I'm working on master branch and deploy two nodes cluster. Data is transferring over RDMA. [admin@server0 ~]$ sudo ceph daemon osd.0 perf dump AsyncMessenger::RDMAWorker-1 { "AsyncMessenger::RDMAWorker-1": { "tx_no_mem": 0, "tx_parital_mem": 0, "tx_failed_post": 0, "tx_chunks": 26966, "tx_bytes": 52789637, "rx_chunks": 26916, "rx_bytes": 52812278, "pending_sent_conns": 0 } } The only difference is that I don’t differentiate public/cluster network in my cluster. You can try to make all public/cluster network use RDMA. Note: 1) If both public/cluster use RDMA, we can’t differentiate them in different subnetwork. This is feature limited. I'm planning to solve it in future) 2) I'll confirm with my colleague that whether cluster network is really used in 14.2.4. We also hit similar problem these days even using TCP async messenger. Below is my cluster's ceph configuration. I also attach the systemd patch used in my side. [admin@server0 ~]$ cat /etc/ceph/ceph.conf [global] cluster = ceph fsid = 24280750-d4f7-4d4f-89e4-f95b8fab87ff auth_cluster_required = cephx auth_service_required = cephx auth_client_required = cephx osd pool default size = 2 osd pool default min size = 2 osd pool default pg num = 64 osd pool default pgp num = 128 osd pool default crush rule = 0 osd crush chooseleaf type = 1 mon_allow_pool_delete=true osd_pool_default_pg_autoscale_mode=off ms_type = async+rdma ms_async_rdma_device_name = mlx5_0 mon_initial_members = server0 mon_host = 172.16.1.4 [mon.rdmarhel0] host = server0 mon addr = 172.16.1.4 [admin@server0 ~]$ B.R. Changcheng On 13:07 Wed 30 Oct, Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) wrote:
1. The current problem is that it still sending data over the ethernet instead of ib. 2. [global] fsid=xxxx mon_initial_members = node1, node2, node3 mon_host = xxx.xx.xxx.ab,xxx.xx.xxx.ac, xxx.xx.xxx.ad auth_cluster_required = cephx auth_service_required = cephx auth_client_required = cephx public_network = xxx.xx.xxx.0/24 cluster_network = xx.xxx.0.0/16 ms_cluster_type = async+rdma ms_type = async+rdma ms_public_type = async+posix [mgr] ms_type = async+posix 3. The ceph cluster is deployed using ceph-deploy then once up all of the daemons are turned off the rdma cluster config is then sent around then once that is complete the daemons are turned back on. The ulimit is set to unlimited, LimitMEMLOCK=infinity is set on the ceph-disk@.service, ceph-mds@.service, ceph-mon@.service, ceph-osd@.service, ceph-radosgw@.service, aswell as PrivateDevices=no on ceph-mds@.service, ceph-mon@.service and ceph-radosgw@.service. The ethernet mtu is set to 1000 __________________________________________________________________
From: Liu, Changcheng <changcheng.liu@intel.com> Sent: 30 October 2019 12:24 To: Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) <gabryel.mason-williams@diamond.ac.uk> Cc: dev@ceph.io <dev@ceph.io> Subject: Re: RMDA Bug?
Liu, Changcheng wrote:
On 07:31 Mon 28 Oct, Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) wrote:
I am using ceph version 12.2.8 (ae699615bac534ea496ee965ac6192cb7e0e07c0) luminous (stable).
I have not checked the master branch do you think this is an
issue in
luminous that has been removed in later versions? I
haven't hit problem on master branch. Ceph/RDMA changed a lot from luminous to master branch.
Is below configuration really needed in luminous/ceph.conf?
ms_async_rdma_local_gid = xxxx On master branch,
1. What's the problem do you hit when using RDMA in 14.2.4? Any log shows the error? 2. What's your ceph.conf? 3. How do you deploy the ceph cluster? RDMA need lock some memory. So, it needs change some system configuration to meet with this requirement? On 11:21 Wed 30 Oct, Gabryel Mason-Williams wrote: this
parameter is not needed at all. B.R. Changcheng
Thanks, the issue of the OSD's falling over seems to have gone away
updating to Nautilus 14.2.4. However, I am still unable to get it to properly communicate over RDMA even with removing ms_async_rdma_local_gid.
_______________________________________________ Dev mailing list -- dev@ceph.io To unsubscribe send an email to dev-leave@ceph.io
--
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