Backoff-based scheduling of storage object deletions

• US 9,063,946 B1
• Filed: 12/14/2012
• Issued: 06/23/2015
• Est. Priority Date: 12/14/2012
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Barber, Jeffrey Michael, Gattu, Praveen Kumar, Elving, Christopher Henning, Perry, Carl Yates, Denny-Brown, Derek Ernest II
• Primary Examiner(s): Jami, Hares
• Application Number:
  US13/715,889
• Time in Patent Office: 921 Days
• Field of Search: 707/607, 707/609, 707/610, 707689-692, 707/703
• US Class Current: 1/1
• CPC Class Codes: G06F 16/125 characterised by the use of retention policies retention policies for HSM systems G06F16/185
• CPC Class Codes: G06F 16/162 Delete operations erasing in storage systems G06F3/0652
• CPC Class Codes: G06F 16/178 Techniques for file synchronisation in file systems

Abstract

Methods and apparatus for backoff-based scheduling of storage object deletions are disclosed. A storage medium stores program instructions that when executed on a processor, obtain an indication of a collection of storage objects of a network-accessible multi-tenant storage service to be deleted in accordance with specified deletion criteria. A deletion of a storage object comprises a metadata deletion operation and one or more other operations. The instructions initiate, corresponding to at least some objects of the collection, respective metadata deletion operations at a metadata node of the storage service. If a metric associated with the metadata node meets a threshold criterion, the instructions delay, by a particular amount of time, an initiation of an operation corresponding to a deletion of another storage object.

Claims

1. A system, comprising one or more computing devices configured to:

implement a network-accessible multi-tenant storage service for storage objects owned by a plurality of clients;

identify a set of storage objects of the storage service to be deleted in accordance with one or more deletion criteria, wherein a deletion of a given storage object from the storage service comprises a metadata deletion operation and a storage release operation asynchronous with respect to the metadata deletion operation;

initiate, corresponding to each storage object of one or more storage objects of the set, a respective metadata deletion operation to delete a metadata entry specifying a data storage location of the storage object;

determine whether a performance metric associated with at least one metadata deletion operation corresponding to a particular storage object of the one or more storage objects meets a threshold criterion; and

in response to a determination that the performance metric meets the threshold criterion, delay, by a particular amount of time, an initiation of an additional metadata deletion operation corresponding to another storage object of the set.

2. The system as recited in claim 1, wherein a particular deletion criterion of the one or more deletion criteria is based at least in part on one or more of:

a time since a particular storage object of the set was created, a time since a particular storage object of the set was most recently modified, a time since a particular storage object of the set was last accessed, a frequency of access of a particular storage object of the set, or a client-specified expiration policy for a particular storage object of the set.

3. The system as recited in claim 1, wherein the performance metric comprises a response time metric, wherein the one or more computing devices are further configured to:

compute the particular amount of time based at least in part on the response time metric and at least in part on a jitter factor.

4. The system as recited in claim 1, wherein the one or more computing devices are further configured to:

compute the particular amount of time based at least in part on an identification of a particular client of the plurality of clients, wherein the particular client owns at least the other storage object.

5. The system as recited in claim 1, wherein the metadata deletion operations corresponding to the one or more storage objects of the set are implemented as background operations, wherein the one or more computing devices are further configured to:

determine at least a second performance metric associated with a foreground operation scheduled in response to a client request; and

delay, by a second amount of time based at least in part from the second performance metric, an initiation of a particular metadata deletion operation corresponding to a different storage object of the set.

6. A method, comprising

implementing a network-accessible multi-tenant storage service for storage objects owned by a plurality of clients;

identifying a set of storage objects of the storage service to be deleted in accordance with one or more deletion criteria, wherein a deletion of a given storage object from the storage service comprises a metadata deletion operation and one or more other operations;

initiating, corresponding to each storage object of one or more storage objects of the set, a respective metadata deletion operation to delete a metadata entry specifying a data storage location of the storage object;

determining whether a metric associated with at least one metadata deletion operation corresponding to a particular storage object of the one or more storage objects meets a threshold criterion; and

in response to determining that the metric meets the threshold criterion, delaying, by a particular amount of time, an initiation of an operation corresponding to a deletion of another storage object.

7. The method as recited in claim 6, wherein a particular deletion criterion of the one or more deletion criteria is based at least in part on one or more of:

a time since a particular storage object of the set was created, a time since a particular storage object of the set was most recently modified, a time since a particular storage object of the set was last accessed, a frequency of access of a particular storage object of the set, or a client-specified expiration policy for a particular storage object of the set.

8. The method as recited in claim 6, wherein the metric comprises a response time metric, further comprising:

computing the particular amount of time based at least in part on the response time metric and at least in part on a jitter factor.

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

computing the particular amount of time based at least in part on an identification of a particular client of the plurality of clients, wherein the particular client owns at least the other storage object.

10. The method as recited in claim 6, wherein the metric comprises an error metric.

11. The method as recited in claim 6, wherein the metadata deletion operations corresponding to the one or more storage objects of the set are implemented as background operations, further comprising:

determining at least a second performance metric associated with a foreground operation scheduled in response to a client request; and

delaying, by a second amount of time determined based at least in part on the second performance metric, an initiation of a particular operation corresponding to a deletion of a different storage object.

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

storing a plurality of replicas of the particular storage object in the storage service in accordance with a durability goal associated with the particular storage object, wherein the at least one metadata deletion operation corresponding to the particular storage object comprises a deletion of a keymap entry indicating a storage location of a particular replica of the plurality of replicas.

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

storing, within the storage service, a deletion job object indicating at least a subset of the identified set of storage objects.

14. The method as recited in claim 6, wherein said initiating is performed asynchronously with respect to said identifying.

15. A non-transitory computer-accessible storage medium storing program instructions that when executed on one or more processors:

obtain an indication of a collection of storage objects of a network-accessible multi-tenant storage service to be deleted in accordance with one or more deletion criteria, wherein a deletion of a given storage object from the storage service comprises a metadata deletion operation and one or more other operations;

initiate, corresponding to each storage object of one or more storage objects of the collection, a respective metadata deletion operation at a metadata node of the storage service to delete a metadata entry of the storage object;

determine, with respect to the respective metadata deletion operation at the metadata node, whether a metric associated with the metadata node meets a threshold criterion; and

in response to determining that the metric meets the threshold criterion, delay, by a particular amount of time, an initiation of an operation corresponding to a deletion of another storage object.

16. The non-transitory computer-accessible storage medium as recited in claim 15, wherein a particular deletion criterion of the one or more deletion criteria is based at least in part on one or more of:

a time since a particular storage object of the collection was created, a time since a particular storage object of the collection was most recently modified, a time since a particular storage object of the collection was last accessed, a frequency of access of a particular storage object of the collection, or a client-specified expiration policy for a particular storage object of the collection.

17. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the metric comprises a response time metric, wherein the instructions when executed on the one or more processors:

compute the particular amount of time based at least in part on the response time metric.

18. The non-transitory computer-accessible storage medium as recited in claim 17, wherein the particular amount of time is computed based at least in part on a randomly-generated jitter factor.

19. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the instructions when executed on the one or more processors:

compute the particular amount of time based at least in part on an identification of a particular client of the storage service, wherein the particular client owns at least one storage object of the collection.

20. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the metric comprises an error metric.

21. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the metadata deletion operations corresponding to the one or more storage objects of the set are implemented as background operations, wherein the instructions when executed on the one or more processors:

determine at least a second metric associated with a foreground operation performed at the metadata node in response to a client request; and

delay, by a second amount of time, an initiation of a particular operation corresponding to a deletion of a different storage object, based at least in part on the second metric.

22. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the indication of a collection of storage objects is obtained from a deletion job object stored in a container of the storage service by a deletion job generator.

23. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the instructions when executed on the one or more processors implement a deletion task dispatcher configured to assign a worker thread to submit a metadata deletion request to the metadata node.

24. The non-transitory computer-accessible storage medium as recited in claim 15, wherein the instructions when executed on the one or more processors implement a worker thread configured to obtain the indication of the collection of storage objects from a deletion task dispatcher of the storage service.