Spark default database migration guide

Default Apache Spark database storage location configured by Lyftrondata is in %ProgramData%\Lyftrondata\warehouse folder. It is okay for development and test environments, but for production, we recommend to change the default location to a different one.

This guide's objective is to describe actions required to migrate Spark default database to a different location.

Introduction

Spark keeps all its metadata in a Microsoft SQL Server database called SparkMetastore located on the database server designated during Lyftrondata installation.

In Lyftrondata's Spark metastore database we need to update three tables: DBS, SDS and FUNC_RU as they log the locations for database, table, and function in that order. Since Lyftrondata does not support custom Spark functions yet, we will update only SDS and DBS tables.

Database migration procedure

Prerequisites

User or users performing the migration steps need to collectively have:

• administrative access to Lyftrondata
• administrative access to Spark metadata database
• administrative access to the folder holding the actual Spark database files
• administrative access to the destination folder

Basic knowledge of MS SQL Server and SQL language is required.

Step 1, shutdown Spark instance

Connect to Lyftrondata's administrative portal with administrative privileges and shutdown Spark instance.

IMPORTANT: It is vital to stop Spark instance before attempting to do any metadata operations.

Step 2, prepare a backup of Spark Metabase

Before updating any data prepare a backup of Spark Metabase.

Step 3, create and configure a destination folder for database files

Create a folder that will hold the database files of the migrated database using tools of your operating system. In this article we assume that our destination folder is d:\warehouse.

WARNING: It is recommended to leave the source files intact, and delete them at the end of the procedure, only when you verify that the migrated database is readable.

The newly created folder does not have proper security Access Control Lists (ACL's) set on it, that are required by Spark to run. To set the necessary ACL's go to %ProgramData%\Lyftrondata\lib\tools\. Make a copy of the file FixHiveACL.bat, name it FixHiveACL-new.bat and edit the copy.

Change the lines inside the FixHiveACL-new.bat:

..\hadoop\hadoop-2.6.0\bin\winutils.exe chmod 777 ../../tmp/hive
..\hadoop\hadoop-2.6.0\bin\winutils.exe chmod 777 ../../warehouse

using our example destination path, into:

..\hadoop\hadoop-2.6.0\bin\winutils.exe chmod 777 d:/warehouse/tmp/hive
..\hadoop\hadoop-2.6.0\bin\winutils.exe chmod 777 d:/warehouse

and save the file.

Open an elevated command prompt, change the current directory to %ProgramData%\Lyftrondata\lib\tools\ and execute the FixHiveACL-new.bat. The script should execute without errors. If "access denied" error occurs, most probably you did not open command line with administrative privileges.

Step 4, copy files to the destination folder.

Copy the contents of %ProgramData%\Lyftrondata\warehouse into the destination folder (e.g.: d:\warehouse).

Step 5, connect to database server holding Spark Metastore

Connect to the database server that holds Spark Metastore database using, for example, a free SQL Server Management Studio. Any SQL Server client able to interactively issue queries to SQL Server will suffice.

You need to have the right to at least modify data in SDB and DBS tables.

Ensure that a recent backup of SparkMetastore database is available.

Step 6, modify file locations in Metabase

Using the SQL client (SSMS), edit and run the following SQL scripts.

List contents of DBS table and verify the location of database files:

SELECT [db_id]
     , [desc]
     , [db_location_uri]
     , [name]
     , [owner_name]
     , [owner_type]
  FROM [SparkMetastore].[dbo].[DBS]

Results of the query should look like shown below:

Edit the following script and change the value of @destinationFolderPathPart variable into the destination folder path (do not add the trailing slash '/'):

declare @destinationFolderPathPart varchar(240); declare @sourceFolderPathPart varchar(240);
 select @sourceFolderPathPart      = 'file:/c:/ProgramData/Lyftrondata/warehouse'
 select @destinationFolderPathPart = 'file:/c:/ProgramData/Lyftrondata/warehouse'; -- <<< change the destination folder path here

declare @qualifyingDbsEntriesBeforeUpdate int
declare @dbsEntriesAfterUpdate            int

-- get information about the count of database entries to modify
select @qualifyingDbsEntriesBeforeUpdate = count(*)
  from [SparkMetastore].[dbo].[DBS]
 where DB_LOCATION_URI like @sourceFolderPathPart + '%'
;

update [SparkMetastore].[dbo].[DBS]
   set DB_LOCATION_URI = replace(DB_LOCATION_URI, @sourceFolderPathPart, @destinationFolderPathPart)
 where DB_LOCATION_URI like @sourceFolderPathPart + '%'
;

select @dbsEntriesAfterUpdate = count(*)
  from [SparkMetastore].[dbo].[DBS]
 where DB_ID = 1
   and DB_LOCATION_URI like @sourceFolderPathPart + '%'
;

-- SDS

declare @qualifyingSdsEntriesBeforeUpdate int
declare @sdsEntriesAfterUpdate            int

select @qualifyingSdsEntriesBeforeUpdate = count(*)
  from [SparkMetastore].[dbo].[SDS]
 where [LOCATION] like @sourceFolderPathPart + '%'
;

update [SparkMetastore].[dbo].[SDS]
   set [LOCATION] = replace([LOCATION], @sourceFolderPathPart, @destinationFolderPathPart)
 where [LOCATION] like @sourceFolderPathPart + '%'
;

 select @sdsEntriesAfterUpdate = count(*)
   from [SparkMetastore].[dbo].[SDS]
  where [LOCATION] like @sourceFolderPathPart + '%'
;

set nocount on;
-- show execution statistics report
select 'DBS' as TableName
     , @qualifyingDbsEntriesBeforeUpdate as QualifyingEntries
     , @dbsEntriesAfterUpdate            as EntriesAfterUpdate
 union 
   all 
select 'SDS'
     , @qualifyingSdsEntriesBeforeUpdate
     , @sdsEntriesAfterUpdate
;
set nocount off;

The execution result should be similar to this:

(1 row affected)

(4632 rows affected)

TableName QualifyingEntries EntriesAfterUpdate
--------- ----------------- ------------------
DBS       1                 0
SDS       4632              0

The EntriesAfterUpdate column holds information about how many rows satisfy the update condition AFTER update - it should be zero.

Step 7, start Spark and verify that, data is readable

Using Lyftrondata Administrative Portal start Spark and verify that data from migrated database is available, for example, by executing queries in Lyftrondata.

Step 8, plan to delete the source folder if not needed

After the successful verification of migration, source files may be deleted, if not needed.

That concludes the Spark database migration procedure.