aajisaka commented on a change in pull request #3065:
URL: https://github.com/apache/hadoop/pull/3065#discussion_r647396463
##########
File path:
hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/namenode/FSNamesystem.java
##########
@@ -1996,7 +1996,12 @@ private void metaSave(PrintWriter out) {
LightWeightHashSet<Long> openFileIds = new LightWeightHashSet<>();
for (DatanodeDescriptor dataNode :
blockManager.getDatanodeManager().getDatanodes()) {
- for (long ucFileId : dataNode.getLeavingServiceStatus().getOpenFiles()) {
+ // Sort open files
+ LightWeightHashSet<Long> dnOpenFiles =
+ dataNode.getLeavingServiceStatus().getOpenFiles();
+ Long[] dnOpenFileIds = new Long[dnOpenFiles.size()];
+ Arrays.sort(dnOpenFiles.toArray(dnOpenFileIds));
+ for (Long ucFileId : dnOpenFileIds) {
INode ucFile = getFSDirectory().getInode(ucFileId);
if (ucFile == null || ucFileId <= prevId ||
openFileIds.contains(ucFileId)) {
Review comment:
Agreed. It must be sorted.
I think HDFS-11847 has a bug. If the DataNodes have the following open files
and we want to list all the open files:
DN1: [1001, 1002, 1003, ... , 2000]
DN2: [1, 2, 3, ... , 1000]
At first `getFilesBlockingDecom(0, "/")` is called and it returns [1001,
1002, ... , 2000] because it reached max size (=1000), and next
`getFilesBlockingDecom(2000, "/")` is called because the last inode Id of the
previous result is 2000. That way the open files of DN2 is missed
--
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:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org
