3

I've installed an NFS server on Ubuntu 20.04 and a FreeIPA Ubuntu 20.04 client with the users home directories served by the NFS server. Performance is extremely slow when accessing files. When I strace the process with time spent in syscalls, I find that openat can take over 1 second at times for the NFS files! (see below). Needless to say file access on the server displays no such issue. openat is the only slow operation.

I've attached a histogram of the times spent in openat (I've trimmed the top bin so the tail can be seen). There are over 800 openat calls that complete in under 10mSec, but it's the tail that makes the difference in the total time, and there are a lot of calls that take over 100mSec which is unreasonable.

I suspect it may have to do with the Kerberos authorisation or something of the kind, but I don't know how to investigate the issue.

Options in /etc/exports:

/home   *(rw,sec=krb5:krb5i:krb5p,async,no_subtree_check)

Mount on client:

server:/home/... on /home/... type nfs4 (rw,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=krb5,clientaddr=xx.xx.xx.x1,local_lock=none,addr=xx.xx.xx.x2)

Any help or clue would be appreciated,

Yuval.

0.000064 : stat("/home/.../lib/python3.8/site-packages/pandas/core", {st_mode=S_IFDIR|0775, st_size=4096, ...}) = 0
0.000040 : stat("/home/.../lib/python3.8/site-packages/pandas/core/nanops.py", {st_mode=S_IFREG|0664, st_size=50002, ...}) = 0
0.000095 : stat("/home/.../lib/python3.8/site-packages/pandas/core/nanops.py", {st_mode=S_IFREG|0664, st_size=50002, ...}) = 0
0.664737 : openat(AT_FDCWD, "/home/.../lib/python3.8/site-packages/pandas/core/__pycache__/nanops.cpython-38.pyc", O_RDONLY|O_CLOEXEC) = 6
0.000122 : fstat(6, {st_mode=S_IFREG|0664, st_size=36431, ...}) = 0
0.000116 : ioctl(6, TCGETS, 0x7ffed1278d60)        = -1 ENOTTY (Inappropriate ioctl for device)
0.000049 : lseek(6, 0, SEEK_CUR)                   = 0
0.000024 : lseek(6, 0, SEEK_CUR)                   = 0
0.000028 : fstat(6, {st_mode=S_IFREG|0664, st_size=36431, ...}) = 0
0.000052 : read(6, "U\r\r\n\0\0\0\0\216t\362aR\303\0\0\343\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 36432) = 36431
0.000024 : read(6, "", 1)                          = 0
0.000438 : close(6)                                = 0
0.000083 : mmap(NULL, 262144, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f23fcbaf000
0.000100 : stat("/home/.../bin", {st_mode=S_IFDIR|0775, st_size=4096, ...}) = 0
0.000120 : stat("/usr/lib/python3.8", {st_mode=S_IFDIR|0755, st_size=20480, ...}) = 0
0.000122 : stat("/usr/lib/python3.8/lib-dynload", {st_mode=S_IFDIR|0755, st_size=12288, ...}) = 0
0.000037 : getcwd("/home/yuval/src/themis", 1024)  = 23
0.000046 : stat("/home/yuval/src/themis", {st_mode=S_IFDIR|0775, st_size=130, ...}) = 0
0.000037 : stat("/home/.../lib/python3.8/site-packages", {st_mode=S_IFDIR|0775, st_size=12288, ...}) = 0
0.000051 : stat("/home/.../lib/python3.8/site-packages/pandas/core/array_algos", {st_mode=S_IFDIR|0775, st_size=163, ...}) = 0
0.000041 : stat("/home/.../lib/python3.8/site-packages/pandas/core/array_algos/masked_reductions.py", {st_mode=S_IFREG|0664, st_size=3721, ...}) = 0
0.000085 : stat("/home/.../lib/python3.8/site-packages/pandas/core/array_algos/masked_reductions.py", {st_mode=S_IFREG|0664, st_size=3721, ...}) = 0
0.411113 : openat(AT_FDCWD, "/home/.../lib/python3.8/site-packages/pandas/core/array_algos/__pycache__/masked_reductions.cpython-38.pyc", O_RDONLY|O_CLOEXEC) = 6
0.000053 : fstat(6, {st_mode=S_IFREG|0664, st_size=3329, ...}) = 0
0.000027 : ioctl(6, TCGETS, 0x7ffed1278d60)        = -1 ENOTTY (Inappropriate ioctl for device)
0.000043 : lseek(6, 0, SEEK_CUR)                   = 0
0.000037 : lseek(6, 0, SEEK_CUR)                   = 0
0.000025 : fstat(6, {st_mode=S_IFREG|0664, st_size=3329, ...}) = 0
0.000032 : read(6, "U\r\r\n\0\0\0\0\216t\362a\211\16\0\0\343\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 3330) = 3329
0.000025 : read(6, "", 1)                          = 0
0.000438 : close(6)                                = 0
0.000105 : stat("/home/.../lib/python3.8/site-packages/pandas/core/arrays", {st_mode=S_IFDIR|0775, st_size=4096, ...}) = 0
0.000102 : stat("/home/.../lib/python3.8/site-packages/pandas/core/arrays/categorical.py", {st_mode=S_IFREG|0664, st_size=94502, ...}) = 0
0.000101 : stat("/home/.../lib/python3.8/site-packages/pandas/core/arrays/categorical.py", {st_mode=S_IFREG|0664, st_size=94502, ...}) = 0
0.413090 : openat(AT_FDCWD, "/home/.../lib/python3.8/site-packages/pandas/core/arrays/__pycache__/categorical.cpython-38.pyc", O_RDONLY|O_CLOEXEC) = 6
0.000063 : fstat(6, {st_mode=S_IFREG|0664, st_size=77947, ...}) = 0
0.000041 : ioctl(6, TCGETS, 0x7ffed127b180)        = -1 ENOTTY (Inappropriate ioctl for device)
0.000037 : lseek(6, 0, SEEK_CUR)                   = 0
0.000023 : lseek(6, 0, SEEK_CUR)                   = 0
0.000031 : fstat(6, {st_mode=S_IFREG|0664, st_size=77947, ...}) = 0
0.000085 : read(6, "U\r\r\n\0\0\0\0\216t\362a&q\1\0\343\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 77948) = 77947

Histogram of time spent in openat for opening Python and importing pytorch

2
  • I have a very similar and even more extreme issue. The openat() to actually open the file takes some 5 seconds! Everything related to walking the directory tree or stat is in the µs range. Did you come any closer to the issue?
    – Lars Hanke
    Apr 14, 2022 at 13:47
  • Yes. But I'm not certain how. I've reattached the client to the domain, and found out chronyd was not working correctly (I don't have access to an external NTP and it wouldn't update from a single source for some reason). I got so frustrated with it that I ended up running an ssh command to the domain server in crontab and updated from the date from the returned. date Not very accurate but it was good enough. I'm not sure which change resolved the issue.
    – YuvGM
    May 12, 2022 at 13:11

1 Answer 1

1

What version of the kernel is installed? This might be related to a known NFS client server bug which was just patched a couple of weeks ago:

https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2009325

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .