### What changes were proposed in this pull request? Update default value of `celeborn.<module>.io.mode` to whether epoll mode is available. Meanwhile, the io mode of transport is `NIO` for unavailable epoll mode. ### Why are the changes needed? The JDK NIO bug produces the situation that empty polling of `Selector` could cause CPU 100%, which refers to 1. [JDK-2147719 : (se) Selector doesn't block on Selector.select(timeout) (lnx)](https://bugs.java.com/bugdatabase/view_bug.do?bug_id=2147719) 2. [JDK-6403933 : (se) Selector doesn't block on Selector.select(timeout) (lnx)](https://bugs.java.com/bugdatabase/view_bug.do?bug_id=6403933) When the epoll mode is available, the default IO mode should be `EPOLL`, which backports [NettyServer.java#L92](https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/io/network/netty/NettyServer.java#L92). Meanwhile, the transport io mode should be `NIO` when the epoll mode is unavailable. ### Does this PR introduce _any_ user-facing change? Change the default value of `celeborn.<module>.io.mode`. ### How was this patch tested? CI. Closes #2994 from SteNicholas/CELEBORN-1774. Authored-by: SteNicholas <programgeek@163.com> Signed-off-by: Shuang <lvshuang.xjs@alibaba-inc.com> |
||
|---|---|---|
| .. | ||
| client.md | ||
| columnar-shuffle.md | ||
| ha.md | ||
| index.md | ||
| master.md | ||
| metrics.md | ||
| network-module.md | ||
| network.md | ||
| quota.md | ||
| worker.md | ||