Junfan Zhang
Junfan Zhang
When the data type in the DataFrame does not meet the requirements of the Analyzer(like `completeness`, `compliance`), the final analysis result may be empty or may not be empty. This...
**Check** has hint param, but **Analyzer** not. it's necessary to support it.
### What changes were proposed in this pull request? Support storing shuffle data to secured HDFS cluster by spark job user's own permission in shuffle server side. ### Why are...
The retry mechanism is introduced by #127. But in some access checker, there's no need to retry, like candidates checker. But the health checker maybe need. So I think we...
### What changes were proposed in this pull request? Introduce more grpc server metrics 1. grpc_server_executor_active_threads 2. grpc_server_executor_blocking_queue_size 3. grpc_server_connection_size ### Why are the changes needed? Now if the executors...
### Motivation Now uniffle's delegation shuffle manager support fallback to sort shuffle manager in app level. We could implement it in shuffle level.
### What changes were proposed in this pull request? Avoid starting unused threads in spark driver ### Why are the changes needed? In Spark RssShuffleManager, for the role of driver,...
### What changes were proposed in this pull request? Use NIO's Files API to replace FileInputStream/FileOutputStream in some paths. ### Why are the changes needed? Follow this PR of spark:...
### Motivation Now the executor only will use the single TCP connection with the specified shuffle server, so when multiple tasks are running concurrently, it will share this channel. Maybe...
Could the community provide the official benchmark with ESS and uniffle?