Eliminating redundant file system operations
Techniques to eliminate redundant file system operations are disclosed. In various embodiments, an indication is received to send locally-stored file system operations to a remote file system metadata server. A set of related locally-stored file system operations is determined. Operations in the set are determined to satisfy a merger criteria and are merged into a single merged file system operation. The merged file system operation is sent to the remote file system metadata server.
1. A method of processing file system operations, comprising:
writing, by one or more processors of a client connected to a distributed file system including a remote file system metadata server, one or more file system operations affecting metadata for which a lease is held to a buffer at the client, the lease being held with respect to at least a portion of a file system metadata;
obtaining, by the client, an indication that the lease is to end;
in response to obtaining the indication that the lease is to end, determining, by the client, to send locally-stored file system operations to the remote file system metadata server, the locally-stored file system operations comprising at least one of the one or more file system operations written to the buffer; and
in response to determining to send locally-stored file system operations to the remote file system metadata server,
determining, by the client, a set of related locally-stored file system operations to be communicated to the file system metadata server,
wherein:
the determining the set of related locally-stored file system operations comprises identifying from among a plurality of file system operations a linear chain of dependent file system operations based at least in part on a dependency graph or other data structure reflecting dependencies between the locally-stored file system operations,
the identifying the linear chain of dependent file system operations processing the dependency graph to find the linear chain,
the identifying the linear chain being based at least in part on a determination that a sequence of operations each have at most a single dependency on a prior operation in the linear chain and none of which has more than one inbound dependency;
identifying, by the client, at least a subset of operations in the set as satisfying a merger criteria;
merging, by the client, the subset of operations into a single merged file system operation; and
sending, by the client, the merged file system operation to the remote file system metadata server, wherein the merged file system operation is sent based at least in part on the receiving of the indication, wherein sending the merged file system operation includes ensuring that an operation of the one or more file system operations is not sent to the remote file system metadata server unless all operations on which the operation depends have already been sent to the remote file system metadata server or are being sent in connection with the operation.
2. The method of claim 1 , wherein identifying at least a subset of operations in the set as satisfying a merger criteria comprises determining that two or more operations comprising the subset would result in making successive overwrites of a same file system metadata attribute.
3. The method of claim 2 , wherein merging the subset of operations into a single merged file system operation comprises removing all but a most recently received one of the file system operations in the subset.
4. The method of claim 2 , wherein merging the subset of operations into a single merged file system operation comprises modifying an operation in the subset to reflect a final value for the file system metadata attribute that is the same as an ending value to which the file system metadata attribute would have been set had the operations in the subset been performed without merger.
5. The method of claim 1 , wherein the locally-stored file system operations comprise file system operations that have been buffered locally to permit asynchronous transmission of at least one of the one or more file system operations from the client to the remote file system metadata server.
6. The method of claim 5 , wherein the lease corresponds to a metadata write lease, and wherein the locally-stored file system operations are associated with a portion of file system metadata with respect to which the metadata write lease is held.
7. The method of claim 6 , wherein the determining to send the locally-stored file system operations to the remote file system metadata server is based at least in part on a lease break communication with respect to the metadata write lease.
8. The method of claim 1 , wherein sending the merged file system operation to the remote file system metadata server includes combining the merged file system operation with one or more other operations in the set of related locally-stored file system operations that were not merged into the merged file system operation to create a compound file system operation; and sending the compound operation to the file system metadata server.
9. A system, comprising:
a communication interface; and
a processor of a client connected to a distributed file system including a remote file system metadata server, the processor coupled to the communication interface and configured to:
write one or more file system operations affecting metadata for which a lease is held to a buffer at the client, the lease being held with respect to at least a portion of a file system metadata;
obtain an indication that the lease is to end
in response to obtaining the indication that the lease is to end, determine to send locally-stored file system operations to the remote file system metadata server, the locally-stored file system operations comprising at least one of the one or more file system operations written to the buffer; and
in response to determining to send locally-stored file system operations to the remote file system metadata server,
determine a set of related locally-stored file system operations to be communicated to the file system metadata server,
wherein:
to determine the set of related locally-stored file system operations comprises identifying from among a plurality of file system operations a linear chain of dependent file system operations based at least in part on a dependency graph or other data structure reflecting dependencies between the locally-stored file system operations,
the identifying the linear chain of dependent file system operations processing the dependency graph to find the linear chain,
the linear chain being based at least in part on a determination that a sequence of operations each have at most a single dependency on a prior operation in the linear chain and none of which has more than one inbound dependency;
identify at least a subset of operations in the set as satisfying a merger criteria;
merge the subset of operations into a single merged file system operation; and
send the merged file system operation to the remote file system metadata server via the communication interface, wherein the merged file system operation is sent based at least in part on the receiving of the indication, wherein sending the merged file system operation includes ensuring that an operation of the one or more file system operations is not sent to the remote file system metadata server unless all operations on which the operation depends have already been sent to the remote file system metadata server or are being sent in connection with the operation.
10. The system of claim 9 , wherein identifying at least a subset of operations in the set as satisfying a merger criteria comprises determining that two or more operations comprising the subset would result in making successive overwrites of a same file system metadata attribute.
11. The system of claim 10 , wherein merging the subset of operations into a single merged file system operation comprises removing all but a most recently received one of the file system operations in the subset.
12. The system of claim 10 , wherein merging the subset of operations into a single merged file system operation comprises modifying an operation in the subset to reflect a final value for the file system metadata attribute that is the same as an ending value to which the file system metadata attribute would have been set had the operations in the subset been performed without merger.
13. The system of claim 9 , wherein the locally-stored file system operations comprise file system operations that have been buffered locally to permit asynchronous transmission to the remote file system metadata server.
14. The system of claim 13 , wherein the lease corresponds to a metadata write lease, and wherein the locally-stored file system operations are associated with a portion of file system metadata with respect to which the metadata write lease is held.
15. The system of claim 14 , wherein the determining to send the locally-stored file system operations to the remote file system metadata server is based at least in part on a lease break communication with respect to the metadata write lease.
16. The system of claim 9 , wherein sending the merged file system operation to the remote file system metadata server includes combining the merged file system operation with one or more other operations in the set of related locally-stored file system operations that were not merged into the merged file system operation to create a compound file system operation; and sending the compound operation to the file system metadata server.
17. A computer program product to process file system operations, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:
writing, by one or more processors of a client connected to a distributed file system including a remote file system metadata server, one or more file system operations affecting metadata for which a lease is held to a buffer at the client, the lease being held with respect to at least a portion of a file system metadata;
obtaining, by the client, an indication that the lease is to end;
in response to obtaining the indication that the lease is to end, determining, by the client, to send locally-stored file system operations to the remote file system metadata server, the locally-stored file system operations comprising at least one of the one or more file system operations written to the buffer; and
in response to determining to send locally-stored file system operations to the remote file system metadata server,
determining, by the client, a set of related locally-stored file system operations to be communicated to the file system metadata server,
wherein:
the determining the set of related locally-stored file system operations comprises identifying from among a plurality of file system operations a linear chain of dependent file system operations based at least in part on a dependency graph or other data structure reflecting dependencies between the locally-stored file system operations,
the identifying the linear chain of dependent file system operations processing the dependency graph to find the linear chain,
the identifying the linear chain being based at least in part on a determination that a sequence of operations each have at most a single dependency on a prior operation in the linear chain and none of which has more than one inbound dependency;
identifying, by the client, at least a subset of operations in the set as satisfying a merger criteria;
merging, by the client, the subset of operations into a single merged file system operation; and
sending, by the client, the merged file system operation to the remote file system metadata server, wherein the merged file system operation is sent based at least in part on the receiving of the indication, wherein sending the merged file system operation includes ensuring that an operation of the one or more file system operations is not sent to the remote file system metadata server unless all operations on which the operation depends have already been sent to the remote file system metadata server or are being sent in connection with the operation.
18. The method of claim 1 , wherein communication of the single merged file system operation operatively communicates the subset of operations that are merged into the single merged file system operation in a same packet.
19. The method of claim 1 , wherein the single merged file system operation is communicated in a single packet.
20. The method of claim 1 , wherein the determining, by one or more processors, a set of related locally-stored file system operations comprises determining to cancel a set of operations bounded by a create operation and a delete operation pair.
21. The method of claim 1 , wherein the identifying the at least a subset of operations in the set as satisfying the merger criteria comprises determining whether any net change would be made to file system metadata if the set of operations in the chain were performed.
22. The method of claim 1 , wherein the merging the subset of operations into the single merged file system operation reduces a number of operations the remote file system metadata server has to perform to achieve a result corresponding to the set of operations.
23. The method of claim 1 , wherein the writing the one or more file system operations affecting metadata for which a lease is held to a buffer affecting comprises intercepting the one or more file system operations by a process running in a kernel of an operating system executed by the one or more processors, and redirecting the one or more file system operations to a file system client by the process running in the kernel.
24. The method of claim 1 , further comprising:
in response to identifying the linear chain, performing a validation processing to ensure that all operations in the linear chain would succeed if all the operations in the linear chain were sent to the remote file system metadata server in a correct order.