[ https://issues.apache.org/jira/browse/HDFS-8791?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15035317#comment-15035317 ]
Joep Rottinghuis commented on HDFS-8791: ---------------------------------------- Totally agree that if the patch goes into 2.6.3 it should go into 2.7.2 as well. While I appreciate the sentiment that layout format changes would normally warrant a new minor release (2.8.0 in this case?), this approach leaves us with a dilemma. We feel that we cannot move from 2.4 to 2.6 despite all if the efforts to validate and test without this fix. Luckily we're in the position that we roll out own internal build, so technically we're not blocked on this. We're already happy this fix will go in upstream. That said, it would block us from rolling cleanly to 2.7.2+ without manually applying this patch. Similarly, what do we tell other users? Don't use 2.6.3 or 2.6.4 because it has a fundamental perf problem? Then why even do a 2.6.3 maintenance release? Isn't the point of these releases that you can avoid trade-off beteren manually applying a list of patches on top of a release? Similarly, do we tell the HBase community to not use this version of Hadoop and just wait for a 2.8.x release and perhaps longer until that has been stabilized to the point where folks can run that comfortably in production knowing it has been battle tested? A format release between dot releases is perhaps not ideal either, but if a release manager is willing to pull it in, and coordinate with the release of an equivalent maintenance release with a newer minor version, then is that not a practical workable outcome? Sent from my iPhone > block ID-based DN storage layout can be very slow for datanode on ext4 > ---------------------------------------------------------------------- > > Key: HDFS-8791 > URL: https://issues.apache.org/jira/browse/HDFS-8791 > Project: Hadoop HDFS > Issue Type: Bug > Components: datanode > Affects Versions: 2.6.0, 2.8.0, 2.7.1 > Reporter: Nathan Roberts > Assignee: Chris Trezzo > Priority: Critical > Attachments: 32x32DatanodeLayoutTesting-v1.pdf, > 32x32DatanodeLayoutTesting-v2.pdf, HDFS-8791-trunk-v1.patch, > HDFS-8791-trunk-v2-bin.patch, HDFS-8791-trunk-v2.patch, > HDFS-8791-trunk-v2.patch, hadoop-56-layout-datanode-dir.tgz > > > We are seeing cases where the new directory layout causes the datanode to > basically cause the disks to seek for 10s of minutes. This can be when the > datanode is running du, and it can also be when it is performing a > checkDirs(). Both of these operations currently scan all directories in the > block pool and that's very expensive in the new layout. > The new layout creates 256 subdirs, each with 256 subdirs. Essentially 64K > leaf directories where block files are placed. > So, what we have on disk is: > - 256 inodes for the first level directories > - 256 directory blocks for the first level directories > - 256*256 inodes for the second level directories > - 256*256 directory blocks for the second level directories > - Then the inodes and blocks to store the the HDFS blocks themselves. > The main problem is the 256*256 directory blocks. > inodes and dentries will be cached by linux and one can configure how likely > the system is to prune those entries (vfs_cache_pressure). However, ext4 > relies on the buffer cache to cache the directory blocks and I'm not aware of > any way to tell linux to favor buffer cache pages (even if it did I'm not > sure I would want it to in general). > Also, ext4 tries hard to spread directories evenly across the entire volume, > this basically means the 64K directory blocks are probably randomly spread > across the entire disk. A du type scan will look at directories one at a > time, so the ioscheduler can't optimize the corresponding seeks, meaning the > seeks will be random and far. > In a system I was using to diagnose this, I had 60K blocks. A DU when things > are hot is less than 1 second. When things are cold, about 20 minutes. > How do things get cold? > - A large set of tasks run on the node. This pushes almost all of the buffer > cache out, causing the next DU to hit this situation. We are seeing cases > where a large job can cause a seek storm across the entire cluster. > Why didn't the previous layout see this? > - It might have but it wasn't nearly as pronounced. The previous layout would > be a few hundred directory blocks. Even when completely cold, these would > only take a few a hundred seeks which would mean single digit seconds. > - With only a few hundred directories, the odds of the directory blocks > getting modified is quite high, this keeps those blocks hot and much less > likely to be evicted. -- This message was sent by Atlassian JIRA (v6.3.4#6332)