Re: Ceph on RDMA

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi,

I used  'https://community.mellanox.com/docs/DOC-2721', and was under the impression I followed all steps.. but I somehow skipped over the /usr/lib/systemd/system/ (/lib/systemd/system/ in case of Ubuntu) ceph-xxx@.service files. I did update the /etc/security/limits.conf file for unlimited memlock, but as I am using systemd for start/stop control, that ofcourse had no effect.

After the alterations in de systemd service files, RDMA works!! The 'RDMAStack RDMAStack!!! WARNING !!!' notice was legit afterall :)

I do encounter some oddites I would like to share.
After upgrading (using apt update/upgrade) the kernel to 4.4.0-93-generic, ceph-OSD barfes out again, see below log snippet.

    -9> 2017-08-29 09:40:21.538709 7f7254dde700  2 -- 10.0.0.2:6801/2828 >> - conn(0xc56a78e800 :6801 s=STATE_ACCEPTING pgs=0 cs=0 l=0)._stop     -8> 2017-08-29 09:40:21.538696 7f72565e1700  2 -- 10.0.0.2:6805/2828 >> 10.0.0.3:0/27835 conn(0xc56a630800 :6805 s=STATE_ACCEPTING_WAIT_SEQ pgs=3 cs=1 l=1).handle_connect_msg accept write reply msg done     -7> 2017-08-29 09:40:21.538772 7f72565e1700  2 -- 10.0.0.2:6805/2828 >> 10.0.0.3:0/27835 conn(0xc56a630800 :6805 s=STATE_ACCEPTING_WAIT_SEQ pgs=3 cs=1 l=1)._process_connection accept get newly_acked_seq 0     -6> 2017-08-29 09:40:21.539284 7f7254dde700  1 -- 10.0.0.2:6801/2828 >> - conn(0xc56a5ab000 :6801 s=STATE_ACCEPTING pgs=0 cs=0 l=0)._process_connection sd=50 -     -5> 2017-08-29 09:40:21.539319 7f7254dde700  5 Infiniband recv_msg recevd: 6, 924, 2116118, 0, fe80000000000000001e0bffff4cff55     -4> 2017-08-29 09:40:21.539354 7f7254dde700 -1 RDMAConnectedSocketImpl activate failed to transition to RTR state: (22) Invalid argument     -3> 2017-08-29 09:40:21.541306 7f72555df700  1 -- 10.0.0.2:6800/2828 >> - conn(0xc56a5a8000 :6800 s=STATE_ACCEPTING pgs=0 cs=0 l=0)._process_connection sd=52 -     -2> 2017-08-29 09:40:21.541355 7f72555df700  5 Infiniband recv_msg recevd: 6, 927, 11225430, 0, fe80000000000000001e0bffff4cff55     -1> 2017-08-29 09:40:21.541379 7f72555df700 -1 RDMAConnectedSocketImpl activate failed to transition to RTR state: (22) Invalid argument      0> 2017-08-29 09:40:21.544688 7f72555df700 -1 /build/ceph-12.1.4/src/msg/async/rdma/RDMAConnectedSocketImpl.cc: In function 'void RDMAConnectedSocketImpl::handle_connection()' thread 7f72555df700 time 2017-08-29 09:40:21.541394 /build/ceph-12.1.4/src/msg/async/rdma/RDMAConnectedSocketImpl.cc: 244: FAILED assert(!r)


Reverting back to 4.4.0-92-generic solves it though.. so I stay with that for now.
I will have a go with linux 4.6.0.

Thanks again!

Best regards,
Jeroen Oldenhof


Op 28-8-2017 om 16:12 schreef Haomai Wang:
do you follow this instruction(https://community.mellanox.com/docs/DOC-2693)?

On Mon, Aug 28, 2017 at 6:40 AM, Jeroen Oldenhof <jeroen@xxxxxx> wrote:
Hi All!

I'm trying to run CEPH over RDMA, using a batch of Infiniband Mellanox
MT25408 20GBit (4x DDR) cards.

RDMA is running, rping works between all hosts, and I've configured 10.0.0.x
addressing on the ib0 interfaces.

But when enabling RMDA in ceph.conf:

   ms_type = async+rdma
   ms_async_rdma_device_name = mlx4_0

OSD and MON on all hosts barf:
     -5> 2017-08-28 12:03:37.623110 7f2326c9de00  1  Processor -- start
     -4> 2017-08-28 12:03:37.624232 7f23205fe700  1 Infiniband binding_port
found active port 1
     -3> 2017-08-28 12:03:37.624250 7f23205fe700  1 Infiniband init
assigning: 1024 receive buffers
     -2> 2017-08-28 12:03:37.624260 7f23205fe700  1 Infiniband init
assigning: 1024 send buffers
     -1> 2017-08-28 12:03:37.624262 7f23205fe700  1 Infiniband init device
allow 4194303 completion entries
      0> 2017-08-28 12:03:37.628379 7f23205fe700 -1
/build/ceph-12.1.4/src/msg/async/rdma/Infiniband.cc: In function 'int
Infiniband::MemoryManager::Cluster::fill(uint32_t)' thread 7f23205fe700 time
2017-08-28 12:03:37.624433
/build/ceph-12.1.4/src/msg/async/rdma/Infiniband.cc: 599: FAILED assert(m)


As suggested in the thread
http://lists.ceph.com/pipermail/ceph-users-ceph.com/2017-June/018943.html
I also tried with lower values for receive and send buffer:
     -4> 2017-08-28 12:36:46.997001 7fa28b270700  1 Infiniband binding_port
found active port 1
     -3> 2017-08-28 12:36:46.997026 7fa28b270700  1 Infiniband init
assigning: 256 receive buffers
     -2> 2017-08-28 12:36:46.997029 7fa28b270700  1 Infiniband init
assigning: 256 send buffers
     -1> 2017-08-28 12:36:46.997030 7fa28b270700  1 Infiniband init device
allow 4194303 completion entries
      0> 2017-08-28 12:36:47.001835 7fa28b270700 -1
/build/ceph-12.1.4/src/msg/async/rdma/Infiniband.cc: In function 'int
Infiniband::MemoryManager::Cluster::fill(uint32_t)' thread 7fa28b270700 time
2017-08-28 12:36:46.997231
/build/ceph-12.1.4/src/msg/async/rdma/Infiniband.cc: 599: FAILED assert(m)




$ ceph -v
ceph version 12.1.4 (a5f84b37668fc8e03165aaf5cbb380c78e4deba4) luminous (rc)

In the osd logs I also see some of these:
   -236> 2017-08-28 12:17:34.507315 7f7f4815ce00 -1 RDMAStack RDMAStack!!!
WARNING !!! For RDMA to work properly user memlock (ulimit -l) must be big
enough to allow large amount of registered memory. We recommend setting this
parameter to infinity

but, the memlock has been set to infinity:

$ ulimit -l
unlimited

Any suggestions..?

Best regards,
Jeroen Oldenhof
The Netherlands

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com




[Index of Archives]     [Information on CEPH]     [Linux Filesystem Development]     [Ceph Development]     [Ceph Large]     [Linux USB Development]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [xfs]


  Powered by Linux