День добрий! Sat, Oct 26, 2019 at 01:04:28AM +0800, changcheng.liu wrote:
What's your ceph version? Have you verified whether the problem could be reproduced on master branch?
As an option, it can be Jumbo Frames related bug. I had completely disabled JF in order to use RDMA over Ethernet on mlx4.
On 08:33 Fri 25 Oct, Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) wrote:
I am currently trying to run Ceph on RDMA, either RoCE 1 or 2. However, I am experiencing issues with this.
When using Ceph on RDMA I experience issues where OSD’s will randomly become unreachable even if the cluster is left alone alone, it also is not properly talking over RDMA and using Ethernet when the config states it should as shown by the same results in the bench marking of the two setups.
After reloading the cluster [cid:36020940-0085-40fc-bb5b-d91de6ace453]
After 5m 9s the cluster went from being healthy to down.
[cid:ed084bcc-0b97-44bd-9648-ce2e06859cd5]
This problem even happens when running a bench mark test on the cluster, OSD’s will just fall over. Another curious issue is that it is not properly talking over RDMA as well and instead using the Ethernet.
[cid:05e9dc68-075e-425d-b76b-ce7fa1d2f7a8]
Next test:
[cid:4183557e-b1da-41f3-afc3-f081b9fb4034]
The config used for the RDMA is a so:
[global]
fsid = aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa
mon_initial_members = node1, node2, node3
mon_host =xxx.xxx.xxx.xxx,xxx.xxx.xxx.xxx, xxx.xxx.xxx.xxx
auth_cluster_required = cephx
auth_service_required =cephx
auth_client_required = cephx
public_network = xxx.xxx.xxx.xxx/24
cluster_network = yyy.yyy.yyy.yyy/16
ms_cluster_type =async+rdma
ms_public_type = async+posix
ms_async_rdma_device_name = mlx4_0
[osd.0]
ms_async_rdma_local_gid = xxxx
[osd.1]
ms_async_rdma_local_gid = xxxx
[osd.2]
ms_async_rdma_local_gid =xxxx
Tests to check the system is using RDMA
sudo ceph --admin-daemon /var/run/ceph/ceph-osd.0.asok config show | grep ms_cluster
OUTPUT
"ms_cluster_type": "async+rdma",
sudo ceph daemon osd.0 perf dump AsyncMessenger::RDMAWorker-1
OUTPUT
{
"AsyncMessenger::RDMAWorker-1": {
"tx_no_mem": 0,
"tx_parital_mem": 0,
"tx_failed_post": 0,
"rx_no_registered_mem": 0,
"tx_chunks": 9,
"tx_bytes": 2529,
"rx_chunks": 0,
"rx_bytes": 0,
"pending_sent_conns": 0
}
}
When running over Ethernet I have a completely stable system with the current benchmarks as so
[cid:544ecbbc-10d9-43e6-ab2f-aa7c2bcd88c0]
Config setup when using Ethernet is
The Config setup when using Ethernet is
[global]
fsid = aaaaaaaa-aaaa-aaaa-aaaa-aaaaaaaaaaaa
mon_initial_members = node1, node2, node3
mon_host =xxx.xxx.xxx.xxx,xxx.xxx.xxx.xxx, xxx.xxx.xxx.xxx
auth_cluster_required = cephx
auth_service_required =cephx
auth_client_required = cephx
public_network = xxx.xxx.xxx.xxx/24
cluster_network = yyy.yyy.yyy.yyy/16
ms_cluster_type =async+posix
ms_public_type = async+posix
ms_async_rdma_device_name = mlx4_0
[osd.0]
ms_async_rdma_local_gid = xxxx
[osd.1]
ms_async_rdma_local_gid = xxxx
[osd.2]
ms_async_rdma_local_gid =xxxx Tests to check the system is using async+posix
sudo ceph --admin-daemon /var/run/ceph/ceph-osd.0.asok config show | grep ms_cluster
OUTPUT
"ms_cluster_type": "async+posix"
sudo ceph daemon osd.0 perf dump AsyncMessenger::RDMAWorker-1
OUTPUT
{}
This clearly a issue with RDMA and not with the OSD's shown by the fact the system is completely fine over Ethernet and not with RDMA.
Any guidance or ideas on how to approach this problem to make Ceph work with RDMA would be greatly appreciated.
Regards
Gabryel Mason-Williams, Placement Student
Address: Diamond Light Source Ltd., Diamond House, Harwell Science & Innovation Campus, Didcot, Oxfordshire OX11 0DE
Email: gabryel.mason-williams@diamond.ac.uk
--
This e-mail and any attachments may contain confidential, copyright and or privileged material, and are for the use of the intended addressee only. If you are not the intended addressee or an authorised recipient of the addressee please notify us of receipt by returning the e-mail and do not use, copy, retain, distribute or disclose the information in or attached to the e-mail. Any opinions expressed within this e-mail are those of the individual and not necessarily of Diamond Light Source Ltd. Diamond Light Source Ltd. cannot guarantee that this e-mail or any attachments are free from viruses and we cannot accept liability for any damage which you may sustain as a result of software viruses which may be transmitted in or with the message. Diamond Light Source Limited (company no. 4375679). Registered in England and Wales with its registered office at Diamond House, Harwell Science and Innovation Campus, Didcot, Oxfordshire, OX11 0DE, United Kingdom
_______________________________________________ Dev mailing list -- dev@ceph.io To unsubscribe send an email to dev-leave@ceph.io
ceph-users mailing list -- ceph-users@ceph.io To unsubscribe send an email to ceph-users-leave@ceph.io