Re: fio main thread got stuck over the weekend

[Jens Axboe <axboe@kernel.dk>]

On 2014-08-11 11:04, scameron@beardog.cce.hp.com wrote:
> On Mon, Aug 11, 2014 at 10:44:23AM -0500, scameron@beardog.cce.hp.com wrote:
>>
>> Hi,
>>
>> I was running an fio job over the weekend and the main thread
>> seems to have hung.  This in on RHEL6u5 with kernel
>> 2.6.32-431.20.3.el6.x86_64 but with a highly modified hpsa
>> driver, just in case it matters.
>>
>> [root@msablackburn saperf]# uname -a
>> Linux msablackburn.americas.hpqcorp.net 2.6.32-431.20.3.el6.x86_64 #1 SMP Fri Jun 6 18:30:54 EDT 2014 x86_64 x86_64 x86_64 GNU/Linux
>> [root@msablackburn saperf]# /usr/local/bin/fio --version
>> fio-2.1.11-3-g789d
>>
>> The jobfile looks like this (note the runtime=216000, that's 60 hours)
>> It seems to have hung with about 25 hours remaining.)
>>
>> [root@msablackburn saperf]# cat 2drive_sdr.fio
>> [global]
>> direct=1
>> ioengine=libaio
>> norandommap
>> randrepeat=0
>> bs=4096
>> iodepth=1011
>> #iodepth=96
>> #numjobs=1
>> numjobs=12
>> #numjobs=24
>> # number_ios=1
>> runtime=216000
>> #runtime=54000
>> time_based=1
>> group_reporting
>> thread
>> gtod_reduce=1
>> #iodepth_batch=4
>> #iodepth_batch_complete=4
>> cpus_allowed=0-5
>> cpus_allowed_policy=split
>> rw=randread
>>
>> [4_KiB_RR_drive_r]
>> filename=/dev/sdr
>>
>> [4_KiB_RR_drive_s]
>> filename=/dev/sds
>>
>>
>> Here is all the output so far:
>>
>> ---------------8<------------8<-------------8<-----------------
>> [root@msablackburn saperf]# fio 2drive_sdr.fio
>> 4_KiB_RR_drive_r: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=libaio, iodepth=1011
>> ...
>> 4_KiB_RR_drive_s: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=libaio, iodepth=1011
>> ...
>> fio-2.1.11-3-g789d
>> Starting 24 threads
>> ^Cbs: 24 (f=24): [r(24)] [1.2% done] [1113MB/0KB/0KB /s] [285K/0/0 iops] [eta 14h:48m:47s]
>> fio: terminating on signal 2
>>
>> 4_KiB_RR_drive_r: (groupid=0, jobs=24): err= 0: pid=7696: Fri Aug  8 16:49:19 2014
>>    read : io=881568MB, bw=1309.1MB/s, iops=335341, runt=672989msec
>>    cpu          : usr=1.75%, sys=23.15%, ctx=1266117, majf=9, minf=835
>>    IO depths    : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=0.1%, >=64=100.0%
>>       submit    : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
>>       complete  : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.1%
>>       issued    : total=r=225681428/w=0/d=0, short=r=0/w=0/d=0
>>       latency   : target=0, window=0, percentile=100.00%, depth=1011
>>
>> Run status group 0 (all jobs):
>>     READ: io=881568MB, aggrb=1309.1MB/s, minb=1309.1MB/s, maxb=1309.1MB/s, mint=672989msec, maxt=672989msec
>>
>> Disk stats (read/write):
>>    sdr: ios=112299535/0, merge=0/0, ticks=13349007/0, in_queue=13054755, util=98.57%
>>    sds: ios=113359581/0, merge=0/0, ticks=13710834/0, in_queue=13414931, util=98.71%
>> [root@msablackburn saperf]# vi !$
>> vi 2drive_sdr.fio
>> [root@msablackburn saperf]# fio 2drive_sdr.fio
>> 4_KiB_RR_drive_r: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=libaio, iodepth=1011
>> ...
>> 4_KiB_RR_drive_s: (g=0): rw=randread, bs=4K-4K/4K-4K/4K-4K, ioengine=libaio, iodepth=1011
>> ...
>> fio-2.1.11-3-g789d
>> Starting 24 threads
>> Jobs: 24 (f=24): [r(24)] [57.1% done] [1408MB/0KB/0KB /s] [361K/0/0 iops] [eta 01d:01h:43m:07s]
>> ---------------8<------------8<-------------8<-----------------
>>
>> And there it sits, the eta does not update.
>>
>>
>> I tried connecting to the process with gdb, and doing "info threads" and "bt":
>>
>> PID of main thread is 8530:
>>
>> root      8530  598  1.0 2159116 170812 pts/0  Sl+  Aug08 23570:56 fio 2drive_sdr.fio
>>
>> ...
>> (gdb) info threads
>>    26 Thread 0x7f65da431700 (LWP 8531)  0x0000003aa4cacced in nanosleep () from /lib64/libc.so.6
>>    25 Thread 0x7f65d9a30700 (LWP 8532)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    24 Thread 0x7f65d902f700 (LWP 8533)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    23 Thread 0x7f65d862e700 (LWP 8534)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    22 Thread 0x7f65d7c2d700 (LWP 8535)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    21 Thread 0x7f65d722c700 (LWP 8536)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    20 Thread 0x7f65d682b700 (LWP 8537)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    19 Thread 0x7f65d5e2a700 (LWP 8538)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    18 Thread 0x7f65d5429700 (LWP 8539)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    17 Thread 0x7f65d4a28700 (LWP 8540)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    16 Thread 0x7f65d4027700 (LWP 8541)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    15 Thread 0x7f65d3626700 (LWP 8542)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    14 Thread 0x7f65d2c25700 (LWP 8543)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    13 Thread 0x7f65d2222700 (LWP 8544)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    12 Thread 0x7f65d1820700 (LWP 8545)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    11 Thread 0x7f65d0e1e700 (LWP 8546)  td_io_queue (td=0x7f65da496410, io_u=0x7f656804d010) at ioengines.c:344
>>    10 Thread 0x7f65d041c700 (LWP 8547)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    9 Thread 0x7f65cfa1a700 (LWP 8548)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    8 Thread 0x7f65cf018700 (LWP 8549)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    7 Thread 0x7f65ce616700 (LWP 8550)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    6 Thread 0x7f65cdc14700 (LWP 8551)  fio_libaio_event (td=0x7f65da4b06a8, event=<value optimized out>) at engines/libaio.c:65
>>    5 Thread 0x7f65cd212700 (LWP 8552)  __get_next_offset (td=0x7f65da4b5a60) at io_u.c:389
>>    4 Thread 0x7f65cc810700 (LWP 8553)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>>    3 Thread 0x7f65cbe0e700 (LWP 8554)  td_io_u_unlock (td=0x7f65da4c01d0, min_evts=<value optimized out>, bytes=0x7f65cbe0dd10) at fio.h:619
>>    2 Thread 0x7f65cb40c700 (LWP 8555)  0x0000003c69800667 in io_submit () from /lib64/libaio.so.1
>> * 1 Thread 0x7f65e34c6720 (LWP 8530)  0x0000003aa4cacced in nanosleep () from /lib64/libc.so.6
>> (gdb) bt
>> #0  0x0000003aa4cacced in nanosleep () from /lib64/libc.so.6
>> #1  0x0000003aa4ce1e44 in usleep () from /lib64/libc.so.6
>> #2  0x000000000044afe4 in do_usleep () at backend.c:1727
>> #3  run_threads () at backend.c:1965
>> #4  0x000000000044b84d in fio_backend () at backend.c:2068
>> #5  0x0000003aa4c1ed1d in __libc_start_main () from /lib64/libc.so.6
>> #6  0x0000000000409c39 in _start ()
>> (gdb) detach
>> Detaching from program: /usr/local/bin/fio, process 8530
>> (gdb) quit
>> [root@msablackburn saperf]#
>>
>> I am guessing the 8530 thread is the one that prints out the eta updates?
>> Well, maybe not, I see a call to create_disk_util_thread() in backend.c.
>>
>> i/o still seems to be running, and there are no messages in dmesg about
>> hung i/o or the scsi mid layer getting angry and trying to abort things
>> or anything like that.  Seems to be happily cruising along, except that
>> the eta isn't updating, and the job should have finished several hours
>> ago now.  I'm leaving it running for now just to try to figure out
>> what happened.
>>
>> Any ideas?  Anything else I should poke at?
>
> A bit more info:
>
> thread 26 looks to be the one that's supposed to print the eta info:
>
> (gdb) thread 26
> [Switching to thread 26 (Thread 0x7f65da431700 (LWP 8531))]#0  0x0000003aa4cacced in nanosleep () from /lib64/libc.so.6
> (gdb) bt
> #0  0x0000003aa4cacced in nanosleep () from /lib64/libc.so.6
> #1  0x0000003aa4ce1e44 in usleep () from /lib64/libc.so.6
> #2  0x00000000004474a0 in disk_thread_main (data=<value optimized out>) at backend.c:1992
> #3  0x0000003aa50079d1 in start_thread () from /lib64/libpthread.so.0
> #4  0x0000003aa4ce8b5d in clone () from /lib64/libc.so.6
> (gdb)
>
> static void *disk_thread_main(void *data)
> {
>          int ret = 0;
>
>          fio_mutex_up(startup_mutex);
>
>          while (threads && !ret) {
>                  usleep(DISK_UTIL_MSEC * 1000);
>                  if (!threads)
>                          break;
>                  ret = update_io_ticks();
>
>                  if (!is_backend)
>                          print_thread_status();
>          }
>
>          fio_mutex_up(disk_thread_mutex);
>          return NULL;
> }
>
> that usleep() is where it's in nanosleep, I guess.
>
> from eta.c:
>
> void print_thread_status(void)
> {
>          struct jobs_eta *je;
>          size_t size;
>
>          je = get_jobs_eta(0, &size);
>          if (je)
>                  display_thread_status(je);
>
>          free(je);
> }
>
> Maybe that je is coming back false?  which is
> probably the return value of calc_thread_status() which, well,
> at a glance, I'm not sure what calc_thread_status() is doing.

I'll take a look at this next week, been away at a conference since last 
weekend.


-- 
Jens Axboe