Hi Williams, Besides usign same port(both publid and cluster network use RDMA) for RDMA messenger, I also tried to use public-network-TCP-messenger and cluster-network-RDMA-messenger. There's no serious problem happen. The ceph is built by self based on master commit 8cb1f6bd(Wed Nov 6 18:43:41 2019 -0500). I don't have too many nodes to check your problem. BTW, on "ceph-users Digest, Vol 82, Issue 27", there's below item: 2. Re: mgr daemons becoming unresponsive (Gregory Farnum) However, I haven't hit mgr problem in my side. B.R. Changcheng On 08:53 Mon 11 Nov, Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) wrote:
@Changcheng
Sorry for the late reply as well.
I followed your setup and I have an issue where the MGR cannot connect to the cluster and RDMA does not work, I believe the MGR is not supported on RDMA.
Thank you for your time but I believe we may be hitting a dead end with this approach as we seem to get different results.
Kind regards
Gabryel Mason-Williams __________________________________________________________________
From: Liu, Changcheng <changcheng.liu@intel.com> Sent: 01 November 2019 06:24 To: Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) <gabryel.mason-williams@diamond.ac.uk> Cc: dev@ceph.io <dev@ceph.io> Subject: Re: RMDA Bug?
@Williams, Sorry for late reply. I'm busy on working getting Ceph/RDMA performance data these days. I'm using Intel RDMA NIC with small cluster, there's no serious issue happened. For Mellanox NIC, there's no problem with your ceph.conf from my perspective. Below is the steps that I used to deploy cluster 1. server0: 172.16.1.4, /dev/nvme0n1, /dev/nvme1n1 2. server1: 172.16.1.2, /dev/nvme0n1, /dev/nvme1n1
Below is my deploy steps: [admin@server0 deploy]$ ceph-deploy new server0 --fsid 24280750-d4f7-4d4f-89e4-f95b8fab87ff [admin@server0 deploy]$ #change ceph.conf as below: [admin@server0 deploy]$ cat 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=on
ms_type = async+rdma ;----changcheng: change device to your dev name---------- ms_async_rdma_device_name = irdma1 ;----changcheng: ignore below parameters with Mellanox NIC-------- ;ms_async_rdma_support_srq = false
mon_initial_members = server0 mon_host = 172.16.1.4
[mon.rdmarhel0] host = server0 mon addr = 172.16.1.4 [admin@server0 deploy]$ ceph-deploy mon create-initial [admin@server0 deploy]$ ceph-deploy admin server0 server1 [admin@server0 deploy]$ ceph-deploy mgr create server0 [admin@server0 deploy]$ ceph-deploy osd create --data /dev/nvme0n1 server0 [admin@server0 deploy]$ ceph-deploy osd create --data /dev/nvme1n1 server0 [admin@server0 deploy]$ ceph-deploy osd create --data /dev/nvme0n1 server1 [admin@server0 deploy]$ ceph-deploy osd create --data /dev/nvme1n1 server1 B.R. Changcheng On 08:27 Thu 31 Oct, Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) wrote:
1. When not defining a public and cluster network the OSD and MGR nodes do not get recognised
sudo ceph -s
cluster:
id: 820f1573-bc4a-4ee0-b702-80ba5ac13c25
health: HEALTH_WARN
3 osds down
3 hosts (3 osds) down
1 root (3 osds) down
no active mgr
too few PGs per OSD (21 < min 30)
services:
mon: 3 daemons, quorum cs04r-sc-com99-05,cs04r-sc-com99-07,cs04r-sc-com99-08 (age 5m)
mgr: no daemons active (since 4m)
osd: 3 osds: 0 up (since 9m), 3 in (since 9m)
data:
pools: 1 pools, 64 pgs
objects: 0 objects, 0 B
usage: 3.0 GiB used, 114 GiB / 117 GiB avail
pgs: 44 stale+active+clean
20 active+clean
This is an issue within the ms_type being async+rdma as the
daemons are
running:
sudo systemctl status ceph-osd.target
$B!|(B ceph-osd.target - ceph target allowing to start/stop all ceph-osd@.service instances at once
Loaded: loaded (/usr/lib/systemd/system/ceph-osd.target; enabled; vendor preset: enabled)
Active: active since Thu 2019-10-31 08:13:42 GMT; 8min ago
sudo systemctl status ceph-mgr.target
$B!|(B ceph-mgr.target - ceph target allowing to start/stop all ceph-mgr@.service instances at once Loaded: loaded (/usr/lib/systemd/system/ceph-mgr.target; enabled; vendor preset: enabled) Active: active since Thu 2019-10-31 08:13:33 GMT; 11min ago
With the config being
[global]
fsid = 820f1573-bc4a-4ee0-b702-80ba5ac13c25
mon_initial_members = node1, node2, node3
mon_host = xxx.xx.xxx.aa,xxx.xx.xxx.ac, xxx.xx.xxx.ad
auth_cluster_required = cephx
auth_service_required = cephx
auth_client_required = cephx
ms_type = async+rdma
ms_async_rdma_device_name = mlx4_0
__________________________________________________________________
From: Liu, Changcheng <changcheng.liu@intel.com> Sent: 31 October 2019 01:09 To: Mason-Williams, Gabryel (DLSLtd,RAL,LSCI) <gabryel.mason-williams@diamond.ac.uk> Cc: dev@ceph.io <dev@ceph.io> Subject: Re: RMDA Bug?
2) I'll confirm with my colleague that whether cluster network
really used in 14.2.4. We also hit similar problem these days even using TCP async messenger. [Changcheng]: 1) The problem should be already sovled in 14.2.4. We hit the
in 14.2.1 2) I'll try to verify your problem when I have time(I'm working on other affairs). There should be no problem when unifying both
I'm working on master branch and deploy two nodes cluster. Data is
network with RDMA device. On 23:22 Wed 30 Oct, Liu, Changcheng wrote: transferring over RDMA.
[admin@server0 ~]$ 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$B!G(Bt differentiate
network in my cluster.
You can try to make all public/cluster network use RDMA. Note: 1) If both public/cluster use RDMA, we can$B!G(Bt 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
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
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?
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:
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
is an
> 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,
issue in 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
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
is problem public/cluster public/cluster the then this the 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
From 40fa0d7096364b410e8242c46967029fb949876a Mon Sep 17 00:00:00 2001 From: Changcheng Liu <changcheng.liu@aliyun.com> Date: Tue, 23 Jul 2019 18:50:57 +0800 Subject: [PATCH] rdma systemd: grant access to /dev and unlimit mem
Signed-off-by: Changcheng Liu <changcheng.liu@aliyun.com>
diff --git a/systemd/ceph-fuse@.service.in b/systemd/ceph-fuse@.service.in index d603042b12..ff2e9072f6 100644 --- a/systemd/ceph-fuse@.service.in +++ b/systemd/ceph-fuse@.service.in @@ -12,6 +12,7 @@ ExecStart=/usr/bin/ceph-fuse -f --cluster ${CLUSTER} %I LockPersonality=true MemoryDenyWriteExecute=true NoNewPrivileges=true +LimitMEMLOCK=infinity # ceph-fuse requires access to /dev fuse device PrivateDevices=no ProtectControlGroups=true diff --git a/systemd/ceph-mds@.service.in b/systemd/ceph-mds@.service.in index 39a2e63105..0e58dfeeea 100644 --- a/systemd/ceph-mds@.service.in +++ b/systemd/ceph-mds@.service.in @@ -14,7 +14,8 @@ ExecReload=/bin/kill -HUP $MAINPID LockPersonality=true MemoryDenyWriteExecute=true NoNewPrivileges=true -PrivateDevices=yes +LimitMEMLOCK=infinity +PrivateDevices=no ProtectControlGroups=true ProtectHome=true ProtectKernelModules=true diff --git a/systemd/ceph-mgr@.service.in b/systemd/ceph-mgr@.service.in index c98f6378b9..682c7ecef3 100644 --- a/systemd/ceph-mgr@.service.in +++ b/systemd/ceph-mgr@.service.in @@ -18,7 +18,8 @@ LockPersonality=true MemoryDenyWriteExecute=false
NoNewPrivileges=true -PrivateDevices=yes +LimitMEMLOCK=infinity +PrivateDevices=no ProtectControlGroups=true ProtectHome=true ProtectKernelModules=true diff --git a/systemd/ceph-mon@.service.in b/systemd/ceph-mon@.service.in index c95fcabb26..51854fad96 100644 --- a/systemd/ceph-mon@.service.in +++ b/systemd/ceph-mon@.service.in @@ -21,7 +21,8 @@ LockPersonality=true MemoryDenyWriteExecute=true # Need NewPrivileges via `sudo smartctl` NoNewPrivileges=false -PrivateDevices=yes +LimitMEMLOCK=infinity +PrivateDevices=no ProtectControlGroups=true ProtectHome=true ProtectKernelModules=true diff --git a/systemd/ceph-osd@.service.in b/systemd/ceph-osd@.service.in index 1b5c9c82b8..06c20d7c83 100644 --- a/systemd/ceph-osd@.service.in +++ b/systemd/ceph-osd@.service.in @@ -16,6 +16,8 @@ LockPersonality=true MemoryDenyWriteExecute=true # Need NewPrivileges via `sudo smartctl` NoNewPrivileges=false +LimitMEMLOCK=infinity +PrivateDevices=no ProtectControlGroups=true ProtectHome=true ProtectKernelModules=true diff --git a/systemd/ceph-radosgw@.service.in b/systemd/ceph-radosgw@.service.in index 7e3ddf6c04..fe1a6b9159 100644 --- a/systemd/ceph-radosgw@.service.in +++ b/systemd/ceph-radosgw@.service.in @@ -13,7 +13,8 @@ ExecStart=/usr/bin/radosgw -f --cluster ${CLUSTER} --name client.%i --setuser ce LockPersonality=true MemoryDenyWriteExecute=true NoNewPrivileges=true -PrivateDevices=yes +LimitMEMLOCK=infinity +PrivateDevices=no ProtectControlGroups=true ProtectHome=true ProtectKernelModules=true diff --git a/systemd/ceph-volume@.service b/systemd/ceph-volume@.service index c21002cecb..e2d1f67b85 100644 --- a/systemd/ceph-volume@.service +++ b/systemd/ceph-volume@.service @@ -9,6 +9,7 @@ KillMode=none Environment=CEPH_VOLUME_TIMEOUT=10000 ExecStart=/bin/sh -c 'timeout $CEPH_VOLUME_TIMEOUT /usr/sbin/ceph-volume-systemd %i' TimeoutSec=0 +LimitMEMLOCK=infinity
[Install] WantedBy=multi-user.target -- 2.17.1
_______________________________________________ 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
participants (1)
-
Liu, Changcheng