Independent evictions from datastore accelerator fleet nodes

Patent metadata

Field	Value
Publication number	US-2020081867-A1
Application number	US-201916684943-A
Country	US
Kind code	A1
Filing date	Nov 15, 2019
Priority date	Mar 30, 2016
Publication date	Mar 12, 2020
Grant date	—

Abstract

A fleet of query accelerator nodes is established for a data store. Each accelerator node caches data items of the data store locally. In response to determining that an eviction criterion has been met, one accelerator node removes a particular data item from its local cache without notifying any other accelerator node. After the particular data item has been removed, a second accelerator node receives a read query for the particular data item and provides a response using a locally-cached replica of the data item.

First claim

1 .- 21 . (canceled)

22 . A method, comprising:
obtaining, at a first accelerator node of a plurality of accelerator nodes associated with one or more data stores, a replica of a data item of a first data store of the one or more data stores; storing the replica at a cache at the first accelerator node; and removing the replica from the cache without providing an indication of the removal to at least some other accelerator nodes of the plurality of accelerator nodes.

23 . The method as recited in claim 22 , wherein said removing the replica is responsive to determining, at the first accelerator node, that the data item meets an eviction criterion, wherein the eviction criterion comprises one or more of: (a) a criterion based on a time at which the data item was accessed, (b) a time-to-live criterion, (c) a size criterion, (d) a locality criterion or (e) a criterion based on a property of a client-side component of an acceleration service.

24 . The method as recited in claim 22 , further comprising:
instantiating at least one accelerator node of the plurality of accelerator nodes in response to a programmatic request.

25 . The method as recited in claim 22 , further comprising:
determining that a probability of obtaining a response to a read request from a second accelerator node of the plurality of accelerator nodes is greater than the probability of obtaining the response from the first accelerator node; and directing the read request to the second accelerator node.

26 . The method as recited in claim 22 , further comprising:
storing another replica of the data item at another cache at a second accelerator node of the plurality of accelerator nodes, wherein the other cache at the second accelerator node differs from the cache at the first accelerator node in one or more of: (a) a type of storage device used, (b) a performance capability or (c) a size.

27 . The method as recited in claim 22 , wherein the plurality of accelerator nodes comprises a master node and one or more non-master nodes, wherein the master node is configured to respond to write requests and read requests, and wherein the first accelerator node is a non-master node, the method further comprising:
requesting another replica of the data item from the master node by the first accelerator node.

28 . The method as recited in claim 22 , wherein the first data store comprises at least one of: (a) an instance of a relational database or (b) an instance of a non-relational database.

29 . A system, comprising:
a plurality of accelerator nodes implemented at respective computing devices, including a first accelerator node, wherein the one or more accelerator nodes are associated with one or more data stores; wherein the first accelerator node is configured to:
obtain a replica of a data item of a first data store of the one or more data stores;
store the replica in a cache at the first accelerator node; and
remove the replica from the cache without providing an indication of the removal to at least some other accelerator nodes of the plurality of accelerator nodes.

30 . The system as recited in claim 29 , wherein the replica is removed from the cache in response to determining, at the first accelerator node, that the data item meets an eviction criterion, wherein the eviction criterion comprises one or more of: (a) a criterion based on a time at which the data item was accessed, (b) a time-to-live criterion, (c) a size criterion, (d) a locality criterion or (e) a criterion based on a property of a client-side component of an acceleration service.

31 . The system as recited in claim 29 , further comprising one or more control plane components of an acceleration service, wherein the one or more control plane components are configured to:
instantiate at least one accelerator node of the plurality of accelerator nodes in response to a programmatic request.

32 . The system as recited in claim 29 , wherein the cache at the first accelerator node does not utilize a persistent storage device.

33 . The system as recited in claim 29 , wherein the plurality of accelerator nodes comprises a second accelerator node configured to:
store another replica of the data item at another cache at the second accelerator node, wherein the cache at the second accelerator node differs from the cache at the first accelerator node in one or more of: (a) a type of storage device used, (b) a performance capability or (c) a size.

34 . The system as recited in claim 29 , wherein the plurality of accelerator nodes comprises a master node and one or more non-master nodes, wherein the master node is configured to respond to write requests and read requests, wherein the first accelerator node is a non-master node, wherein the master node is configured to:
transmit, in response to receiving a request for another replica of the data item from the first accelerator node, a respective replica of the data item to the first accelerator node and at least one other non-master node.

35 . The system as recited in claim 29 , wherein the one or more data stores comprise at least one of: (a) an instance of a relational database or (b) an instance of a non-relational database.

36 . One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors cause one or more computer systems to:
obtain, at a first accelerator node of a plurality of accelerator nodes associated with one or more data stores, a replica of a data item of a first data store of the one or more data stores; store the replica in a cache at the first accelerator node; and remove the replica from the cache without providing an indication of the removal to at least some other accelerator nodes of the plurality of accelerator nodes.

37 . The one or more non-transitory computer-accessible storage media as recited in claim 36 , wherein the replica is removed from the cache in response to determining, at the first accelerator node, that the data item meets an eviction criterion, wherein the eviction criterion comprises one or more of: (a) a criterion based on a time at which the data item was accessed, (b) a time-to-live criterion, (c) a size criterion, (d) a locality criterion or (e) a criterion based on a property of a client-side component of an acceleration service.

38 . The one or more non-transitory computer-accessible storage media as recited in claim 36 , wherein the cache at the first accelerator node comprises at least a portion of a persistent storage device.

39 . The one or more non-transitory computer-accessible storage media as recited in claim 36 , wherein at least the first accelerator node is configured at an acceleration service, and wherein one or more data stores comprise at least one of: (a) a file system, (b) a data store configured at least in part at a premise of a client of the acceleration service, or (c) a data store configured at least in part at a third-party premise.

40 . The one or more non-transitory computer-accessible storage media as recited in claim 36 , storing further program instructions that when executed on or across one or more processors cause the one or more computer systems to:
submit, from the first accelerator node after removal of the replica from the cache, a request for another replica of the data item to a second accelerator node of the plurality of accelerator nodes, wherein the plurality of accelerator nodes includes a master node configured to respond to write requests and re

Assignees

• Amazon Tech Inc

Inventors

• Muniswamy Reddy Kiran Kumar
• Sasidharan Anand
• Zaki Omer Ahmed
• O'Neill Brian

Classifications

• G06F12/0868

  Data transfer between cache memory and other subsystems, e.g. storage devices or host systems · CPC title

• G06F3/0656

  Data buffering arrangements · CPC title

• G06F3/061

  Improving I/O performance · CPC title

• G06F3/067 Primary

  Distributed or networked storage systems, e.g. storage area networks [SAN], network attached storage [NAS] · CPC title

• G06F16/24552

  Database cache management · CPC title