[GitHub] [kudu] acelyc111 opened a new pull request, #50: KUDU-3371 [fs] Use RocksDB to store LBM metadata

["acelyc111 (via GitHub)" <gi...@apache.org>]

acelyc111 opened a new pull request, #50:
URL: https://github.com/apache/kudu/pull/50

   Since LogBlockContainer store block records sequentially in metadata file, the live blocks maybe in a very low ratio, and it cause disk space wasting and long time bootstrap.
   
   This patch use RocksDB to store LBM metadata, a new item will be Put() into RocksDB when a new block created in LBM, and the item will be Delete()d from RocksDB when the block removed from LBM. Data in RocksDB can be maintained in RocksDB itself, i.e. deleted items will be GCed so doesn't need rewriting as how we do it in current LBM.
   
   The implemention also reuse most logic of LBM, the main difference is store Block records metadata in RocksDB.
   1. Make LogBlockManager as a super class
   2. The former LBM that stores metadata in a append only file, is separeted from LBM and specified a new name LogfBlockManager. Its behavior has no change.
   3. Introduce a new class LogrBlockManager that stores metadata in RocksDB, the main idea: a. Create container Data file is created as before, metadata is stored in keys prefixed by the container's id, append the block id, e.g. <container_id>.<block_id>. Make sure there is no such keys in RocksDB before this container created. b. Open container Make sure the data file is healthy. c. Deconstruct container If the container is dead (full and no live blocks), remove the data file, and clean up keys prefixed by the container's id. d. Load container (by ProcessRecords()) Iterate the RocksDB in the key range [<container_id>, <next_container_id>), only live block records will be populated, we can use them as before. e. Create blocks in a container Put() serialized BlockRecordPB records into RocksDB in batch, keys are in form of '<container_id>.<block_id>' as mentioned above. f. Remove blocks from a container Contruct the keys by container's id and block's id, Delete() them from RocksDB in 
 batch.
   
   4. Some refactors, such as create and delete blocks in batch to reduce lock consult times.
   
   This patch contains the following changes:
   - Adds a new block manager type named 'logr', it use RocksDB to store LBM metadata, it is also specified by flag '--block_manager'.
   - block_manager-test supports to test LogrBlockManager
   - block_manager-stress-test supports to test LogrBlockManager
   - log_block_manager-test supports to test LogrBlockManager
   - tablet_server-test supports to test LogrBlockManager
   - dense_node-itest supports to test LogrBlockManager
   - kudu-tool-test supports to test LogrBlockManager
   
   It's optional to use RocksDB, we can use the former LBM as before, we can introduce more tools to convert data between the two implemention in the future.
   
   The optimization is obvious as shown in JIRA KUDU-3371, it shows that reopen staged reduced upto 90% time cost.
   
   Change-Id: Ie72f6914eb5653a9c034766c6cd3741a8340711f


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: dev-unsubscribe@kudu.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org

[GitHub] [kudu] acelyc111 closed pull request #50: KUDU-3371 [fs] Use RocksDB to store LBM metadata

["acelyc111 (via GitHub)" <gi...@apache.org>]

acelyc111 closed pull request #50: KUDU-3371 [fs] Use RocksDB to store LBM metadata
URL: https://github.com/apache/kudu/pull/50


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: dev-unsubscribe@kudu.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org