[Bro] Bro's limitations with high worker count and memory exhaustion
Baxter Milliwew
baxter.milliwew at gmail.com
Wed Jul 1 13:50:04 PDT 2015
No. The cluster is six 48-core/64GB servers with the manager as one and 3
proxies and 28 workers for the others. I enabled profiling but didn't see
anything wrong. The bro process that is consuming all memory is not the
same process detailed by prof.log.
On Wed, Jul 1, 2015 at 12:27 PM, Jan Grashofer <jan.grashofer at cern.ch>
wrote:
> You are not trying to run 140 workers on a single machine with 64GB
> memory, right?
>
>
> ------------------------------
> *From:* Baxter Milliwew [baxter.milliwew at gmail.com]
> *Sent:* Wednesday, July 01, 2015 20:39
> *To:* Siwek, Jon
> *Cc:* Jan Grashofer; bro at bro.org
> *Subject:* Re: [Bro] Bro's limitations with high worker count and memory
> exhaustion
>
> Do you think a high worker count with the current implementation of
> select() would cause high memory usage ?
>
> I'm trying to figure out why the manager always exhausts all memory:
>
> top - 18:36:13 up 1 day, 14:42, 1 user, load average: 12.67, 10.83,
> 10.95
>
> Tasks: 606 total, 5 running, 601 sleeping, 0 stopped, 0 zombie
>
> %Cpu(s): 15.3 us, 6.4 sy, 1.3 ni, 76.3 id, 0.0 wa, 0.0 hi, 0.7 si,
> 0.0 st
>
> KiB Mem: 65939412 total, 65251768 used, 687644 free, 43248 buffers
>
> KiB Swap: 67076092 total, 54857880 used, 12218212 free. 4297048 cached
> Mem
>
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+
> COMMAND
>
>
>
> 35046 logstash 20 0 10.320g 511600 3784 S 782.1 0.8 5386:34 java
>
>
>
>
> 9925 bro 25 5 97.504g 0.045t 1508 R 99.7 73.8 814:58.88 bro
>
>
>
>
> 9906 bro 20 0 22.140g 3.388g 3784 S 73.2 5.4 1899:18 bro
>
>
>
>
> 2509 root 20 0 308440 44064 784 R 48.5 0.1 1029:56
> redis-server
>
>
>
> 2688 bro 30 10 4604 1440 1144 R 44.8 0.0 0:00.49 gzip
>
>
>
>
> 180 root 20 0 0 0 0 S 8.2 0.0 4:26.54
> ksoftirqd/8
>
>
>
> 2419 debug 20 0 25376 3564 2600 R 7.3 0.0 0:00.76 top
>
>
>
>
> 2689 logstash 20 0 8 4 0 R 5.5 0.0 0:00.06 bro
>
>
>
> On Tue, Jun 30, 2015 at 11:37 AM, Baxter Milliwew <
> baxter.milliwew at gmail.com> wrote:
>
>> Thanks. Some limited reading says it's not possible to increase
>> FD_SETSIZE on linux and it's time to migrate to poll().
>>
>>
>>
>> On Tue, Jun 30, 2015 at 7:44 AM, Siwek, Jon <jsiwek at illinois.edu> wrote:
>>
>>> A guess is that you’re bumping into an FD_SETSIZE limit — the way remote
>>> I/O is currently structured has at least 5 file descriptors per remote
>>> connection from what I can see at a glance (a pair of pipes, 2 fds each,
>>> for signaling read/write readiness related to ChunkedIO and one fd for the
>>> actual socket). Typically, FD_SETSIZE is 1024, so with ~150-200 remote
>>> connections and 5 fds per connection plus whatever other descriptors Bro
>>> may need to have open (e.g. for file I/O), it seems reasonable to guess
>>> that’s the problem. But you could easily verify w/ some code modifications
>>> to check whether the FD_SET call is using a fd >= FD_SETSIZE.
>>>
>>> Other than making involved code changes to Bro (e.g. to move away from
>>> select() for I/O event handling), the only suggestions I have are 1)
>>> reducing number of remote connections 2) see if you can increase FD_SETSIZE
>>> via preprocessor stuff or CFLAGS/CXXFLAGS upon ./configure’ing (I’ve never
>>> done this myself to know if it works, but I’ve googled around before and
>>> think the implication was that it may work on Linux).
>>>
>>> - Jon
>>>
>>> > On Jun 29, 2015, at 6:22 PM, Baxter Milliwew <
>>> baxter.milliwew at gmail.com> wrote:
>>> >
>>> > The manager still crashes. Interesting note about a buffer overflow.
>>> >
>>> >
>>> > [manager]
>>> >
>>> > Bro 2.4
>>> > Linux 3.16.0-38-generic
>>> >
>>> > core
>>> > [New LWP 18834]
>>> > [Thread debugging using libthread_db enabled]
>>> > Using host libthread_db library
>>> "/lib/x86_64-linux-gnu/libthread_db.so.1".
>>> > Core was generated by `/usr/local/3rd-party/bro/bin/bro -U .status -p
>>> broctl -p broctl-live -p local -'.
>>> > Program terminated with signal SIGABRT, Aborted.
>>> > #0 0x00007f163bb46cc9 in __GI_raise (sig=sig at entry=6) at
>>> ../nptl/sysdeps/unix/sysv/linux/raise.c:56
>>> >
>>> > Thread 1 (Thread 0x............ (LWP 18834)):
>>> > #0 0x00007f163bb46cc9 in __GI_raise (sig=sig at entry=6) at
>>> ../nptl/sysdeps/unix/sysv/linux/raise.c:56
>>> > #1 0x00007f163bb4a0d8 in __GI_abort () at abort.c:89
>>> > #2 0x00007f163bb83394 in __libc_message (do_abort=do_abort at entry=2,
>>> fmt=fmt at entry=0x............ "*** %s ***: %s terminated\n") at
>>> ../sysdeps/posix/libc_fatal.c:175
>>> > #3 0x00007f163bc1ac9c in __GI___fortify_fail (msg=<optimized out>,
>>> msg at entry=0x............ "buffer overflow detected") at
>>> fortify_fail.c:37
>>> > #4 0x00007f163bc19b60 in __GI___chk_fail () at chk_fail.c:28
>>> > #5 0x00007f163bc1abe7 in __fdelt_chk (d=<optimized out>) at
>>> fdelt_chk.c:25
>>> > #6 0x00000000005e962a in Set (set=0x............,
>>> this=0x............) at /home/bro/Bro-IDS/bro-2.4/src/iosource/FD_Set.h:59
>>> > #7 SocketComm::Run (this=0x............) at
>>> /home/bro/Bro-IDS/bro-2.4/src/RemoteSerializer.cc:3406
>>> > #8 0x00000000005e9c31 in RemoteSerializer::Fork (this=0x............)
>>> at /home/bro/Bro-IDS/bro-2.4/src/RemoteSerializer.cc:687
>>> > #9 0x00000000005e9d4f in RemoteSerializer::Enable
>>> (this=0x............) at
>>> /home/bro/Bro-IDS/bro-2.4/src/RemoteSerializer.cc:575
>>> > #10 0x00000000005b6943 in BifFunc::bro_enable_communication
>>> (frame=<optimized out>, BiF_ARGS=<optimized out>) at bro.bif:4480
>>> > #11 0x00000000005b431d in BuiltinFunc::Call (this=0x............,
>>> args=0x............, parent=0x............) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Func.cc:586
>>> > #12 0x0000000000599066 in CallExpr::Eval (this=0x............,
>>> f=0x............) at /home/bro/Bro-IDS/bro-2.4/src/Expr.cc:4544
>>> > #13 0x000000000060ceb4 in ExprStmt::Exec (this=0x............,
>>> f=0x............, flow=@0x............: FLOW_NEXT) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Stmt.cc:352
>>> > #14 0x000000000060b174 in IfStmt::DoExec (this=0x............,
>>> f=0x............, v=<optimized out>, flow=@0x............: FLOW_NEXT) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Stmt.cc:456
>>> > #15 0x000000000060ced1 in ExprStmt::Exec (this=0x............,
>>> f=0x............, flow=@0x............: FLOW_NEXT) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Stmt.cc:356
>>> > #16 0x000000000060b211 in StmtList::Exec (this=0x............,
>>> f=0x............, flow=@0x............: FLOW_NEXT) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Stmt.cc:1696
>>> > #17 0x000000000060b211 in StmtList::Exec (this=0x............,
>>> f=0x............, flow=@0x............: FLOW_NEXT) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Stmt.cc:1696
>>> > #18 0x00000000005c042e in BroFunc::Call (this=0x............,
>>> args=<optimized out>, parent=0x0) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Func.cc:403
>>> > #19 0x000000000057ee2a in EventHandler::Call (this=0x............,
>>> vl=0x............, no_remote=no_remote at entry=false) at
>>> /home/bro/Bro-IDS/bro-2.4/src/EventHandler.cc:130
>>> > #20 0x000000000057e035 in Dispatch (no_remote=false,
>>> this=0x............) at /home/bro/Bro-IDS/bro-2.4/src/Event.h:50
>>> > #21 EventMgr::Dispatch (this=this at entry=0x...... <mgr>) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Event.cc:111
>>> > #22 0x000000000057e1d0 in EventMgr::Drain (this=0xbbd720 <mgr>) at
>>> /home/bro/Bro-IDS/bro-2.4/src/Event.cc:128
>>> > #23 0x00000000005300ed in main (argc=<optimized out>, argv=<optimized
>>> out>) at /home/bro/Bro-IDS/bro-2.4/src/main.cc:1147
>>> >
>>> >
>>> >
>>> > On Mon, Jun 29, 2015 at 4:09 PM, Baxter Milliwew <
>>> baxter.milliwew at gmail.com> wrote:
>>> > Nevermind... new box, default nofile limits. Thanks for the malloc
>>> tip.
>>> >
>>> >
>>> > On Mon, Jun 29, 2015 at 4:03 PM, Baxter Milliwew <
>>> baxter.milliwew at gmail.com> wrote:
>>> > Switching to jemalloc fixed the stability issue but not the worker
>>> count limitation.
>>> >
>>> > On Sun, Jun 28, 2015 at 7:18 PM, Baxter Milliwew <
>>> baxter.milliwew at gmail.com> wrote:
>>> > Looks like malloc from glibc, default on Ubuntu. I will try jemalloc
>>> and others.
>>> >
>>> >
>>> >
>>> > On Sun, Jun 28, 2015 at 1:03 AM, Jan Grashofer <jan.grashofer at cern.ch>
>>> wrote:
>>> > I experienced similar problems (memory gets eaten up quickly and
>>> workers crash with segfault) using tcmalloc. Which malloc do you use?
>>> >
>>> >
>>> > Regards,
>>> >
>>> > Jan
>>> >
>>> >
>>> > From: bro-bounces at bro.org [bro-bounces at bro.org] on behalf of Baxter
>>> Milliwew [baxter.milliwew at gmail.com]
>>> > Sent: Friday, June 26, 2015 23:03
>>> > To: bro at bro.org
>>> > Subject: [Bro] Bro's limitations with high worker count and memory
>>> exhaustion
>>> >
>>> > There's some sort of association between memory exhaustion and a high
>>> number of workers. The poor man's fix would be to purchase new servers
>>> with higher CPU speeds as that would reduce the worker count. Issues with
>>> high worker count and/or memory exhaustion appears to be a well know
>>> problem based on the mailing list archives.
>>> >
>>> > In the current version of bro-2.4 my previous configuration
>>> immediately causes the manager to crash: 15 proxies, 155 workers. To
>>> resolve this I've lowered the count to 10 proxies and 140 workers. However
>>> even with this configuration the manager process will exhaust all memory
>>> and crash within about 2 hours.
>>> >
>>> > The manager is threaded; I think this is an issue with the threading
>>> behavior between manager, proxies, and workers. Debugging threading
>>> problems is complex and I'm a complete novice.. my current tutorial is
>>> using information from a stack overflow thread:
>>> >
>>> >
>>> http://stackoverflow.com/questions/981011/c-programming-debugging-with-pthreads
>>> >
>>> > Does anyone else have this problem ? What have you tried and what do
>>> you suggest ?
>>> >
>>> > Thanks
>>> >
>>> >
>>> >
>>> >
>>> > 1435347409.458185 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] peer sent class "control"
>>> > 1435347409.458185 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] phase: handshake
>>> > 1435347409.661085 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] request for unknown event save_results
>>> > 1435347409.661085 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] registered for event
>>> Control::peer_status_response
>>> > 1435347409.694858 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] peer does not support 64bit PIDs;
>>> using compatibility mode
>>> > 1435347409.694858 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] peer is a Broccoli
>>> > 1435347409.694858 worker-2-18 parent - - -
>>> info [#10000/10.1.1.1:36994] phase: running
>>> >
>>> >
>>> >
>>> >
>>> >
>>> > _______________________________________________
>>> > Bro mailing list
>>> > bro at bro-ids.org
>>> > http://mailman.ICSI.Berkeley.EDU/mailman/listinfo/bro
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ICSI.Berkeley.EDU/pipermail/bro/attachments/20150701/15072ed2/attachment-0001.html
More information about the Bro
mailing list