshold;

in response to receiving a metadata version response from at least one of the multiple fleet nodes within the gossip group time threshold, determine whether the local version is stale based at least in part on whether the metadata version response indicates a version of the dynamic metadata that is more recent than the local version by at least a staleness threshold;

determine whether the local version is indeterminate based at least in part on whether at least the gossip group size threshold of metadata version responses were received from the multiple fleet nodes within the gossip group time threshold;

in response to determining that the local version is stale, indeterminate, or both, halt the particular fleet node from further performance of a function of the one or more functions, wherein the function is dependent on the local version; and

in response to determining that the local version is not stale and is not indeterminate, continue further performance of the function by the particular fleet node.

10. The system of claim 9, wherein to perform the function, the particular fleet node is configured to route one or more client requests to one or more storage devices of a plurality of storage devices, and wherein the dynamic metadata is indicative of a plurality of routes to corresponding storage devices.

11. The system of claim 9, wherein the particular fleet node is configured to request the metadata version information, to determine whether the local version is stale, and to determine whether the local version is indeterminate at least once per gossip group time threshold.

12. The system of claim 9, wherein to halt further performance of the function, the particular fleet node is configured to halt one or more heartbeat notifications that indicate that the particular fleet node is available to perform the function.

13. The system of claim 9, wherein to continue further performance of the function, the particular fleet node is configured to:

send one or more heartbeat notifications that indicate that the particular fleet node is available to perform the function; and

in response to the one or more heartbeat notifications, receive one or more requests to perform the function.

14. The system of claim 9, further comprising:

one or more metadata mutation devices implemented by one or more respective computing devices, wherein a particular metadata mutation device is configured to;

determine a desired modification to the dynamic metadata;

identify a current dynamic metadata state at a metadata store storing the dynamic metadata;

request metadata information from a quantity of the fleet nodes, wherein the quantity includes at least a mutation node check threshold of the fleet nodes, and wherein the mutation node check threshold is related to the gossip group size threshold;

in response to receiving fewer than the mutation node check threshold of metadata responses indicative of corresponding local versions within a freshness threshold, determine not to modify the dynamic metadata at the metadata store;

in response to receiving at least the mutation node check threshold of metadata responses indicative of corresponding local versions within the freshness threshold, wait a gossip group time threshold to avoid a negative impact of the desired modification on fleet nodes having out of date local versions; and

after waiting the gossip group time threshold, modify the dynamic metadata at the metadata store.

15. A non-transitory, computer-readable storage medium storing program instructions that, when executed on one or more computers, cause the one or more computers to implement a particular fleet node to:

request, within a gossip group time threshold, metadata version information from a quantity of multiple fleet nodes of a plurality of fleet nodes, wherein the metadata information pertains to a local version of dynamic metadata stored by the particular fleet node and the quantity is at least equal to a gossip group size threshold;

in response to receiving a metadata version response from a at least one of the multiple fleet nodes within the gossip group time threshold, determine whether the local version is stale based at least in part on whether the metadata version response indicates a version of the dynamic metadata that is more recent than the local version by at least a staleness threshold;

determine whether the local version is indeterminate based at least in part on whether at least the gossip group size threshold of metadata version responses were received from the multiple fleet nodes within the gossip group time threshold;

in response to determining that the local version is stale, indeterminate, or both, halt the particular fleet node from further performance of a function of the particular fleet node, wherein the function is dependent on the local version; and

in response to determining that the local version is not stale and is not indeterminate, continue further performance of the function by the fleet node.

16. The non-transitory, computer-readable storage medium of claim 15, wherein to perform the function, the program instructions when executed on the one or more computers cause the particular fleet node to route one or more client requests to one or more storage devices of a plurality of storage devices, and wherein the dynamic metadata is indicative of a plurality of routes to corresponding storage devices.

17. The non-transitory, computer-readable storage medium of claim 15, wherein the program instructions when executed on the one or more computers cause the particular fleet node to request the metadata version information, to determine whether the local version is stale, and to determine whether the local version is indeterminate at least once per gossip group time threshold.

18. The non-transitory, computer-readable storage medium of claim 15, wherein the program instructions when executed on the one or more computers cause the particular fleet node to:

subsequent to determining that the local version is stale, receive an update to the local version; and

based at least in part on receiving the update to the local version, resume performance of the one or more functions.

19. The non-transitory, computer-readable storage medium of claim 15, wherein the program instructions when executed on the one or more computers cause the particular fleet node to:

request one or more metadata updates from a different fleet node in response to determining, based on a metadata version response from the different fleet node, that the local version is stale.

20. The non-transitory, computer-readable storage medium of claim 15, wherein the program instructions when executed on the one or more computers cause the particular fleet node to:

receive updates to the local version of the metadata broadcasted from a metadata store that stores the dynamic metadata.