Re: panic: Journal overflow

On 04/26/07 03:32, Rong-en Fan wrote:
> On 4/26/07, Eric Anderson <anderson_at_freebsd.org> wrote:
>> On 04/25/07 13:24, Rong-en Fan wrote:
>>> This is a i386 current SMP box as of Apr 14 or 15. Got
>>> a panic with geom journal.
>>>
>>> panic: Journal overflow (joffset=3246704015360 active=3246705852416 inactive=324
>>> cpuid = 2
>>> KDB: stack backtrace:
>>> db_trace_self_wrapper(c06b3e2c,e5321944,c04fc86e,c06c70c8,2,...) at db_trace_sel
>>> kdb_backtrace(c06c70c8,2,c06ad0d2,e5321950,5,...) at kdb_backtrace+0x2f
>>> panic(c06ad0d2,eea3b800,2f3,eebfc000,2f3,...) at panic+0x11f
>>> g_journal_check_overflow(c4fc5c00,cb030a00,eb,95428000,eb,...) at g_journal_chec
>>> g_journal_flush(c4fc5c00,0,eb,95428000,eb,...) at g_journal_flush+0x60d
>>> g_journal_add_current(c4fc5c00,c9cb0948,ca8e818c,c4fc5c00,e5321cbc,...) at g_jou
>>> g_journal_release_delayed(c4fc5c00,0,ca8e818c,c4fdadc0,2,...) at g_journal_relea
>>> g_journal_flush_send(c4fc5c00,c8521c60,205d0000,1b2,205d4000,...) at g_journal_f
>>> g_journal_worker(c4fc5c00,e5321d38,0,0,0,...) at g_journal_worker+0x7f7
>>> fork_exit(c04b5813,c4fc5c00,e5321d38) at fork_exit+0x83
>>> fork_trampoline() at fork_trampoline+0x8
>>> --- trap 0, eip = 0, esp = 0xe5321d70, ebp = 0 ---
>>>
>>> Sorry, that I don't have core dump available. Will set up next time.
>>> After the panic, the system hangs at single user prompt:
>>>
>>> Trying to mount root from ufs:/dev/da0s1a
>>> WARNING: / was not properly dismounted
>>> Loading configuration files.
>>> kernel dumps on /dev/da0s1b
>>> Entropy harvesting: interrupts ethernet point_to_point kickstart.
>>> swapon: adding /dev/da0s1b as swap device
>>> Starting file system checks:
>>> /dev/da0s1a: 1704 files, 56880 used, 196935 free (631 frags, 24538 blocks, 0.2%
>>> Can't stat /dev/concat/data.journal: No such file or directory
>>> /dev/da0s1f: 237 files, 1208 used, 11877856 free (64 frags, 1484724 blocks, 0.0%
>>> GEOM_JOURNAL: Journal concat/data consistent.
>>> /dev/da0s1e: UNREF FILE I=548536  OWNER=root MODE=100644
>>> /dev/da0s1e: SIZE=717466 MTIME=Apr  4 07:58 2007  (CLEARED)
>>> /dev/da0s1e: FREE BLK COUNT(S) WRONG IN SUPERBLK (SALVAGED)
>>> /dev/da0s1e: SUMMARY INFORMATION BAD (SALVAGED)
>>> /dev/da0s1e: BLK(S) MISSING IN BIT MAPS (SALVAGED)
>>> /dev/da0s1e: 190105 files, 763850 used, 2281197 free (36557 frags, 280580 blocks
>>> /dev/da0s1d: 17507 files, 40158 used, 972857 free (641 frags, 121527 blocks, 0.1
>>> THE FOLLOWING FILE SYSTEM HAD AN UNEXPECTED INCONSISTENCY:
>>>         ufs: /dev/concat/data.journal (/data)
>>> Unknown error; help!
>>> AEnter full pathname of shell or RETURN for /bin/sh:
>>> # mount -a
>>> WARNING: R/W mount of /backup denied.  Filesystem is not clean - run fsck
>>> mount: /dev/concat/data.journal : Operation not permitted
>>> # fsck_ffs -p /dev/concat/data.journal
>>>
>>> I need to issue 'fsck_ffs -p' myself... any idea about why this happens?
>>>
>>> The geom journal setup:
>>>
>>> Geom name: gjournal 68372861
>>> ID: 68372861
>>> Providers:
>>> 1. Name: concat/data.journal
>>>    Mediasize: 3246137539072 (3.0T)
>>>    Sectorsize: 512
>>>    Mode: r1w1e1
>>> Consumers:
>>> 1. Name: concat/data
>>>    Mediasize: 3247211281408 (3.0T)
>>>    Sectorsize: 512
>>>    Mode: r1w1e1
>>>    Jend: 3247211280896
>>>    Jstart: 3246137539072
>>>    Role: Data,Journal
>>>
>>> The gconcat consists two scsi disk (actually, it's raid) da0 and da1.
>>> Oh no, it panics with journal overflow again while writing this message :(
>>>
>>> The data.journal is shared by nfs, and there are two boxes that are
>>> doing a tar writing operation on this partition.
>> You need to change your journal switch and cache switch times, to
>> something like this:
>>
>> kern.geom.journal.force_switch=50
>> kern.geom.journal.cache.switch=75
>>
>> Try that and see if that eases your pain a bit.
> 
> This does not help :(
> In the past two hours, I tried tuning this two sysctls a bit.
> The result is panic over 10 times :(


How low did you try them?


> I have to remove gjournal on concat/data. I do this following
> 
> gjournal stop concat/data
> gjournal clear concat/data
> tunefs -J disable /dev/concat/data
> 
> If I want to turn on gjournal someday, can I do it without
> recreate the filesystem?


Without doing any research, I would say 'sure', because I can't think of 
a reason why not.  I think just relabeling it, then turning it on via 
tunefs would do it..

Eric