tomcat NIO connector fails #293
Comments
|
More backtrace: #0 0x00000000003be012 in cli_hlt () at /home/asias/src/cloudius-systems/osv/arch/x64/processor.hh:241 |
|
On Tue, May 6, 2014 at 5:01 AM, Asias He [email protected] wrote:
The only thing I can think of that can cause this (but this is just a wild It's been a while since I looked at this code, but looking now at By the way, it also seems that the epoll_file_closed() function, if called Asias, do you think that this workload really does epoll_wait() |
|
On Sun, May 11, 2014 at 7:35 PM, nyh [email protected] wrote:
I am not sure. Is there any way I can verify it? Asias |
|
On 05/12/2014 08:45 AM, Asias He wrote:
You can examine the type of the object: (gdb) set print object on If it says 'file', then Nadav is correct. How we actually get there can |
|
(gdb) f 20 |
|
Please reopen it since the commit does not really fix the issue. On Thu, Jul 3, 2014 at 2:35 PM, Avi Kivity [email protected] wrote:
Asias |
|
Reopened, while the patch is needed, it is not sufficient. |
|
Could it be related to #424 ? |
|
It may. Hard to see exactly how. What I am sure about is all the reference counting in that area is On 08/04/2014 09:16 PM, Tomasz Grabiec wrote:
|
|
''' ''' [backtrace] ''' (gdb) i th
|
|
On Mon, Aug 11, 2014 at 10:54 AM, Asias He [email protected] wrote:
|
|
On Mon, Aug 11, 2014 at 4:11 PM, nyh [email protected] wrote:
I am not testing this particular commit. I am testing if we have fixed the Asias |
|
On Mon, Aug 11, 2014 at 11:16 AM, Asias He [email protected] wrote:
Oh, sorry. I thought you did a bisection and came up with this comment :-) Nadav Har'El |
|
Sometimes I saw assert on A, sometimes on B. |
|
I added the following to verify that we reached Line X twice. |
|
I saved backtrace to fp->bt using backtrace_safe in the first time we hit line X. We are doing dup3 -> fdset -> fdrop in the first time. |
|
Aha. So while the first fdrop() is calling file::close() (which is We need to move the call to file::close() from fdrop() to fdset() and On 08/19/2014 11:03 AM, Asias He wrote:
|
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we move fp->stop_polls and fp->close() from fdrop() into in fdset() and fdclose(). This will eliminate the multipe call to fp->close(), as there can be only one call to fdclose() or fdset(). Also, we set f_count to INT_MIN using CAS, this way, we make sure rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we move fp->stop_polls and fp->close() from fdrop() into in fdset() and fdclose(). This will eliminate the multipe call to fp->close(), as there can be only one call to fdclose() or fdset(). Also, we set f_count to INT_MIN using CAS, this way, we make sure rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we move fp->stop_polls and fp->close() from fdrop() into in fdset() and fdclose(). This will eliminate the multipe call to fp->close(), as there can be only one call to fdclose() or fdset(). Also, we set f_count to INT_MIN using CAS, this way, we make sure rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we set f_count to INT_MIN using CAS, this way, we make sure fp->close() and rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we set f_count to INT_MIN using CAS, this way, we make sure fp->close() and rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
There is a race window between f_count reaches 0 and we set f_count to INT_MIN. This race would result in multiple call to fp->close() and rcu_dispose(fp) if we call them in fdrop(). To fix, we set f_count to INT_MIN using CAS, this way, we make sure fp->close() and rcu_dispose(fp) won't be called twice. Fixes cloudius-systems#293. Signed-off-by: Asias He <[email protected]>
Change tomcat to use NIO connector instead of BIO by editing:
$ vim osv/apps/tomcat/upstream/apache-tomcat-7.0.42/conf/server.xml
$ make image=tomcat
$ sudo make scripts/run.py -nv
May 06, 2014 12:14:36 AM org.apache.catalina.startup.Catalina start
INFO: Server startup in 1753 ms
pure virtual method called
terminate called without an active exception
Aborted
[backtrace]
0x417694 <__gnu_cxx::__verbose_terminate_handler()+340>
0x3bb0c2 <do_poll(std::vector<poll_file, std::allocator<poll_file> >&, int)+290>
0x3bc867 <epoll_wait+391>
0x100003209a82 <Java_sun_nio_ch_EPollArrayWrapper_epollWait+210>
0x200000590b70 <???+5835632>
0x200000598b1f <???+5868319>
0x4083f8e96f <???+-2080839313>
$ sh perform-one-test.sh
$ cat perform-one-test.sh
!/usr/bin/env bash
GUEST_IP=192.168.122.89
if [ -z "$GUEST_IP" ]; then
echo "Please set GUEST_IP"
exit 1
fi
TIMESTAMP=$(date +%Y%m%d_%H%M%S)
echo "=== Warmup ==="
wrk -t4 -c256 -d30s http://${GUEST_IP}:8081/servlet/json
echo "=== Main test ==="
wrk_out_file=wrk.${TIMESTAMP}
echo "Output goes to $wrk_out_file"
wrk --latency -t4 -c128 -d1m http://${GUEST_IP}:8081/servlet/json > ${wrk_out_file}
echo "=== Results ==="
./wrk-parse.py ${wrk_out_file}
The text was updated successfully, but these errors were encountered: