[HBASE-14368] New TestWALLockup broken by addendum added to parent issue

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.0.0
• Component/s: test
• Hadoop Flags: Reviewed
• Assignee: Michael Stack
• Reporter: Michael Stack
• Created: 04/Sep/15 22:42
• Updated: 01/Jul/22 21:05
• Resolved: 05/Sep/15 15:54

Description

My second addendum broke TestWALLockup, the one that did this: https://issues.apache.org/jira/browse/HBASE-14317?focusedCommentId=14730301&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-14730301

diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/wal/FSHLog.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/wal/FSHLog.java
index 5708c30..c421f5c 100644
--- a/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/wal/FSHLog.java
+++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/regionserver/wal/FSHLog.java
@@ -878,8 +878,19 @@ public class FSHLog implements WAL {
         // Let the writer thread go regardless, whether error or not.
         if (zigzagLatch != null) {
           zigzagLatch.releaseSafePoint();
-          // It will be null if we failed our wait on safe point above.
-          if (syncFuture != null) blockOnSync(syncFuture);
+          // syncFuture will be null if we failed our wait on safe point above. Otherwise, if
+          // latch was obtained successfully, the sync we threw in either trigger the latch or it
+          // got stamped with an exception because the WAL was damaged and we could not sync. Now
+          // the write pipeline has been opened up again by releasing the safe point, process the
+          // syncFuture we got above. This is probably a noop but it may be stale exception from
+          // when old WAL was in place. Catch it if so.
+          if (syncFuture != null) {
+            try {
+              blockOnSync(syncFuture);
+            } catch (IOException ioe) {
+              if (LOG.isTraceEnabled()) LOG.trace("Stale sync exception", ioe);
+            }
+          }

It broke the test because the test hand feeds appends and syncs with when they should throw exceptions. In the test we manufactured the case where an append fails and we then asserted the following sync would fail.

Problem was that we expected the failure to be a dropped snapshot failure because fail of sync is a catastrophic event... but our hand feeding actually reproduced the case where a sync goes into the damaged file... before it had rolled... which is no longer a catastrophic event... we just catch and move on.

The attached patch just removes check for dropped snapshot and that abort was called.