Automatic repair of corrupt HBases

• US 9,842,126 B2
• Filed: 03/15/2013
• Issued: 12/12/2017
• Est. Priority Date: 04/20/2012
• Status: Active Grant

• Current Assignee: Cloudera Incorporated
• Original Assignee: Cloudera Incorporated
• Inventors: Hsieh, Jonathan Ming-Cyn
• Primary Examiner(s): Rostami, Mohammad S
• Assistant Examiner(s): HARMON, COURTNEY N
• Application Number:
  US13/842,967
• Publication Number:
  US 20130282668A1
• Time in Patent Office: 1,733 Days
• CPC Class Codes: G06F 16/10 File systems; File servers
• CPC Class Codes: G06F 16/182 Distributed file systems
• CPC Class Codes: G06F 16/215 Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors

Abstract

Systems and methods for checking for region consistency and table integrity problems and automatically repairing a corrupted HBase cluster. The methods and systems operate in a diagnostic mode and a diagnostic and repair mode. The methods include fixing table integrity problems, such as backwards table regions, table region holes, table region overlap, and the like to restore table integrity invariant. Once the table integrity has been restored, each row key resolves to exactly one region. The methods further include fixing region inconsistencies, such as bad region assignment, no region present in the meta table, region information not in the Hadoop Distributed File System (HDFS), and the like to restore region consistency invariant. The information in the HDFS is taken as ground truth and any meta table or assignment problems that are inconsistent with the HDFS is deemed wrong and removed.

Claims

1. A method for maintaining table integrity of a datastore table in a distributed data cluster that relies on the datastore table for locating data, the datastore table having rows and being partitioned into regions, each region having a start key and a stop key for identifying which rows map to which region, the distributed data cluster including (1) a number of region servers, each region server maintaining one or more of the regions and (2) a distributed file system (DFS) that stores the data, the method comprising:

identifying, by scanning all rows in the datastore table, whether each possible row in the datastore table maps to one and only one region;

upon identifying that a particular row does not map to one and only one region, determining that a table integrity problem exists;

determining a type of the table integrity problem;

deciding a repair option based on the type of the table integrity problem, wherein the repair option is to cause the particular row to become mapped to one and only one correct region, in consistency with the data stored in the DFS; and

resolving the table integrity problem by executing the repair option.

2. The method of claim 1, wherein the type of table integrity problem includes at least one of:

a degenerate region, a backward region, a hole between regions, or a region overlap.

3. The method of claim 1, wherein the particular row maps to more than one region, and wherein the repair option includes:

moving affected data to a temporary directory;

modifying a start key or a stop key of a particular region associated with the table integrity problem such that the particular row becomes mapped to the one and only one correct region; and

restoring the affected data from the temporary directory to the one and only one correct region.

4. The method of claim 3, wherein the type of the table integrity problem is a degenerate region, which includes a start key that equals a stop key.

5. The method of claim 3, wherein the type of the table integrity problem is a backward region, which includes a start key that is smaller than a stop key.

6. The method of claim 3, wherein the type of the table integrity problem is a region overlap which includes two regions having a same start key.

7. The method of claim 1, wherein the particular row maps to no region, and wherein the repair option includes:

fabricating a new region in the DFS;

updating a meta table to reflect the new region in the DFS; and

assigning the new region to a particular region server.

8. The method of claim 7, wherein the type of the table integrity problem is a hole between regions.

9. The method of claim 1, further comprising:

automatically performing said scanning of the datastore table, said determining of the type of table integrity problem, said deciding of the repair option, and said executing the repair option, based on a time schedule.

10. The method of claim 1, wherein the distributed data cluster implements a non-relational database management system.

11. The method of claim 1, wherein the datastore table is an Apache HBase™

table.

12. A method for maintaining region consistency in a distributed data cluster that relies on a datastore table for locating data, the datastore table having rows and being partitioned into regions, each region having a start key and a stop key for identifying which rows map to which region, the distributed data cluster including (1) a number of region servers, each region server maintaining one or more of the regions and (2) a distributed file system (DFS) that stores the data, the method comprising:

gathering information associated with the regions from locations including the region servers, the DFS, and a meta table that records which region server has a particular row for a particular access request,

identifying, based on the information associated with the regions, whether each available region is assigned to one and only one region server;

upon identifying that a particular region is not assigned to one and only one region server, or that any information associated with the regions from one location is inconsistent with such information from another location, determining that a region consistency problem exists;

determining a type of the region consistency problem;

deciding a repair option based on the type of the region consistency problem, wherein the repair option is to cause region information from the region servers and region information in the meta table to be consistent with the data stored in the DFS; and

resolving the region consistency problem by executing the repair option.

13. The method of claim 12, wherein the type of region consistency problem includes at least one of:

assignment inconsistency in the region servers, no region information present in the DFS, no region information in the meta table, or an orphan region in the DFS.

14. The method of claim 13, wherein the assignment inconsistency includes at least one of:

unassigned regions, incorrectly assigned regions, or multiply assigned regions.

15. The method of claim 12, wherein the type of region consistency problem is assignment inconsistency, and wherein the repair option includes:

examining region information on the DFS to determine, for a particular region associated with the assignment inconsistency problem, the one and only one correct region server to which the particular region should be assigned; and

modifying region information on region servers associated with the assignment inconsistency problem such that the particular region becomes assigned to the one and only one correct region server.

16. The method of claim 12, wherein the type of region consistency problem is no region information present in the DFS, and wherein the repair option includes:

removing region information of a particular region associated with the assignment inconsistency problem in the meta table; and

removing region information of the particular region associated with the assignment inconsistency problem in the region servers.

17. The method of claim 12, wherein the type of region consistency problem is no region information in the meta table, and wherein the repair option includes:

adding or updating region information of a particular region associated with the assignment inconsistency problem in the meta table.

18. The method of claim 12, wherein the type of region consistency problem is an orphan region in the DFS, and wherein the repair option includes:

upon detecting that a region in the DFS has no region information in the meta table and has no region server assignment, adopting the region into the data cluster by;

creating a corresponding entry for the region in the meta table; and

assigning the region to a region server.

19. The method of claim 12, wherein the distributed data cluster implements a non-relational database management system.

20. The method of claim 12, wherein the DFS is an Apache Hadoop™

Distributed File System.

21. A distributed database system having a datastore table for locating data, the datastore table having rows and being partitioned into regions, each region having a start key and a stop key for identifying which rows map to which region, the system comprising:

a number of region servers, each region server maintaining one or more of the regions;

a distributed file system (DFS) that stores the data; and

a master node configured to,

(A) in a first mode, perform;

identifying, by scanning all rows in the datastore table, whether each possible row in the datastore table maps to one and only one region;

upon identifying that a particular row does not map to one and only one region, determining that a table integrity problem exists;

determining a type of the table integrity problem;

deciding a repair option based on the type of the table integrity problem, wherein the repair option is to cause the particular row to become mapped to one and only one correct region, in consistency with the data stored in the DFS; and

resolving the table integrity problem by executing the repair option;

(B) in a second mode, perform;

gathering information associated with the regions from locations including the region servers, the DFS, and a meta table that records which region server has a particular row for a particular access request,

identifying, based on the information associated with the regions, whether each available region is assigned to one and only one region server;

upon identifying that a particular region is not assigned to one and only one region server, or that any information associated with the regions from one location is inconsistent with such information from another location, determining that a region consistency problem exists;

determining a type of the region consistency problem;

deciding a repair option based on the type of the region consistency problem, wherein the repair option is to cause region information from the region servers and region information in the meta table to be consistent with the data stored in the DFS; and

resolving the region consistency problem by executing the repair option.