This repository has been archived by the owner on Jan 13, 2022. It is now read-only.
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Summary: This patch enables bufferedIO for hadoop daemons, and sets the buffer size to 4096, which should be slightly below 20 lines (the average size in dfs1 is about 250bytes/line). So obvoiusly we will not see synchronously those last 20 lines, but I don't think it's a huge issue. On the other hand, my experiments show, that with this settings we can speed up logging by a factor of 2. I also added a hook which when the node is shutdown with hadoopctl hadoop-daemon.sh stop - it will close the log manager which flushes the contents of the buffer, so we won't miss those messages. Of course if we kill a node with -9, we will loose up to those 20 lines. -limited the classname to one from previous 2. The logs on dfs1 are huge, we need to make a pass I think over messages and limit these. In the last few hours it's like 5GB per hour. Test Plan: manually local setup, validated that log is flushed at normal shutdown Reviewers: hkuang Reviewed By: hkuang
- Loading branch information