tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Charles Richard <>
Subject Re: Tomcat thread dump analysis
Date Wed, 08 May 2013 17:57:15 GMT
On Wed, May 8, 2013 at 2:33 PM, Christopher Schultz <> wrote:

> Hash: SHA256
> Chuck,
> On 5/8/13 11:25 AM, Caldarale, Charles R wrote:
> >> From: Charles Richard []
> >> Subject: Re: Tomcat thread dump analysis
> >
> >> Here is a full thread dump
> >
> > Which again shows no Tomcat involvement in the locking hang.
> +1
> At least it confirms OP is actually running Tomcat, which wasn't
> evident from the first stack trace.
> Charles, please understand that what you have presented is a stack
> trace, not a thread dump: the difference is that a thread dump is a
> stack trace for every live thread in the JVM (which can show
> deadlocks) and you have provided only a single stack trace which
> cannot show a deadlock (it can only show that the thread is waiting on
> some condition).
> To demonstrate deadlock, you need to be able to see the condition
> being awaited in one thread, a lock held in another thread, and then
> the opposite scenario in another thread (or a single thread running
> into its own locks). Your stack trace does not show that the thread
> already holds the lock being awaited at the top of the stack trace.
> I appreciate the friendly feedback! How do I show a lock? I don't see any
threads that have a "BLOCKED" status. I do get this when I do a grep:

[root@web01 stacks]# grep locked tomcat1_20130507_14\:38.stack | wc -l

My connection pool is 150 and I do see 150 locks in the following type
scenario which has a "locked":

"TP-Processor416" daemon prio=10 tid=0x00002aff2939a800 nid=0x5ae7 in
Object.wait() [0x00002aff632ae000]
   java.lang.Thread.State: WAITING (on object monitor)
        at java.lang.Object.wait(Native Method)
        - waiting on <0x00002afe240fbc58> (a
        - locked <0x00002afe240fbc58> (a

> It is entirely possible that both Spring and Terrecotta attempt to
> synchronize on the same object (HttpSession) and that the thread is
> deadlocking on itself. If this is trivially-repeatable every time
> Terrecotta attempts to do whatever it is doing (hard to tell...
> Hibernate is executing a SQL query which ... uses the session?!?),
> then perhaps you have identified the problem and the Spring
> configuration to synchronize on a session attribute rather than the
> session itself will fix the problem.
> On the other hand, if the problem is different and yet you have
> arrived at your "Spring + Terrecotta = deadlock" conclusion due to
> treating Google as a consultant, then you may be barking up the wrong
> tree.
> Sorry: we like to get enough facts before we tell you a) what the
> problem is and b) how to fix it rather than just saying "I Googled for
> 'terra cotta deadlock' and someone said that if you use Spring, you
> get deadlocks" and telling you to go away. Instead, we'd rather help
> you fix your actual problem and educate you at the same time.
> I agree. At times desperate times call for desperate actions when I've
used all existing knowledge, Google can give me ideas of where I could
possibly look next. Given the choice, I would rather be educated on how to
diagnose this issue properly.

> It would appear that logic in your application threads has either
> > created a deadlock, or failed to unlock something before
> > returning,
> That's a tall order unless native code is involved, I believe. Though
> I don't have much experience with the java.lang.concurrent package...
> > or something else entirely (e.g, broken network causing grief with
> > Terracotta's distributed caching).  If it's a deadlock, you need
> > to examine the entire thread dump (all threads) and look for a
> > call stack that isn't like the others but holds the lock everyone
> > else is waiting for.  If it's simply a failure to unlock somewhere,
> > you will likely have to do a thorough code review to find it.  You
> > will most likely need Terracotta's help here, since it's their code
> > waiting on the lock.
> +1
> Start with finding out what Terrecotta is trying to lock on: if it's
> not the session, then you know you're entirely wrong about the cause.
> Also, knowing whether the thread is deadlocking itself versus another
> thread deadlocking with it is a very important piece of information to
> have.
> - -chris
> Version: GnuPG/MacGPG2 v2.0.17 (Darwin)
> Comment: GPGTools -
> Comment: Using GnuPG with Thunderbird -
> yqgyB0JOkb9pZ42P2bVovJJtO5A15tWw8NigEbKc+CwJMv9rgOzqRY7y30oih9mK
> yhu5C/sEFDPlPPzt0H5CT3YFqVOoJMpstmgFBA0FBLKEaqmzdc61WIohQqJmv2J6
> 2z3vNvmH1xBucEqgOrOEn8mp9rbaOHVmQtf6Zh1QaxhBzA0CREnnwz0FvFEhklGx
> rlpwkQafjcKjCCP6eaiurtZJtv4S0hJQJSOmLQKBdm3DY/dRokeN/ROnxn3DpqfX
> 0WwONHAw/OY0MzCJ6M9cZicX9XhNDNeRUzYlkORQpeAloM5w7Tp9q5fKo3iabFcC
> NwuWN5O5H5TZ2MSH7UHV9Kf0EA1Zk7Howi2Ea2zV4EnzutdgM5yxq//ow8LTcOdi
> ED376ckgSMgfV/b8u2tfJ1M9bD5WjD8v+kVbu+ErYzuM6i28iAGN7wrfgO+/Zn2S
> BuQgcEq/3usUXwnQ5IBAYpVF5DAKv/GeAaV+C8Ft+mb+jgo00Hb7Eqn5eY/w6CL/
> 1DM0mgmfcHTlSP4JK5noV73Eip2LmxdPwDFvNBkIiRPq5Av+FSIfIKclJu/piMQ/
> =3SsJ
> ---------------------------------------------------------------------
> To unsubscribe, e-mail:
> For additional commands, e-mail:

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message