Re: Linux Loopback Interface
Date: Wed, 11 Sep 2019 22:53:28 -0400
Message-ID: <CAMHX9JKY4r-hMWKXSPoVknm0s-4w6ZhA1KPUHQbBhxYrqPRm6g_at_mail.gmail.com>
You could use lsof to see who's using any TCP/UDP sockets with IP 127.0.0.1 as one endpoint (or any other IP if any have been assigned to loopback interface)
$ *sudo lsof -nPi | grep 127.0.0.1*
chronyd 5334 chrony 1u IPv4 35307 0t0 UDP 127.0.0.1:323 master 6535 root 13u IPv4 38297 0t0 TCP 127.0.0.1:25(LISTEN) Regarding the memory errors and reducing MTU size, there have been similar issues with jumbo frames. "Large" frames and MTU sizes require kernel to allocate large *contiguous* memory chunks from kernel memory. Since such allocation sizes can vary (different kernel drivers & allocation reasons using them), this kernel memory can get fragmented, thus finding a "large" chunk on demand gets problematic. This can cause paging/reclamation issues (even if there's some memory free) or kernel memory allocation failures.
One solution would be to tune the Linux kernel settings to keep more memory free at any time (vm.min_free_kbytes) or use lower MTU/frame/kernel memory allocation sizes.
You can monitor the fragmentation of (some) kernel memory via
*/proc/buddyinfo*. This is a good article explaining the buddy allocator
basics and how to read /proc/buddyinfo + a python script:
$ *cat /proc/buddyinfo *
$ *./buddyinfo.py *
Node 0, zone DMA 1 0 0 0 2 1 1
0 1 1 3
Node 0, zone DMA32 2656 4401 3224 4288 1480 277 29
0 0 0 0
Node 0, zone Normal 23 76 25 16 5 0 0
0 1 0 0
Node: 0
Zone: DMA
Free KiB in zone: 15876.00
Fragment size Free fragments Total available KiB 4096 1 4.0 8192 0 0.0 16384 0 0.0 32768 0 0.0 65536 2 128.0 131072 1 128.0 262144 1 256.0 524288 0 0.0 1048576 1 1024.0 2097152 1 2048.0 4194304 3 12288.0Zone: DMA32
Free KiB in zone: 368760.00
Fragment size Free fragments Total available KiB 4096 1788 7152.0 8192 4401 35208.0 16384 3224 51584.0 32768 4288 137216.0 65536 1480 94720.0 131072 277 35456.0 262144 29 7424.0 524288 0 0.0 1048576 0 0.0 2097152 0 0.0 4194304 0 0.0Zone: Normal
Free KiB in zone: 2956.00
Fragment size Free fragments Total available KiB 4096 23 92.0 8192 76 608.0 16384 25 400.0 32768 16 512.0 65536 5 320.0 131072 0 0.0 262144 0 0.0 524288 0 0.0 1048576 1 1024.0 2097152 0 0.0 4194304 0 0.0
Thanks
-- Tanel Poder https://blog.tanelpoder.com/seminar On Tue, Sep 10, 2019 at 10:20 PM Rakesh Ra <rakeshra.tr_at_gmail.com> wrote:Received on Thu Sep 12 2019 - 04:53:28 CEST
> Hi All,
>
> Can someone shed some light on for what operations Oracle can use loopback
> interface? We had the below issue on one of our primary production database
> running on 12.1.0.2 running on Red Hat Linux 7.5.
>
>
> skgxpvfynet: mtype: 61 process 75262 failed because of a resource problem
> in the OS. The OS has most likely run out of buffers (rval: 4) opiodr
> aborting process unknown ospid (75262) as a result of ORA-603
>
> Oracle asked us to fix the MTU by lowering it from 64k to 16k by following
> the below note.There were just application user connections (jsut less than
> 400 user processes) and couple of scp process which was shipping the backup
> files from the this primary database server to standby database server as
> part of DR build.
>
> lo: flags=73<UP,LOOPBACK,RUNNING> mtu 16436 <<<<<<<<<<<< This value was
> 64K during the issue time. It looks starting 6.5 kernel , MTU is bumped to
> 64K. We lowered this to 16K
> inet 127.0.0.1 netmask 255.0.0.0
> loop txqueuelen 1000 (Local Loopback)
> RX packets 2689657 bytes 17458642228 (16.2 GiB)
> RX errors 0 dropped 0 overruns 0 frame 0
> TX packets 2689657 bytes 17458642228 (16.2 GiB)
> TX errors 0 dropped 0 overruns 0 carrier 0 collisions 0
>
> Oracle Linux: ORA-27301:OS Failure Message: No Buffer Space Available (Doc
> ID 2041723.1)
>
> I followed up with Oracle to see, in what way loopback interface is
> contributing to the issue here, but they failed to give more details saying
> it's OS specific and they don't have idea on the same. :(
>
> Appreciate if someone has some previous experience working on similar
> issue provide some information related to this.
>
> BR,
> Rakesh RA
>
-- http://www.freelists.org/webpage/oracle-l