FILE SYSTEM MODE SWITCHING IN A DISTRIBUTED STORAGE SERVICE

• US 20170277715A1
• Filed: 03/25/2016
• Published: 09/28/2017
• Est. Priority Date: 03/25/2016
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: STRAUSS, Jacob A., LUSZCZ, Jocob David, JONES, Peter, FRASCA, Michael Robert, CHARBONNEAU, Neal John
• Granted Patent: US 10,545,927 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/1865 Transactional file systems

Abstract

A distributed file storage services (DFSS) that provides low latency (LL) and high throughput (HT) modes for file systems. A HT metadata subsystem (HTMS) may use a distributed transaction system for data and metadata reads and writes for file systems in HT mode. A LL metadata subsystem (LLMS) may process metadata requests for file systems in LL mode. For metadata read requests, the LLMS may locally cache at least some of the file system metadata. For metadata write requests, the LLMS may write entries to a journal and notify the access node after the entries are committed to the journal. Data reads and writes for file systems in LL mode may be handled similarly to data reads and writes for file systems in HT mode. A file system may be converted from LL to HT mode, or from HT to LL mode.

Claims

1. A distributed storage service, comprising:

a plurality of computing devices that implement;

one or more low latency (LL) servers configured to support one or more of a plurality of file systems that are in LL mode, wherein the one or more LL servers are configured to;

determine write operations according to metadata requests received from one or more sources;

write journal entries for the write operations to metadata journals for respective file systems; and

send notifications to respective ones of the sources indicating results of the metadata requests, wherein metadata write operations specified by the journal entries are committed in a storage system of the distributed storage service subsequent to sending the notifications; and

a high throughput (HT) metadata subsystem configured to support one or more of the plurality of file systems that are in HT mode, wherein the HT metadata subsystem is configured to communicate with the storage system using distributed transactions to perform metadata operations indicated by metadata requests received from one or more sources and to notify respective sources after completion of the transactions.

2. The distributed storage service as recited in claim 1, wherein the one or more LL servers are further configured to:

determine read operations according to the metadata requests received from the one or more sources;

obtain metadata for at least one read operation from a local metadata cache; and

fetch metadata for at least one read operation from the storage system and store the fetched metadata to the local metadata cache.

3. The distributed storage service as recited in claim 1, wherein the one or more sources include an access subsystem of the distributed storage service, wherein the access subsystem is configured to:

receive file system requests for the plurality of file systems;

for at least one file system request directed to a file system in LL mode, send a metadata request to an LL server for the respective file system; and

for at least one file system request directed to a file system in HT mode, send the metadata request to the HT metadata subsystem.

4. The distributed storage service as recited in claim 3, wherein the access subsystem is further configured to, for file system requests directed to file system data in either mode:

communicate with the storage subsystem using distributed transactions to perform file system data operations indicated by the file system requests; and

notify respective clients after completion of the transactions.

5. The distributed storage service as recited in claim 3,

wherein the distributed storage service is configured to:

determine that a file system in HT mode is to be converted to LL mode;

convert the file system to LL mode; and

assign the file system to one of the one or more LL servers;

wherein the access subsystem is configured to;

send a metadata request for the file system in LL mode to one of the one or more LL servers;

receive a message from the LL server indicating that the LL server is not assigned to the file system; and

in response to the message, either send the metadata request to another one of the one or more LL servers or request the distributed storage service to assign the file system to another LL server.

6. The distributed storage service as recited in claim 1, wherein the distributed storage service is configured to:

determine that a file system in LL mode is to be converted to HT mode;

stop journal writes to the metadata journal for the file system by a respective LL server;

flush content of the metadata journal to the storage subsystem; and

convert the file system to HT mode so that metadata requests for the file system are sent to the HT metadata subsystem.

7. A method, comprising:

performing, by a plurality of computing devices;

receiving requests for a plurality of file systems in a distributed storage service;

for at least one request;

communicating with a storage system of the distributed storage service using one or more distributed transactions to perform one or more operations indicated by the request; and

notifying a source of the request after completion of the one or more distributed transactions; and

for at least one other request;

writing at least one journal entry specifying one or more write operations indicated by the request to a metadata journal for the respective file system; and

notifying a source of the request to indicate results of the request, wherein the write operations specified by the at least one journal entry are committed in the storage system subsequent to the notification.

8. The method as recited in claim 7, further comprising:

determining read operations according to the requests;

obtaining metadata for at least one read operation from a metadata cache; and

fetching metadata for at least one other read operation from the storage system and

storing the fetched metadata to the metadata cache.

9. The method as recited in claim 7, further comprising:

receiving file system requests for the plurality of file systems;

for one or more of the file system requests;

determining a mode of the file system at which the file system request is targeted;

for file systems in a first mode, sending one or more requests to two or more of the plurality of computing devices that perform said communicating and said notifying; and

for file systems in a second mode, sending one or more requests to one of the plurality of computing devices that performs said writing and said notifying.

10. The method as recited in claim 9, further comprising, for at least one file system request, communicating with the storage system to perform file system data operations indicated by the file system request.

11. The method as recited in claim 7, wherein said communicating and notifying are performed by a metadata subsystem implemented by two or more of the plurality of computing devices for file systems in a first mode, and wherein said writing and notifying are performed by one or more metadata servers implemented by one or more of the plurality of computing devices for file systems in a second mode, the method further comprising:

determining that a file system in the first mode is to be converted to the second mode;

converting the file system to the second mode; and

assigning one of the metadata servers to the file system in the second mode.

12. The method as recited in claim 11, further comprising the metadata server assigned to the file system:

reading metadata pages of the file system from the storage system, wherein the metadata pages includes slot ID fields; and

for at least one metadata page read from the storage system, writing a new slot ID to the slot ID field of the metadata page in the storage system to indicate that the metadata server owns the respective metadata page.

13. The method as recited in claim 7, wherein said communicating and notifying are performed by a metadata subsystem implemented by two or more of the plurality of computing devices for file systems in a first mode, and wherein said writing and notifying are performed by one or more metadata servers implemented by one or more of the plurality of computing devices for file systems in a second mode, the method further comprising:

determining that a file system in the second mode is to be converted to the first mode;

stopping journal writes to the metadata journal for the file system by the respective metadata server;

flushing content of the metadata journal to the storage subsystem; and

converting the file system to the first mode, wherein requests for the file system in the first mode are sent to the metadata subsystem.

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

receiving, by the metadata subsystem, a request for the file system, wherein the request specifies a slot ID for the file system;

in response to the request, the metadata subsystem;

checking the slot ID of the request against a slot ID indicated by a slot ID field of a target file system page in the storage subsystem;

if the slot ID of the request is greater than or equal to the slot ID of the target file system page, performing one or more operations according to the request; and

if the slot ID of the request is less than the slot ID of the target file system page, notifying a source of the request that the file system'"'"'s mode has changed.

15. The method as recited in claim 7, wherein said communicating and notifying are performed for file systems in a first mode, and wherein said writing and notifying are performed for file systems in a second mode, the method further comprising:

determining that a file system is to be converted from the first mode to the second mode or from the second mode to the first mode; and

in response to said determining, converting the file system from its current mode to the other mode.

16. The method as recited in claim 15, wherein said determining that a file system is to be converted from the first mode to the second mode or from the second mode to the first mod comprises analyzing requests to the file system to determine a best mode for the file system.

17. The method as recited in claim 15, wherein said determining that a file system is to be converted from the first mode to the second mode or from the second mode to the first mod comprises receiving a request to convert the file system from its current mode to the other mode from a customer of the distributed storage service that owns the file system.

18. 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 low latency (LL) metadata subsystem comprising one or more servers, the servers associated with one or more of a plurality of file systems stored in a distributed storage service, wherein the one or more servers are configured to;

determine write operations according to metadata requests received from the distributed storage service;

write journal entries for the write operations to metadata journals for respective file systems; and

send notifications indicating results of the metadata requests, wherein write operations specified by the journal entries are committed in the storage subsystem subsequent to sending the notifications; and

a high throughput (HT) metadata subsystem configured to communicate using distributed transactions according to a distributed transaction system to transactionally perform metadata operations indicated by metadata requests and to send notifications after completion of the transactions.

19. The non-transitory computer-readable storage medium as recited in claim 18, wherein the one or more servers are configured to:

receive metadata requests for a file system that has been converted from HT mode to LL mode and assigned to the server; and

in response to the metadata requests;

read metadata pages of the file system from the storage subsystem, wherein each metadata page includes a slot ID field;

cache the metadata pages to a local metadata cache; and

for at least one metadata page read from the storage subsystem, write a new slot ID to the slot ID field of the metadata page in the storage subsystem to indicate that the server owns the respective metadata page.

20. The non-transitory computer-readable storage medium as recited in claim 18, wherein the one or more servers are configured to:

stop journal writes to the metadata journal for a file system that is to be converted from LL mode to HT mode; and

flush content of the metadata journal for the file system to the storage subsystem.

21. The non-transitory computer-readable storage medium as recited in claim 18, wherein the HT metadata subsystem is configured to:

receive a metadata request for a file system, wherein the metadata request specifies a slot ID for the file system;

in response to the metadata request;

check the slot ID of the metadata request against a slot ID indicated by a slot ID field of a target file system page in the storage subsystem;

if the slot ID of the metadata request is greater than or equal to the slot ID of the target file system page, perform one or more operations according to the metadata request; and

if the slot ID of the metadata request is less than the slot ID of the target file system page, notify the access subsystem.