11.0 stuck on high network load
Julien Charbon
jch at freebsd.org
Mon Sep 26 08:51:21 UTC 2016
On 9/25/16 8:58 PM, Slawa Olhovchenkov wrote:
> On Fri, Sep 23, 2016 at 10:16:56PM +0300, Slawa Olhovchenkov wrote:
>
>> On Thu, Sep 22, 2016 at 01:20:45PM +0300, Slawa Olhovchenkov wrote:
>>
>>> On Thu, Sep 22, 2016 at 12:04:40PM +0200, Julien Charbon wrote:
>>>
>>>>>> These paths can indeed compete for the same INP lock, as both
>>>>>> tcp_tw_2msl_scan() calls always start with the first inp found in
>>>>>> twq_2msl list. But in both cases, this first inp should be quickly used
>>>>>> and its lock released anyway, thus that could explain your situation it
>>>>>> that the TCP stack is doing that all the time, for example:
>>>>>>
>>>>>> - Let say that you are running out completely and constantly of tcptw,
>>>>>> and then all connections transitioning to TIME_WAIT state are competing
>>>>>> with the TIME_WAIT timeout scan that tries to free all the expired
>>>>>> tcptw. If the stack is doing that all the time, it can appear like
>>>>>> "live" locked.
>>>>>>
>>>>>> This is just an hypothesis and as usual might be a red herring.
>>>>>> Anyway, could you run:
>>>>>>
>>>>>> $ vmstat -z | head -2; vmstat -z | grep -E 'tcp|sock'
>>>>>
>>>>> ITEM SIZE LIMIT USED FREE REQ FAIL SLEEP
>>>>>
>>>>> socket: 864, 4192664, 18604, 25348,49276158, 0, 0
>>>>> tcp_inpcb: 464, 4192664, 34226, 18702,49250593, 0, 0
>>>>> tcpcb: 1040, 4192665, 18424, 18953,49250593, 0, 0
>>>>> tcptw: 88, 16425, 15802, 623,14526919, 8, 0
>>>>> tcpreass: 40, 32800, 15, 2285, 632381, 0, 0
>>>>>
>>>>> In normal case tcptw is about 16425/600/900
>>>>>
>>>>> And after `sysctl -a | grep tcp` system stuck on serial console and I am reset it.
>>>>>
>>>>>> Ideally, once when everything is ok, and once when you have the issue
>>>>>> to see the differences (if any).
>>>>>>
>>>>>> If it appears your are quite low in tcptw, and if you have enough
>>>>>> memory, could you try increase the tcptw limit using sysctl
>>>>>
>>>>> I think this is not eliminate stuck, just may do it less frequency
>>>>
>>>> You are right, it would just be a big hint that the tcp_tw_2msl_scan()
>>>> contention hypothesis is the right one. As I see you have plenty of
>>>> memory on your server, thus could you try with:
>>>>
>>>> net.inet.tcp.maxtcptw=4192665
>>>>
>>>> And see what happen. Just to validate this hypothesis.
>>>
>>> This is bad way for validate, with maxtcptw=16384 happened is random
>>> and can be waited for month. After maxtcptw=4192665 I am don't know
>>> how long need to wait for verification this hypothesis.
>>>
>>> More frequency (may be 3-5 times per day) happening less traffic drops
>>> (not to zero for minutes). May be this caused also by contention in
>>> tcp_tw_2msl_scan, but fast resolved (stochastic process). By eating
>>> CPU power nginx can't service connection and clients closed
>>> connections and need more TIME_WAIT and can trigered
>>> tcp_tw_2msl_scan(reuse=1). After this we can got live lock.
>>>
>>> May be after I learning to catch and dignostic this validation is more
>>> accurately.
>>
>> Some more bits:
>>
>> socket: 864, 4192664, 30806, 790,28524160, 0, 0
>> ipq: 56, 32802, 0, 1278, 1022, 0, 0
>> udp_inpcb: 464, 4192664, 44, 364, 14066, 0, 0
>> udpcb: 32, 4192750, 44, 3081, 14066, 0, 0
>> tcp_inpcb: 464, 4192664, 38558, 378,28476709, 0, 0
>> tcpcb: 1040, 4192665, 30690, 738,28476709, 0, 0
>> tcptw: 88, 32805, 7868, 772, 8412249, 0, 0
>>
>> last pid: 49575; load averages: 2.00, 2.05, 3.75 up 1+01:12:08 22:13:42
>> 853 processes: 15 running, 769 sleeping, 35 waiting, 34 lock
>> CPU 0: 0.0% user, 0.0% nice, 0.0% system, 100% interrupt, 0.0% idle
>> CPU 1: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 2: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 3: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 4: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 5: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 6: 0.0% user, 0.0% nice, 0.4% system, 0.0% interrupt, 99.6% idle
>> CPU 7: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 8: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 9: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> CPU 10: 0.0% user, 0.0% nice, 0.4% system, 0.0% interrupt, 99.6% idle
>> CPU 11: 0.0% user, 0.0% nice, 0.0% system, 0.0% interrupt, 100% idle
>> Mem: 8659M Active, 8385M Inact, 107G Wired, 1325M Free
>> ARC: 99G Total, 88G MFU, 10G MRU, 32K Anon, 167M Header, 529M Other
>> Swap: 32G Total, 32G Free
>
> I can collect some more info.
> First:
>
> 1046 - - R 125:10.79 nginx: worker process (nginx)
> 1047 kqread - I 150:11.98 nginx: worker process (nginx)
> 1049 kqread - I 145:58.35 nginx: worker process (nginx)
> 1050 kqread - I 136:33.36 nginx: worker process (nginx)
> 1051 kqread - I 140:59.73 nginx: worker process (nginx)
> 1052 kqread - I 137:18.12 nginx: worker process (nginx)
>
> pid 1046 is nginx running on CPU0 (affinity mask set).
>
> # procstat -k -k 1046
> PID TID COMM TDNAME KSTACK
> 1046 100686 nginx - mi_switch+0xd2 critical_exit+0x7e lapic_handle_timer+0xb1 Xtimerint+0x8c __mtx_lock_sleep+0x168 zone_fetch_slab+0x47 zone_import+0x52 zone_alloc_item+0x36 keg_alloc_slab+0x63 keg_fetch_slab+0x16e zone_fetch_slab+0x6e zone_import+0x52 uma_zalloc_arg+0x36e m_getm2+0x14f m_uiotombuf+0x64 sosend_generic+0x356 soo_write+0x42 dofilewrite+0x87
>
> Tracing command nginx pid 1046 tid 100686 td 0xfffff8014485f500
> sched_switch() at 0xffffffff804c956d = sched_switch+0x6ad/frame 0xfffffe20216992a0 /usr/src/sys/kern/sched_ule.c:1973
> mi_switch() at 0xffffffff804a8d92 = mi_switch+0xd2/frame 0xfffffe20216992d0 /usr/src/sys/kern/kern_synch.c:465
> critical_exit() at 0xffffffff804a6bee = critical_exit+0x7e/frame 0xfffffe20216992f0 /usr/src/sys/kern/kern_switch.c:219
> lapic_handle_timer() at 0xffffffff80771701 = lapic_handle_timer+0xb1/frame 0xfffffe2021699330 /usr/src/sys/x86/x86/local_apic.c:1185
> Xtimerint() at 0xffffffff806cbbcc = Xtimerint+0x8c/frame 0xfffffe2021699330 /usr/src/sys/amd64/amd64/apic_vector.S:135
> --- interrupt, rip = 0xffffffff804de424, rsp = 0xfffffe2021699400, rbp = 0xfffffe2021699420 ---
> lock_delay() at 0xffffffff804de424 = lock_delay+0x54/frame 0xfffffe2021699420 /usr/src/sys/kern/subr_lock.c:127
> __mtx_lock_sleep() at 0xffffffff80484dc8 = __mtx_lock_sleep+0x168/frame 0xfffffe20216994a0 /usr/src/sys/kern/kern_mutex.c:512
> zone_fetch_slab() at 0xffffffff806a4257 = zone_fetch_slab+0x47/frame 0xfffffe20216994e0 /usr/src/sys/vm/uma_core.c:2378
> zone_import() at 0xffffffff806a4312 = zone_import+0x52/frame 0xfffffe2021699530 /usr/src/sys/vm/uma_core.c:2501
> zone_alloc_item() at 0xffffffff806a0986 = zone_alloc_item+0x36/frame 0xfffffe2021699570 /usr/src/sys/vm/uma_core.c:2591
> keg_alloc_slab() at 0xffffffff806a2463 = keg_alloc_slab+0x63/frame 0xfffffe20216995d0 /usr/src/sys/vm/uma_core.c:965
> keg_fetch_slab() at 0xffffffff806a48ce = keg_fetch_slab+0x16e/frame 0xfffffe2021699620 /usr/src/sys/vm/uma_core.c:2349
> zone_fetch_slab() at 0xffffffff806a427e = zone_fetch_slab+0x6e/frame 0xfffffe2021699660 /usr/src/sys/vm/uma_core.c:2375
> zone_import() at 0xffffffff806a4312 = zone_import+0x52/frame 0xfffffe20216996b0 /usr/src/sys/vm/uma_core.c:2501
> uma_zalloc_arg() at 0xffffffff806a147e = uma_zalloc_arg+0x36e/frame 0xfffffe2021699720 /usr/src/sys/vm/uma_core.c:2531
> m_getm2() at 0xffffffff8048231f = m_getm2+0x14f/frame 0xfffffe2021699790 /usr/src/sys/kern/kern_mbuf.c:830
> m_uiotombuf() at 0xffffffff80516044 = m_uiotombuf+0x64/frame 0xfffffe20216997e0 /usr/src/sys/kern/uipc_mbuf.c:1535
> sosend_generic() at 0xffffffff8051ce56 = sosend_generic+0x356/frame 0xfffffe20216998a0
> soo_write() at 0xffffffff804fd872 = soo_write+0x42/frame 0xfffffe20216998d0
> dofilewrite() at 0xffffffff804f5c97 = dofilewrite+0x87/frame 0xfffffe2021699920
> kern_writev() at 0xffffffff804f5978 = kern_writev+0x68/frame 0xfffffe2021699970
> sys_writev() at 0xffffffff804f5be6 = sys_writev+0x36/frame 0xfffffe20216999a0
> amd64_syscall() at 0xffffffff806e4051 = amd64_syscall+0x2c1/frame 0xfffffe2021699ab0
> Xfast_syscall() at 0xffffffff806cb2bb = Xfast_syscall+0xfb/frame 0xfffffe2021699ab0
> --- syscall (121, FreeBSD ELF64, sys_writev), rip = 0x8019cc6ba, rsp = 0x7fffffffd688, rbp = 0x7fffffffd6c0 ---
This call stack is quite interesting:
1: A process is calling writev()
2: Kernel calls sosend_generic() that starts allocating memory
3: This allocation is then interrupted by the timer interrupt handler
[that could actually trigger tcp_tw_2msl_scan(reuse=0)]
4: The timer interrupt handler seems to wait on sched_switch()
And fun fact: When sosend_generic() calls m_uiotombuf() it does not
hold INP_WLOCK yet...
--
Julien
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 496 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20160926/9f4f219a/attachment.sig>
More information about the freebsd-stable
mailing list