Continuous namespace verification for cluster filesystems

Patent metadata

Field	Value
Publication number	US-12164475-B2
Application number	US-202217958912-A
Country	US
Kind code	B2
Filing date	Oct 3, 2022
Priority date	Oct 3, 2022
Publication date	Dec 10, 2024
Grant date	Dec 10, 2024

Abstract

Embodiments for providing adaptive namespace verification with high efficiency. The verification process is used in modern filesystems to check the integrity of key data structures, such for backup systems where providing thorough and efficient mechanisms for verification is critical. Embodiments accelerate the verification process by selectively applying continuous, differential, and integral verification techniques for startup and runtime repair purposes to overcome present problems for system capacities that exceed the petabyte range and impose significant time requirements using existing verification methods. Embodiments work with both single-node and distributed filesystems.

First claim

What is claimed is:

1. A method of verifying filesystem data in a cluster filesystem using a continuous verification process, comprising:
checking, upon a system restart after an event, metadata just written to persistent storage, for each node of the cluster filesystem, by monitoring the flow of transactions written to the persistent storage, wherein the cluster filesystem comprises a network of individual nodes having a log-structured file system implemented in a Data Domain Filesystem, and wherein the namespace contains file metadata implemented as a B+ tree comprising a balanced tree of fixed depth, wherein a node is a page of space, and related metadata is stored in leaf pages and separately maintains the metadata and file data, and comprises a snapshot process that provides a consistent point-in-time image between the metadata and corresponding file data, and further wherein the metadata is processed using front-end nodes, and the file data is processed using back-end nodes;
snooping namespace writes that are logged in a transaction log configured to store all namespace operations, and that comprises a persistent transaction log comprising a circular buffer data structure having a fixed size;
storing the logged namespace writes in a verification cache;
receiving a list of modified pages replayed after the system restart;
verifying, using an idle thread, each modified page by backtracking through events of the event to ensure that replayed pages have been correctly written to the persistent storage;
translating the transaction log entries are into page identifiers that are affected by the change;
condensing page information into a stream of identifiers; and
storing, in batches upon a flush operation of the transaction log, the stream in the verification cache.

2. The method of claim 1 wherein the namespace stores names of directories and files, and related information as metadata that includes file ownership, creation time, and wherein the namespace operations comprise filesystem creates, updates, and deletes, and further wherein the namespace of the cluster filesystem contains file metadata implemented as a B+ tree comprising a balanced tree of fixed depth, wherein the node is a page of space and all metadata is stored in leaf pages.

3. The method of claim 2 wherein the verification cache is embodied in a B+ tree and configured to hold a condensed cache of metadata modifications, and has a key/value orientation.

4. The method of claim 1 wherein the keys comprise unique identifiers of the Mtree B+ Tree pages (DMBT pages) and inode objects.

5. The method of claim 4 wherein the keys are provided in a format: <mid><sid><page_no>, and <pid><cid><sid> respectively.

6. The method of claim 5 wherein information regarding modified files and their connection to directories and specific to the continuous verification process is saved in a value portion of key-value structures in a key-value (KV) store.

7. The method of claim 1 wherein the idle thread comprises a thread that is idle when there is no work for it to consume and that is activated when one of: new keys are added to the verification cache, or when a set timer count has expired.

8. A continuous file verification method for a cluster filesystem having a namespace, comprising:
running a continuous verification process on each node of the cluster filesystem comprising a network of individual nodes having a log-structured file system implemented in a Data Domain Filesystem, and wherein the namespace contains file metadata implemented as a B+ tree comprising a balanced tree of fixed depth, wherein a node is a page of space, and related metadata is stored in leaf pages and separately maintains the metadata and file data, and comprises a snapshot process that provides a consistent point-in-time image between the metadata and corresponding file data, and further wherein the metadata is processed using front-end nodes, and the file data is processed using back-end nodes;
logging file create, update and delete operations in a transaction log comprising a persistent transaction log comprising a circular buffer data structure having a fixed size;
snooping namespace write operations to the transaction log;
depositing the namespace write operations in a verification cache; and
reading the verification cache using a verification thread to verify pages in the namespace;
verifying, using an idle thread, modified pages replayed after a system restart by backtracking through events of the event to ensure that replayed pages have been correctly written to the persistent storage;
translating the transaction log entries are into page identifiers that are affected by the change;
condensing page information into a stream of identifiers; and
storing, in batches upon a flush operation of the transaction log, the stream in the verification cache.

9. The method of claim 8 wherein the verification process is performed after a crash of the filesystem, the method further comprising:
restarting, after the crash, the filesystem;
performing a replay operation to construct a list of the modified pages; and
verifying each modified page in the list prior to bringing the filesystem online.

10. The method of claim 9 wherein the verifying step comprises:
backtracking through events of the crash; and
ensuring that replayed commands have been correctly written to persistent storage.

11. The method of claim 8 wherein the cluster filesystem separately maintains the metadata and file data, and comprises a snapshot process that provides a consistent point-in-time image between the metadata and corresponding file data, and further wherein the metadata is processed using front-end nodes, and the file data is processed using back-end nodes.

12. The method of claim 11 wherein the transaction log comprises a transaction log storing a list of metadata changes comprising a modified directory manager B+ tree (DMBT) and modified file inode information, and wherein the verification cache comprises a B+ tree data element storing a condensed cache of the metadata changes with a key/value orientation, and further wherein the continuous verification process is performed by an idle thread comprising a thread that is idle when there is no work for the thread to consume and that is activated when one of: new keys are added to the verification cache, or when a set timer count has expired.

13. A cluster system for verifying filesystem data using a continuous verification process, comprising:
a persistent storage maintained in physical memory storing metadata written to the filesystem, wherein the cluster filesystem comprises a network of individual nodes having a log-structured file system implemented in a Data Domain Filesystem, and wherein the namespace contains file metadata implemented as a B+ tree comprising a balanced tree of fixed depth, wherein a node is a page of space, and related metadata is stored in leaf pages and separately maintains the metadata and file data, and comprises a snapshot process that provides a consistent point-in-time image between the metadata and corresponding file data, and further wherein the metadata is processed using front-end nodes, and the file data is processed using back-end nodes;
a hardware-based monitor component monitoring the flow of transactions written in each node of the cluster filesystem to the persistent storage to check, upon a system restart after an event, metadata just written to the persistent;
a transaction log stored in the physical memory and storing all namespace operations;
a component snooping namespace writes that are logged in the transaction log that comprises a persist

Assignees

• Dell Products Lp

Inventors

• Lakshmikantha Shashikanth
• Hickey Charles J
• Mamidi Murthy V

Classifications

• G06F16/128 Primary

  Details of file system snapshots on the file-level, e.g. snapshot creation, administration, deletion (error detection or correction of the data by redundancy in operations or in hardware G06F11/14, G06F11/16) · CPC title

• G06F16/185

  Hierarchical storage management [HSM] systems, e.g. file migration or policies thereof (details of archiving G06F16/11) · CPC title

• G06F16/16 Primary

  File or folder operations, e.g. details of user interfaces specifically adapted to file systems · CPC title