Message Server Down, File table overflow, SAP Oracle HP UX

[Former Member]

Our Current issue Message Server's Down on CI, with Platform on HP_UX, Oracle, SAP NW04s.

Dev_MS: [Thr 01] Wed Nov 5 16:22:18 2008

[Thr 01] ***LOG Q0I=> NiIAccept: accept (23: File table overflow) [nixxi.cpp 3014]

[Thr 01] *** ERROR => NiIAccept: SiAccept failed for hdl 0 / sock 7

(SI_EMFILE/23; I4; ST; own high 35; max hds 802) [nixxi.cpp 3014]

[Thr 01] *** ERROR => MsSConnect: NiAccept (rc=NIEINVAL) [msxxserv_mt. 10251]

[Thr 01] *** ERROR => NiIInitSocket: SiSocket failed (sirc=17; I4; ST; errno=23: File table overflow) [nixxi.cpp 691]

[Thr 01] ***LOG Q0I=> NiIInitSocket: socket (23: File table overflow) [nixxi.cpp 697]

[Thr 01] *** ERROR => NiIInitSocket: SiSocket failed for hdl 0 / sock -1

(SI_EMFILE/23; I4; ST; own high 35; max hds 802) [nixxi.cpp 697]

[Thr 01] *** ERROR => NiIListen: NiIInitSocket failed (rc=-1) [nixxi.cpp 2287]

[Thr 01] *** ERROR => MsSCommInit: NiBuf2Listen(sapmsL4P) (rc=NIEINTERN) [msxxserv_mt. 10113]

[Thr 01] *** ERROR => MsSConnect: MsSCommInit(sapmsL4P) (rc=-1) [msxxserv_mt. 10262]

[Thr 01] *** ERROR => MsSConnect [msxxserv_mt. 5901]

[Thr 01] ***LOG Q02=> MsSHalt, MSStop (Msg Server 23083) [msxxserv_mt. 5944]

We have referred to SAP Notes 546006, 9748 and increased nfile value to 163840. Now we are operating within the limits of nfile.

We are really concerned about the Oracle and SAP Process why those process are not closing the Opened Files by them. Though there is not much load on the server the nfile value reaches 80000.

I appreciate the proper replies.

[Former Member · Accepted solution]

Well, I have checked the open files by Oracle and SAPWEBDISP were dominant. So what should I have to counter to close the files which were opened by these process automatically.

Do I need to upgrade the Kernel.

Thank you Marks.

[markus_doehr2 · Accepted solution]

I would increase the kernel parameters, not close applications....

Markus

[fidel_vales · Accepted solution]

While oracle is up and running, the shadow process will open the the data files as they are the ones that read the data from them to the shared memory.

They open the files and do not close them until the shadow process is finished/closed ...

It would not make sense that the shadow processes would be opening and closing files several times per second and this would be a bigger overhead.

Regards

[fidel_vales]

Hi,

Can you tell me

1.- total number of SAP workprocesses?

2.- total number of oracle datafiles?

Note that the error is not caused by oracle, but to give you a little idea why Oracle and SAP "are not closing the Opened Files by them"

[Former Member]

Total Number of Wp's in Ci 76,

Total Number of Datafiles in Oracle;108

Thankyou for ur reply.

[former_member204746]

76 processes in CI& I know that this is possible, that sounds like a lot to me.

[Former Member]

Eric, u can go upto 100 wp's for an instance, and were good with the nfile rate until last week.

We have opened the oss message but there was no appropriate support reflecting our issue.

Our concern how to find out what are the files opened by Oracle, and SAP apart from that why they were not closed based on load.

Also we have webdispatcher on CI itself.

[markus_doehr2]

Each network connection and each file that is open consumes a descriptor, you may have ran out of those.

Check the HP-UX kernel parameters (using sam):

nfile

maxfiles

maxfiles_lim

and also the shell limits for <sid>adm and ora<sid>.

Especially if you run a webdispatcher who opens two connections per "web connect" you need to increase those values.

Markus