>>I still (highly)suspect that there is something wrong with the flush queue(some entry pushed into it can't be poll out). Ya I too have that suspect. May be any new logs may help to uncover the issue.
On Thu, Jun 5, 2014 at 11:06 AM, Stack <st...@duboce.net> wrote: > Always the same two regions that get stuck or does it vary? Another set of > example logs may help uncover the sequence of trouble-causing events. > > Thanks, > St.Ack > > > On Wed, Jun 4, 2014 at 7:31 PM, sunweiwei <su...@asiainfo-linkage.com> > wrote: > > > my log is similar as HBASE-10499. > > > > Thanks > > > > -----邮件原件----- > > 发件人: saint....@gmail.com [mailto:saint....@gmail.com] 代表 Stack > > 发送时间: 2014年6月3日 23:10 > > 收件人: Hbase-User > > 主题: Re: 答复: forcing flush not works > > > > Mind posting link to your log? Sounds like HBASE-10499 as Honghua says. > > St.Ack > > > > > > On Tue, Jun 3, 2014 at 2:34 AM, sunweiwei <su...@asiainfo-linkage.com> > > wrote: > > > > > Thanks. Maybe the same as HBase-10499. > > > I stop the regionserver then start it. Then hbase back to normal. > > > This is jstack log when 2 regions can not flush. > > > > > > "Thread-17" prio=10 tid=0x00007f6210383800 nid=0x6540 waiting on > > condition > > > [0x00007f61e0a26000] > > > java.lang.Thread.State: TIMED_WAITING (parking) > > > at sun.misc.Unsafe.park(Native Method) > > > - parking to wait for <0x000000041ae0e6b8> (a > > > java.util.concurrent. > > > locks.AbstractQueuedSynchronizer$ConditionObject) > > > at > > > java.util.concurrent.locks.LockSupport.parkNanos(LockSupport.java:196) > > > at > > > > > > > > > java.util.concurrent.locks.AbstractQueuedSynchronizer$ConditionObject.awaitN > > > anos(AbstractQueuedSynchronizer.java:2025) > > > at java.util.concurrent.DelayQueue.poll(DelayQueue.java:201) > > > at java.util.concurrent.DelayQueue.poll(DelayQueue.java:39) > > > at > > > > > > > > > org.apache.hadoop.hbase.regionserver.MemStoreFlusher$FlushHandler.run(MemSto > > > reFlusher.java:228) > > > at java.lang.Thread.run(Thread.java:662) > > > > > > -----邮件原件----- > > > 发件人: 冯宏华 [mailto:fenghong...@xiaomi.com] > > > 发送时间: 2014年6月3日 16:34 > > > 收件人: user@hbase.apache.org > > > 主题: 答复: forcing flush not works > > > > > > The same symptom as HBase-10499? > > > > > > I still (highly)suspect that there is something wrong with the flush > > > queue(some entry pushed into it can't be poll out). > > > ________________________________________ > > > 发件人: sunweiwei [su...@asiainfo-linkage.com] > > > 发送时间: 2014年6月3日 15:43 > > > 收件人: user@hbase.apache.org > > > 主题: forcing flush not works > > > > > > Hi > > > > > > > > > > > > I'm using a heavy-write hbase0.96 . I find this in regionserver log: > > > > > > 2014-06-03 15:13:19,445 INFO [regionserver60020.logRoller] wal.FSHLog: > > Too > > > many hlogs: logs=33, maxlogs=32; forcing flush of 3 regions(s): > > > 1a7dda3c3815c19970ace39fd99abfe8, aff81bc46aa7d3ed51a01f11f23c8320, > > > d5666e003f598147b4dda509f173a779 > > > > > > 2014-06-03 15:13:23,869 INFO [regionserver60020.logRoller] wal.FSHLog: > > Too > > > many hlogs: logs=34, maxlogs=32; forcing flush of 2 regions(s): > > > aff81bc46aa7d3ed51a01f11f23c8320, d5666e003f598147b4dda509f173a779 > > > > > > ┇ > > > > > > ┇ > > > > > > 2014-06-03 15:18:14,778 INFO [regionserver60020.logRoller] wal.FSHLog: > > Too > > > many hlogs: logs=93, maxlogs=32; forcing flush of 2 regions(s): > > > aff81bc46aa7d3ed51a01f11f23c8320, d5666e003f598147b4dda509f173a779 > > > > > > > > > > > > > > > > > > It seems like 2 regions can’t be flushed and WALs Dir continue to > > increase > > > and Then I find this in client log: > > > > > > INFO | AsyncProcess-waitForMaximumCurrentTasks [2014-06-03 15:30:53] - > : > > > Waiting for the global number of running tasks to be equals or less > than > > 0, > > > tasksSent=15819, tasksDone=15818, currentTasksDone=15818, > > > tableName=BT_D_BF001_201406 > > > > > > > > > > > > Then write speed will become very slow. > > > > > > After I flush 2 regions manually , write speed can back to normal > only > > > a > > > short while. > > > > > > > > > > > > Any suggestion will be appreciated. Thanks. > > > > > > > > > > > > > >
