Data storage system with multi-tier control plane

1. A method, comprising:

servicing, by a designated-primary head node of a data storage unit that comprises a designated-secondary head node configured to replicate data of the designated-primary head node, and one or more data storage sleds for storing data flushed from head nodes of the data storage unit, read requests and write requests directed to a volume stored in the data storage unit; and

generating, by a local control plane of the data storage unit, a recommendation of a sandbox of specified storage, associated with the volume, in the data storage sleds;

wherein the recommendation indicates availability of the sandbox of specified storage in the data storage sleds for selection, by the designated-primary head node, for data storage during a data flush from the designated-primary head node to the data storage sleds.

2. The method of claim 1 , further comprising:

transmitting the recommendation to the designated-primary head node.

3. The method of claim 1 , further comprising:

selecting, by the designated-primary head node, from the sandbox of specified storage, storage to store data during the data flush from the designated-primary head node to the data storage sleds.

4. The method of claim 1 , further comprising:

flushing, by the designated-primary head node, the data to columns within the sandbox of specified storage without requesting, for the flush, extent allocation from the local control plane.

5. The method of claim 1 , further comprising:

determining:

that insufficient space is available in the sandbox of specified storage, or

that a particular placement of the data would cause a data durability of the saved data to fall below of minimum required durability for the volume; and

selecting, based on the determining, one or more columns outside the sandbox of specified storage to write data for the volume.

6. The method of claim 1 , further comprising:

determining that communication between the local control plane and the designated-primary head node has been lost; and

continuing, based on the determining, to flush data from the designated-primary head node to the data storage sleds.

7. The method of claim 1 , further comprising:

collecting performance metrics from one or more of the data storage sleds or the head nodes; and

issuing, based on the performance metrics, a sandbox update to the designated-primary head node;

wherein the recommendation is based at least in part on the sandbox update.

8. The method of claim 1 , further comprising:

identifying a heat collision comprising multiple of the head nodes attempting to access a same data storage sled at a same time; and

to avoid the identified heat collision:

issuing the sandbox update, or

generating and transmitting the sandbox recommendation.

9. A data storage system, comprising:

a data storage unit assigned to service read requests and write requests directed to a volume partition stored in the data storage unit, the data storage unit comprising:

a plurality of head nodes comprising:

a designated-primary head node configured to service the volume partition; and

one or more data storage sleds for storing data flushed from the head nodes;

wherein the designated-primary head node is configured to:

receive a recommendation that indicates availability of a sandbox of specified storage, allocated to the volume and in the data storage sleds, for selection, by the designated-primary head node, for data storage during a data flush from the designated-primary head node to the data storage sleds;

select, from the sandbox of specified storage in the data storage sleds that is allocated to the volume, one or more portions of storage to store data during the data flush from the designated-primary head node to the data storage sleds; and

flush data to the selected storage.

10. The data storage system of claim 9 , wherein the designated-primary head node is configured to:

determine:

that communication between a local control plane implemented by one of the head nodes and the designated-primary head node has been lost,

that a particular placement of the data would cause a data durability of the saved data to fall below of minimum required durability for the volume, or

that insufficient space is available in the specified storage of the sandbox; and

flush, based on the determination, data to columns outside of the specified storage of the sandbox.

11. The data storage system of claim 9 , wherein the designated-primary head node is further configured to flush the data to columns within the sandbox of specified storage without requesting, from a local control plane implemented by one of the head nodes, and for the flush, an allocation of memory in the storage sleds.

12. The data storage system of claim 11 , further comprising:

an extent allocation service configured to allocate extents as sets of columns of space on mass storage devices in different sleds that each store some of the flushed data;

wherein the sandbox comprises extents that include corresponding columns in multiple mass storage devices; and

wherein the recommendation is generated by a local control plane implemented by one of the head nodes and transmitted to the designated-primary head node and indicates availability of extents for selection by the designated-primary head node to select for data storage during the data flush from the designated-primary head node to the data storage sleds.

13. The data storage system of claim 11 , wherein the designated-primary head node is configured to:

determine that communication between the local control plane and the designated-primary head node has been lost; and

continue, based on the determination, to flush data from the designated-primary head node to the data storage sleds.

14. The data storage system of claim 11 , wherein:

the sandbox comprises extents that include corresponding columns in multiple mass storage devices.

15. The data storage system of claim 11 , wherein the local control plane is configured to:

collect performance metrics from one or more of the data storage sleds or the head nodes; and

issue, based on the performance metrics, a sandbox update to the designated-primary head node;

wherein the recommendation is based at least in part on the sandbox update.

16. A data storage system, comprising:

a head node of a data storage unit assigned to service read requests and write requests directed to a volume partition stored in a data storage unit, the data storage unit comprising:

respective designated-primary head nodes configured to service respective volume partitions stored in the data storage unit;

respective designated-secondary head nodes configured to replicate data of the corresponding designated-primary head nodes;

and data storage sleds for storing data flushed from the designated-primary head nodes,

wherein the head node is configured to implement a local control plane for the data storage unit, configured to:

generate and transmit a recommendation of a sandbox of specified storage, associated with a respective volume partition, in the data storage sleds, to a respective designated-primary head node for the respective volume partition;

wherein the recommendation indicates availability of the sandbox of specified storage in the data storage sleds for selection, by the respective designated-primary head node, for data storage during a data flush from the respective designated-primary head node to the data storage sleds.

17. The data storage system of claim 16 ,

wherein the respective designated-primary head node is configured to:

receive the recommendation of the sandbox of specified storage;

select, from the sandbox of specified storage, storage to flush data; and

flush data to the selected storage.

18. The data storage system of claim 16 , wherein the local control plane further comprises:

an extent allocation service configured to allocate extents as sets of columns of space on mass storage devices in different sleds that each store some of the flushed data;

wherein the sandbox comprises extents that include corresponding columns in multiple mass storage devices; and

wherein the recommendation generated by the local control plane and transmitted to the designated-primary head node indicates availability of extents for selection by the respective designated-primary head node to select for data storage during the data flush from the designated-primary head node to the data storage sleds.

19. The data storage system of claim 16 , wherein the local control plane is further configured to:

collect performance metrics from one or more of the data storage sleds or the respective designated-primary head nodes; and

issue, based on the performance metrics, a sandbox update to the respective designated-primary head node for the respective volume partition;

wherein the recommendation is based at least in part on the sandbox update.

20. The data storage system of claim 19 , wherein the local control plane is further configured to:

determine a heat collision comprising multiple of the head nodes attempting to access a same data storage sled at a same time; and

generate and transit the recommendation or issue the sandbox update to avoid the determined heat collision.