hycdong removed a comment on issue #577:
URL: 
https://github.com/apache/incubator-pegasus/issues/577#issuecomment-678071662


   # Apache Pegasus 2.1.0 Release Notes
   
   Apache Pegasus 2.1.0 is a feature release.
   
   ## Upgrading Notes
   
   - 2.1.0 is not compatible in storage format with previous versions. 
ReplicaServers upgraded to this version can't rollback to any previous releases.
   
   - 2.1.0 can only be upgraded from 2.0.0. ReplicaServers with version 1.x 
should upgrade to 2.0.0 firstly before upgrading to 2.1.0.
   
   ## New Features
   
   ### Bulk Load
   
   Bulk Load provides a new way to write massive data into a Pegasus table. 
Using bulk load ingestion eliminates the RocksDB compaction and improves the 
stability of the online Pegasus cluster. See this document for more details: 
https://pegasus.apache.org/2020/02/18/bulk-load-design.html
   
   ## Optimizations and improvements
   
   ### Write latency optimization
   
   - Added a new thread pool to execute shared-log write-callback. This 
optimization significantly reduces the write-latency spike caused by RocksDB, 
which usually resides in isolated disks from shared-log. It was because they 
shared the same thread pools that shared-log latency was negatively impacted by 
RocksDB compaction. Now while the latency spike still exists, it's reduced by 
about 5-times in our experiments. See 
[XiaoMi/rdsn#568](https://github.com/XiaoMi/rdsn/pull/568)
   
   - Pegasus ReplicaServer now supports the rate-limiting of RocksDB 
compaction. It prevents huge disk I/O spike and optimizes read/write latencies 
in write-intensive workload. See #543 
   
   ### Clear modifications to RocksDB
   
   Pegasus used a modified version of RocksDB, which contained some 
disk-incompatible changes. In this version, we reduced most of the changes we 
made, moving towards the goal where finally Pegasus could use the official 
RocksDB. See #535 #556 
   
   ## Fixed Issues
   
   - Fixed a memory leak in the network UDP module. Since UDP is by default 
used in failure detection between MetaServer and ReplicaServer, this issue 
caused the memory usage of MetaServer to increase over time.  See 
[XiaoMi/rdsn#477](https://github.com/XiaoMi/rdsn/pull/477)
   
   ### Known Issues
   
   - We plan to add an important fix (#587) in the 2.1.0 release, this pr is 
not included in this RC version. It will be added after it has been well 
tested. 
   - We plan to add a 2.0.0-compatible version only used for an emergency 
downgrade when upgrading to 2.1.0 meets serious problems.
   


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
[email protected]



---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to