[lustre-discuss] how to set max_pages_per_rpc (I have done something wrong and need help)

[Harald van Pee]

Thank you,

I will start next week and report what helps.
Unfortunately in this cluster we have to wait with the upgrade to 2.10.1
until next year.

Best whishes
Harald


On Wednesday 15 November 2017 23:09:49 Dilger, Andreas wrote:
> On Nov 15, 2017, at 12:56, Harald van Pee <pee at hiskp.uni-bonn.de> wrote:
> > Hello Andreas,
> > 
> > thanks for your information, now I have the feeling I'm not completly
> > lost. With erasing configuration parameters do you mean the
> > writeconf procedure? (chapter 14.4)
> 
> Yes.
> 
> > Or is it possible to erase the unknown parameter?
> 
> You could try "lctl conf_param -d <param>" to delete the parameter.
> 
> Cheers, Andreas
> 
> > On Wednesday 15 November 2017 20:37:20 Dilger, Andreas wrote:
> >> The problem that Lustre clients fail to mount when they get an unknown
> >> parameter is fixed in newer Lustre releases (2.9+) via patch
> >> https://review.whamcloud.com/21112 .
> >> 
> >> The current maintenance release is 2.10.1.
> >> 
> >> You could also work around this by erasing the configuration parameters
> >> (see Lustre manual).
> >> 
> >> Cheers, Andreas
> >> 
> >> On Nov 15, 2017, at 09:26, Harald van Pee
> >> <pee at hiskp.uni-bonn.de<mailto:pee at hiskp.uni-bonn.de>> wrote:
> >> 
> >> Here are more informations:
> >> 
> >> if I try to mount the filesystem on the client I get similar messages as
> >> from the failing conf_param command. It seems one have to remove this
> >> failed configuration but how?
> >> Here the syslog output on the client:
> >> 
> >> kernel: [ 4203.506437] LustreError: 3698:0:
> >> (obd_mount.c:1340:lustre_fill_super()) Unable to mount  (-2)
> >> kernel: [ 5028.547095] LustreError: 3830:0:
> >> (obd_config.c:1202:class_process_config()) no device for:
> >> hiskp3-OST0000-osc- ffff880416680800
> >> kernel: [ 5028.547105] LustreError: 3830:0:
> >> (obd_config.c:1606:class_config_llog_handler()) MGC192.168.128.200 at o2ib:
> >> cfg command failed: rc = -22
> >> kernel: [ 5028.547112] Lustre:    cmd=cf00f 0:hiskp3-OST0000-osc
> >> 1:osc.max_pages_per_rpc=256
> >> kernel: [ 5028.547112]
> >> kernel: [ 5028.547156] LustreError: 15b-f: MGC192.168.128.200 at o2ib: The
> >> configuration from log 'hiskp3-client'failed from the MGS (-22).  Make
> >> sure this client and the MGS are running compatible versions of Lustre.
> >> kernel: [ 5028.547407] LustreError:
> >> 1680:0:(lov_obd.c:946:lov_cleanup()) hiskp3-clilov-ffff880416680800:
> >> lov tgt 1 not cleaned! deathrow=0, lovrc=1 kernel: [ 5028.547415]
> >> LustreError: 1680:0:(lov_obd.c:946:lov_cleanup()) Skipped 3 previous
> >> similar messages
> >> kernel: [ 5028.550906] Lustre: Unmounted hiskp3-client
> >> kernel: [ 5028.551407] LustreError: 3815:0:
> >> (obd_mount.c:1340:lustre_fill_super()) Unable to mount  (-22)
> >> 
> >> 
> >> 
> >> On Wednesday 15 November 2017 16:06:29 Harald van Pee wrote:
> >> Dear all,
> >> 
> >> I want to set max_pages_per_rpc to 64 instead of 256
> >> lustre mgs/mdt version 2.53
> >> lustre oss version 2.53
> >> lustre client 2.6
> >> 
> >> on client I have done:
> >> lctl get_param osc.hiskp3-OST*.max_pages_per_rpc
> >> osc.hiskp3-OST0001-osc-ffff88105dba4800.max_pages_per_rpc=256
> >> osc.hiskp3-OST0002-osc-ffff88105dba4800.max_pages_per_rpc=256
> >> osc.hiskp3-OST0003-osc-ffff88105dba4800.max_pages_per_rpc=256
> >> osc.hiskp3-OST0004-osc-ffff88105dba4800.max_pages_per_rpc=256
> >> lctl set_param osc.hiskp3-OST*.max_pages_per_rpc=64
> >> 
> >> this works, but after remount I get again 256 therefore I want to make
> >> it permant with
> >> lctl conf_param hiskp3-OST*.osc.max_pages_per_rpc=64
> >> 
> >> But I get the message, that this command have to be given on mdt
> >> unfortunately I go to our combined mgs/mdt and get
> >> 
> >> Lustre: Setting parameter hiskp3-OST0000-osc.osc.max_pages_per_rpc in
> >> log hiskp3-client
> >> LustreError: 956:0:(obd_config.c:1221:class_process_config()) no device
> >> for: hiskp3-OST0000-osc-MDT0000
> >> LustreError: 956:0:(obd_config.c:1591:class_config_llog_handler())
> >> MGC192.168.128.200 at o2ib: cfg command failed: rc = -22
> >> Lustre:    cmd=cf00f 0:hiskp3-OST0000-osc-MDT0000
> >> 1:osc.max_pages_per_rpc=64
> >> 
> >> than I can not mount client and want to go back
> >> lctl set_param osc.hiskp3-OST*.max_pages_per_rpc=64
> >> 
> >> Lustre: Modifying parameter hiskp3-OST0000-osc.osc.max_pages_per_rpc in
> >> log hiskp3-client
> >> Lustre: Skipped 1 previous similar message
> >> LustreError: 966:0:(obd_config.c:1221:class_process_config()) no device
> >> for: hiskp3-OST0000-osc-MDT0000
> >> LustreError: 966:0:(obd_config.c:1591:class_config_llog_handler())
> >> MGC192.168.128.200 at o2ib: cfg command failed: rc = -22
> >> Lustre:    cmd=cf00f 0:hiskp3-OST0000-osc-MDT0000
> >> 1:osc.max_pages_per_rpc=256
> >> 
> >> obviously what I have done was completly wrong and I can no longer mount
> >> a client, mounted clients are working.
> >> How can I get it back working?
> >> hiskp3-MDT0000 ist the label of the mgs/mdt but
> >> hiskp3-OST0000-osc-MDT0000 seems to be incorrect
> >> 
> >> What I have to do to get the mgs/mdt working again?
> >> Its your production cluster
> >> Any help is welcome
> >> 
> >> Best
> >> Harald
> >> 
> >> 
> >> 
> >> 
> >> 
> >> 
> >> _______________________________________________
> >> lustre-discuss mailing list
> >> lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
> >> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
> >> 
> >> --
> >> Harald van Pee
> >> 
> >> Helmholtz-Institut fuer Strahlen- und Kernphysik der Universitaet Bonn
> >> Nussallee 14-16 - 53115 Bonn - Tel +49-228-732213 - Fax +49-228-732505
> >> mail: pee at hiskp.uni-bonn.de<mailto:pee at hiskp.uni-bonn.de>
> >> _______________________________________________
> >> lustre-discuss mailing list
> >> lustre-discuss at lists.lustre.org<mailto:lustre-discuss at lists.lustre.org>
> >> http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
> 
> Cheers, Andreas
> --
> Andreas Dilger
> Lustre Principal Architect
> Intel Corporation