[GitHub] [hbase] ddupg commented on a diff in pull request #4756: HBASE-27354 EOF thrown by WALEntryStream causes replication blocking

[GitBox <gi...@apache.org>]

ddupg commented on code in PR #4756:
URL: https://github.com/apache/hbase/pull/4756#discussion_r962458619


##########
hbase-server/src/main/java/org/apache/hadoop/hbase/replication/regionserver/WALEntryStream.java:
##########
@@ -255,15 +255,21 @@ private void dequeueCurrentLog() throws IOException {
    * Returns whether the file is opened for writing.
    */
   private boolean readNextEntryAndRecordReaderPosition() throws IOException {
+    long prePos = reader.getPosition();

Review Comment:
   Thanks for reviewing.
   I haven't found out yet that we read an incomplete entry. EOFException is thrown by `org.apache.hadoop.fs.FSDataInputStream#seek`.
   After reading the uncommitted data, we will reopen `FSDataInputStream` and seek back. If the position where seek back is the sync length, it may cause EOF. As the 3 copies of the hdfs block being written are inconsistent, even if we can read the data from a specific position in one copy, it may not be read in the other one.
   It's not a big problem if it just throws EOF as we'll retry. The big problem is that if we have read some entries into the `WALEntryBatch` and increased the [`totalBufferUsed`](https://github.com/apache/hbase/blob/308cd729d23329e6d8d4b9c17a645180374b5962/hbase-server/src/main/java/org/apache/hadoop/hbase/replication/regionserver/ReplicationSourceWALReader.java#L78), and the `totalBufferUsed` is not subtracted after throwing EOF, all peers will eventually block completely.
   



-- 
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.

To unsubscribe, e-mail: issues-unsubscribe@hbase.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org