ORA-27102 SVR4 Error: 12: Not enough space

lrp

Our Oracle 11.1.0.7 database is running on Solaris 10 / Sunfire hardware. Unfortunately, we've been seeing our database crash and corrupt a datafile in the process several times. The key alert log message is below (which I've blanked out some info for security purposes):

KCF: write/open error block=0xbc28 online=1
file=6 /***************.dbf
error=27063 txt: 'SVR4 Error: 12: Not enough space
Additional information: -1
Additional information: 8192'

KCF: write/open error block=0x1571 online=1
file=57 *************.dbf
error=27063 txt: 'SVR4 Error: 12: Not enough space
Additional information: -1
Additional information: 8192'
Automatic datafile offline due to write error on
file 57: ***********.dbf

I've already done a forum search for the error codes but couldn't find any other post that matched my exact situation/code. The key coincidence here is that the number 8192 matches a value in my ulimit (stack size). Below is what I would currently see as my oracle user:

ulimit -a
core file size (blocks, -c) unlimited
data seg size (kbytes, -d) unlimited
file size (blocks, -f) unlimited
open files (-n) 256
pipe size (512 bytes, -p) 10
stack size (kbytes, -s) 8192  <-------------
cpu time (seconds, -t) unlimited
max user processes (-u) 29995
virtual memory (kbytes, -v) unlimited

Oracle's documentation doesn't give me much information on what ORA-27063 means. Does anybody else know how Oracle reports 'Error 12:" codes coming from the OS? What the "additional information: -1" or "additional information: 8192" means? I need this in my troubleshooting so we can measure and adjust the correct resource, instead of blindly increasing stack size or other resource parameters on the OS.

sb92075

27063, 00000, "number of bytes read/written is incorrect"
// *Cause: the number of bytes read/written as returned by aiowait
// does not match the original number, additional information
// indicates both these numbers
// *Action: check errno

: 'SVR4 Error: 12: Not enough space

Is the volume filling up & running out of free disk space?

lrp

Good thoughts, but we had ruled that out early: Filesystem's not the issue, the disk had plenty of space. Metalink itself (472813.1) points to "the Unix error number (ERRNO) 12 during a Unix write()/open() system call, and this Unix error indicates a lack of *process memory* rather than physical disk space."

- /var/adm/messages has no memory- or disk-related messages around the time of failure.
- SAN administrator saw nothing in their logs at the time of failure

We had already tried raising SGA and raising shared memory in the solaris project, but it seems like we're fishing for an answer by blindly raising a parameter when we don't know what OS limit Oracle had reached. The key numbers I'm looking for are those specified in the 'additional information' section. Oracle's knowledge base has nothing that I can use so far.

sb92075

http://www.lmgtfy.com/?q=oracle+SVR4+Error:+12:+Not+enough+space

lrp

Thanks! We'd definitely been checking google for information for the past couple weeks before checking with Oracle Forums. In fact, there are several blogs and [lazydba/experts-exchange links|http://www.lazydba.com/oracle/0__125336.html] on the subject which point us in the right direction and was the basis for us looking at enlarging the shared memory kernel parameters to start.

At this point, it's more how to interpret how Oracle spits out information, since there wasn't any publicly available information on the format of the error code.

Much like how P1, P2, and P3 in v$session_wait will mean different things, I would guess that the "Additional Information" tokens after the "Error: 12 code" mean different things. That much is evident in my searches, where it appears that:
Error code 12 => memory related things
Error code 11 => resource is temporarily unavailable for whatever reason
Error code 5 => disk/IO issue

.. so drilling down further, Error code 12's two additional information items must mean something:
-1 => some return code?
8192 => the number at which it failed at? Some bit-wise address?

At no point does the stack gets mentioned in our diagnostic text, which is why I'm asking the larger oracle community.

sb92075

Are file=6 & file=57 on same volume?

What is storage architecture containing Oracle's dbf files?
What flavor of file system supports Oracle's dbf files?

lrp

To elaborate, the filesystem is UFS-based, all configured in RAID-6 [(striped disks with dual parity)|http://en.wikipedia.org/wiki/Redundant_array_of_independent_disks] (including the redo logs and archive logs). The storage underneath the filesystem is a Hitachi 9985v SAN, meaning the physical disks themselves were grouped into logical partitions and then divvied up into filesystems. Files #6 & 57 are on different filesystems, but all of them had ample space at the time.

sb92075

Since errors are being logged into alertSID.log file, you know when this problem occurs.
When these errors occur, is it during HEAVY I/O activity.

I am not making any accusation, just making idle observation.
I have never used (or seen) UFS under Oracle.
An ever so slight possibility is a file system bug is inflicting the damage.

From my exeperience, the root cause is outside Oracle at OS or similar layer.
Oracle is just to dumb to lie about errors & it is detecting a SNAFU in underlying system.

Good Luck with your Gremlin hunt.

lrp

I am entirely with you on this.
-- if it was exclusively an oracle memory error, we would have seen an ORA-4031 indicating exactly which pool was compromised.
-- if it was a dbwr error, we would have seen a more descriptive alert saying 'disk full' or 'crc did not match'
-- etc.

The only IO activity we did catch was the fact that RMAN archive log backups were running at the time. Our I/O usage charts for those filesystems (and solaris sar/vmstat/iostat counts) did not spike during those times.

Regarding the filesystem setup, our vote was to use ZFS, but this was a decision made beyond our heads. Unfortunately, without a real error to show the SAN administrator, we are unable to provide them effective evidence to support the claim. Getting proper diagnostic information was the point of this forum post -- notably, Metalink's own article +(22080.1 - An Introduction to Error Message Articles)+ pretty much admits we need to look further than just the error codes:

Please note that the steps included under these additional headings
is generally very terse. Information may be incomplete and may use
abbreviations which are not self explanatory. It is also possible that
there may be references to articles which are not visible to customers.
These additional notes are intended to help give pointers and are NOT
intended to be a complete explanation of the error.

More complete descriptions may be included in separate problem/solution or
bulletin documents.

orafad

ora-27063 mentions "Cause: the number of bytes read/written as returned by aiowait does not match the original number, additional information indicates both these numbers"
So OS returned -1 ("error state") while Oracle expected 8196 bytes (block size, probably).

About the ENOMEM (error code 12) - it might be a good idea to check shared mem settings and system memory overall.

sb92075

When these errors occur, is it during HEAVY I/O activity?
or just randomly across 24 hours?

lrp

to SB: This symptom has showed up across our dev and production databases every 2 months or so. Interpreting from the alert log, it has also happened close to the time when an archive/backup occurs at the same time as either a recompile, auto-stats gather, or snapshot. So, "load" would appear to cause it, but nothing I would consider heavy I/O activity.

orafad: The metalink for ORA-27063 does mention this, and I guess it to be a rather generic post. The real culprit has to be the codes behind the 27063, which is what I'm trying to get to the bottom of:
SVR4 = a header error indicating some OS 'thing'
Error 12 = ???? Solaris memory error. This appears to show up for most 'capacity-related' things on google searches, for both file descriptors, semaphores, swap, and shared memory.
Additional information = -1 error state, you're most likely correct.
Additional information: 8192 = could mean anything, and that's what I need to find out from either Metalink support (in progress) or Solaris's knowledge base.

Update: I apologize, orafad -- you mentioned the ENOMEM, which I missed. Where did you reference this ? I'd love to have an additional resource to look up the error code. Our memory_target/memory_max_size were set at 5G/8GB, with the shared memory set at 20GB and overall physical memory at 32GB. Our sysadmin logs showed no memory usage or swap errors, leading me to believe it was not a general 'out of memory' error so much as a kernel resource setting (semaphores, per process limit of some sort).

Edited by: lrp on Jun 1, 2009 5:50 PM

Edited by: lrp on Jun 1, 2009 5:53 PM

sb92075

This symptom has showed up across our dev

if I were in your shoes, I'd do what I could to change the underlying file system .
If problem still happens on different fs, then file system flavor can be ruled out as possible root cause.

Again, I am fairly certain Oracle is the victim & not the culprit.
Proving who or what is to blame will be a battle.

Happy Hunting!

lrp

Because this happens so infrequently, I need a way to measure what is happening (be it stack, filedescriptors) . I've already got scripts tallying those resources per process on a 5 minute interval, so I'm hoping to prepare myself for the next occurrence (could be next week, could be a month from now). ..unfortunately, moving to another filesystem is going to be rather hard to prove a case for, since we would have no way to really identifying whether the experiment was successful.

Thanks for your time in this.

sb92075

Are there any additional clues in OS messages or dmesg logs?

lrp

Absolutely no /var/adm/messages related to the failure (which is what keeps pissing me off, since Metalink's notes on the 27063 error routinely point me to the OS logs) -- I just spoke with the SA, and most of the messages get piped there. He did have sar available with a viewer, but i also had charts for the disk i/o at the time -- nothing stood out.

561365

solaris 10 require the setting of memory limit

projadd -U oracle -K "project.max-shm-memory=(priv,4096MB,deny)" user.oracle

The memory specified above should be more than the actual sga &pga you are using.

Thanks

lrp

That was the first thing we increased -- Our 11g memory target (which is the combined SGA+PGA, automatically managed by the instance) is at 5GB, with a memory_max_size = 8G. While it is the only database in the solaris project there ARE other databases on the prod server, contained within their own project (ie. oraproj2) and ORACLE_HOME.

If I'm reading the results below correctly, I believe we have 16GB of shared memory available:

[email protected]:PROD01:/var/adm> projects -l oraproj
oraproj
        projid : 102
        comment: ""
        users  : oracle
        groups : (none)
        attribs: project.max-sem-ids=(priv,200,deny)
                 project.max-sem-nsems=(priv,512,deny)
                 project.max-shm-ids=(priv,200,deny)
                 project.max-shm-memory=(priv,17179869184,deny)

jgarry

lrp wrote:
Good thoughts, but we had ruled that out early: Filesystem's not the issue, the disk had plenty of space. Metalink itself (472813.1) points to "the Unix error number (ERRNO) 12 during a Unix write()/open() system call, and this Unix error indicates a lack of *process memory* rather than physical disk space."

- /var/adm/messages has no memory- or disk-related messages around the time of failure.
- SAN administrator saw nothing in their logs at the time of failure

We had already tried raising SGA and raising shared memory in the solaris project, but it seems like we're fishing for an answer by blindly raising a parameter when we don't know what OS limit Oracle had reached. The key numbers I'm looking for are those specified in the 'additional information' section. Oracle's knowledge base has nothing that I can use so far.

On a different version and platform, I had rare issues when running RMAN. Eventually I came to the conclusion that: RMAN uses large pool; OS eventually fragments I/O buffers due to the way RMAN uses I/O on that platform. It is of course wild speculation that this has anything to do with your problem, but given the lack of real information, I'd say you perhaps want to shrink your SGA some, give more to large pool, and pray a lot. Oh, and sample the large pool SGA statistics when you are running RMAN, and wonder if the parallel automatic tuning is shooting your large_pool_size in the leg if you are using it.

sb92075

I just stumble across this

Subject: Upon startup of Linux database get ORA-27102: out of memory Linux-X86_64 Error: 28: No space left on device
Doc ID: 301830.1

Let us know if it helped.

lrp

Subject: Upon startup of Linux database get ORA-27102: out of memory Linux-X86_64 Error: 28: No space left on device

Doc ID: 301830.1>
Thank you. I looked up the article, and it shows a similar message but has a key distinction between error codes -- that OS code was error: 28, while my error was error: 12. The linux and Solaris error code tables are similar, so for OS error 28, the the solaris error code table shows:

"28: ENOSPC No space left on device
While writing an ordinary file or creating a directory entry, there is no free space
left on the device. In the fcntl routine, the setting or removing of record locks
on a file cannot be accomplished because there are no more record entries left
on the system.{code}
By the same token, my OS error code 12 that I'm seeing appears to mean:
{code}
"12 ENOMEM Not enough space
During execution of an exec, brk, or sbrk routine, a program asks for more space
than the system is able to supply. This is not a temporary condition; the maximum
size is a system parameter. On some architectures, the error may also occur if the
arrangement of "text, data, and stack segments requires too many"
"segmentation registers, or if there is not enough swap space" during the fork
routine. If this error occurs on a resource associated with Remote File
Sharing (RFS), it indicates a memory depletion which may be temporary,
dependent on system activity at the time the call was invoked.

Emphasis on the clause "..*text, data, and stack segments requires too many segmentation registers, or if there is not enough swap space*.."

..So my clues from the OS documentation point to some maximum, like swap, segmentation, stack and other resources.
I can only assume the "additional information" of -1 and 8192 are relevant numbers to those resources.

The two things that are *8192* in my environment appear to be # of file descriptors and stack size.

Therefore, my plan of attack is going to change both in my oracle profile to see if this occurs again:

ulimit -n 16834 (raise file descriptors per process from 8kto 16k)
ulimit -s 32767 (raise stack from 8mb to 32mb)

Hopefully, this will give Oracle more leeway to use OS resources AND give me extra clues if the error shows up again. In other words, if something like stack is truly the issue, then I expect to see another crash with addtional information = *32767* instead of 8192:

KCF: write/open error block=0x1571 online=1
file=57 /datafile/DB_001.dbf
error=27063 txt: 'SVR4 Error: 12: Not enough space
Additional information: -1
Additional information: "32767"'

Will update this thread with any relevant results..

orafad

lrp wrote:
..So my clues from the OS documentation point to some maximum, like swap, segmentation, stack and other resources.
I can only assume the "additional information" of -1 and 8192 are relevant numbers to those resources.

The two things that are *8192* in my environment appear to be # of file descriptors and stack size.

As I tried to explain earlier: this means no more than "I (Oracle) asked to read/write 8192 bytes from the buffer but got back -1". Syscall returns number of bytes actually read/written, or -1 indicating that an error occurred.

error=27063 txt: 'SVR4 Error: 12: Not enough space

Is this from a 32-bit Oracle server?

As the software owner user, could you verify system parameters? (prctl)

Specifically, what are your parameters settings that corresponds to shmmax and shmall?

lrp

>
As I tried to explain earlier: this means no more than "I (Oracle) asked to read/write 8192 bytes from the buffer but got back -1". Syscall returns number of bytes actually read/written, or -1 indicating that an error occurred.
error=27063 txt: 'SVR4 Error: 12: Not enough space
>
I do recall the post. Is there a man page describing that error code that I can look into further? I realize that the error number, but I didn't happen to find anywhere which stated the details behind 'errno' or return code. The man page for syscall talks about returning -1 on error, but doesn't say anything about the second return code ( unless you mean the variable errno ).

>
Is this from a 32-bit Oracle server?
As the software owner user, could you verify system parameters? (prctl)
Specifically, what are your parameters settings that corresponds to shmmax and shmall?
>
It's Oracle 64-bit Enterprise on Solaris 10 SPARC 64-bit. I hate pasting the full text of a screendump, but since I cannot really figure which are the important pieces of info, I'll paste the results of prctl on the project that my DB is running under. (if you can narrow which params i'm looking for, i can cull the rest later in an edit). This is only for one of the db's but the idea is the same. Note that project.max-shm-memory = 16 GB, well above my oracle memory_max_size of 8GB. I'm not sure how to obtain the solaris equivalent of shmall.

[email protected]:PROD01:/fs1> prctl -i project oraproj
project: 100: oraproj
NAME    PRIVILEGE       VALUE    FLAG   ACTION                       RECIPIENT
project.max-contracts
        privileged      10.0K       -   deny                                 -
        system          2.15G     max   deny                                 -
project.max-device-locked-memory
        privileged      1.95GB      -   deny                                 -
        system          16.0EB    max   deny                                 -
project.max-locked-memory
        system          16.0EB    max   deny                                 -
project.max-port-ids
        privileged      8.19K       -   deny                                 -
        system          65.5K     max   deny                                 -
project.max-shm-memory
        privileged      16.0GB      -   deny                                 -          <----------
        system          16.0EB    max   deny                                 -
project.max-shm-ids
        privileged        200       -   deny                                 -
        system          16.8M     max   deny                                 -
project.max-msg-ids
        privileged        258       -   deny                                 -
        system          16.8M     max   deny                                 -
project.max-sem-ids
        privileged        200       -   deny                                 -
        system          16.8M     max   deny                                 -
project.max-crypto-memory
        privileged      7.81GB      -   deny                                 -
        system          16.0EB    max   deny                                 -
project.max-tasks
        system          2.15G     max   deny                                 -
project.max-lwps
        system          2.15G     max   deny                                 -
project.cpu-cap
        system          4.29G     inf   deny                                 -
project.cpu-shares
        privileged          1       -   none                                 -
        system          65.5K     max   none                                 -
zone.max-swap
        system          16.0EB    max   deny                                 -
zone.max-locked-memory
        system          16.0EB    max   deny                                 -
zone.max-shm-memory
        system          16.0EB    max   deny                                 -
zone.max-shm-ids
        system          16.8M     max   deny                                 -
zone.max-sem-ids
        system          16.8M     max   deny                                 -
zone.max-msg-ids
        system          16.8M     max   deny                                 -
zone.max-lwps
        system          2.15G     max   deny                                 -
zone.cpu-cap
        system          4.29G     inf   deny                                 -
zone.cpu-shares
        privileged          1       -   none                                 -

Edited by: lrp on Jun 4, 2009 4:23 PM

lrp1

Hello--I'm just updating the post with further information.
There is a Nov 9 2009 SUN Blog post (http://blogs.sun.com/hippy/entry/problems_with_solaris_and_a) which mentions similar symptoms to our problem and mentions basically to either:
a) upgrade to solaris 10 update 8 (we are at update 4)
b) disable oracle DISM

The article appears to advise turning off only SGA_MAX_SIZE. We currently have several memory settings:

memory_max_target                    big integer 8000M
memory_target                        big integer 5056M
shared_memory_address                integer     0
sga_max_size                         big integer 8000M
sga_target                           big integer 0
pga_aggregate_target                 big integer 0

If we were to disable DISM, does that mean disabling ONLY SGA_MAX_SIZE, or should we also remove the MEMORY_MAX_SIZE? If we wanted to maintain the same memory settings, would we then set memory_target to 8000M and leave SGA_TARGET/PGA_AGGREGATE_TARGET completely alone?

All in all, Oracle Support still does not give us many clues beyond saying that it is an OS file-resource error, not recognizing that it is a solaris kernel memory limit.