Snapshot isolation in a distributed storage system

1. A method comprising:

receiving a request for a data object stored at a data storage system comprising a plurality of storage nodes, the request comprising a key associated with the data object and a request timestamp;

identifying, based on the key and a mapping of keys to storage nodes, a first storage node in the plurality of storage nodes from which to read the data object;

determining that the data object is associated with a transaction comprising at least one operation associated with a second storage node in the plurality of storage nodes, the determining based on an entry for the data object in a key-value table, wherein the entry for the data object includes the key and a transaction identifier that identifies the transaction and a transaction table;

identifying a transaction timestamp based on a query of the transaction table using the transaction identifier;

determining that the data object is ready to be read based on a comparison of a marker of the first storage node to the transaction timestamp; and

providing the data object when the data object is ready to be read.

2. The method of claim 1 , wherein the key-value table is stored on the first storage node.

3. The method of claim 1 , wherein the key comprises a sharding value and an object identifier.

4. The method of claim 3 , wherein each storage node in the plurality of storage nodes is associated with a range of sharding values.

5. The method of claim 1 , further comprising identifying the transaction table from a plurality of transaction tables based on a transaction map, wherein each transaction table in the plurality of transaction tables is stored on a storage node in the plurality of storage nodes.

6. The method of claim 1 , further comprising identifying the transaction table from a plurality of transaction tables based on an encoding of the transaction table in the transaction identifier.

7. The method of claim 1 , further comprising:

identifying a data object timestamp from the entry for the data object; and

determining that the data object is ready to be read based on a comparison of the marker of the first storage node to the data object timestamp.

8. The method of claim 1 , further comprising:

updating a data object timestamp in the entry for the data object with the transaction timestamp.

9. A non-transitory computer-readable medium comprising instructions that, when executed by one or more processors, cause a computing device to:

receive a request to perform a transaction to write data to a data storage system comprising a plurality of storage nodes, wherein the transaction comprises at least a first write operation on a first storage node in the plurality of storage nodes and a second write operation on a second storage node in the plurality of storage nodes;

create, for the first write operation, a first entry in a first key-value table, wherein the first entry includes a transaction identifier that identifies the transaction and a transaction table;

create, for the second write operation, a second entry in a second key-value table, wherein the second entry includes the transaction identifier;

create, for the transaction, a third entry in the transaction table, wherein the third entry includes the transaction identifier;

perform the first write operation on the first storage node and the second write operation on the second storage node;

in response to a read request to read a data object associated with the data, identify a transaction timestamp by querying the transaction table using the transaction identifier;

determine that the data object is ready to be read based on a comparison of a marker of the first storage node to the transaction timestamp; and

provide the data object when the data object is ready to be read.

10. The non-transitory computer-readable medium of claim 9 , wherein the first key-value table is stored on the first storage node and the second key-value table is stored on the second storage node.

11. The non-transitory computer-readable medium of claim 9 , wherein the instructions further cause the computing device to:

identify the first storage node and the second storage node for handling the request to write data to the data storage system based on a mapping of key ranges to storage nodes.

12. The non-transitory computer-readable medium of claim 9 , wherein the first entry further includes a first timestamp generated by the first storage node, and wherein the second entry further includes a second timestamp generated by the second storage node.

13. The non-transitory computer-readable medium of claim 9 , wherein the first write operation comprises an operation to write metadata associated with a content item.

14. A distributed storage system comprising:

a plurality of storage nodes; and

at least one front end device including at least a first processor configured to:

receive a request for a data object from a client device, the request comprising a key associated with the data object, and

identify, based on the key and a mapping of keys to storage nodes, a first storage node in the plurality of storage nodes from which to read the data object; and

wherein the first storage node including at least a second processor in the plurality of storage nodes is configured to:

determine that the data object is associated with a transaction comprising at least one operation associated with a second storage node in the plurality of storage nodes, the determining based on an entry for the data object in a key-value table, wherein the entry for the data object includes the key and a transaction identifier that identifies the transaction and a transaction table;

identify a transaction timestamp based on a query of the transaction table using the transaction identifier;

determine that the data object is ready to be read based on a comparison of a marker of the first storage node to the transaction timestamp; and

provide the data object when the data object is ready to be read.

15. The distributed storage system of claim 14 , wherein the first storage node is further configured to:

identify the transaction identifier in the entry for the data object in the key-value table indicating that the data object is associated with the transaction comprising at least one operation associated with the second storage node in the plurality of storage nodes;

wherein the data object is ready to be read based on a status of the at least one operation.

16. The distributed storage system of claim 15 , wherein the key-value table is stored on the first storage node.

17. The distributed storage system of claim 15 , wherein the front end device is further configured to identify the transaction table from a plurality of transaction tables based on an encoding of the transaction table in the transaction identifier.

18. The distributed storage system of claim 15 , wherein the first storage node is further configured to:

update a data object timestamp in the entry for the data object with the transaction timestamp.

19. The distributed storage system of claim 14 , wherein the at least one front end device and the plurality of storage nodes are a part of a content management system.