Developer Playgrounds

Developer playground machines are used for developing Ceph.

All machines have cephfs mounts for accessing the teuthology, scratch, and postfile file systems. It is encouraged to use these machines to view teuthology logs as they will be significantly faster than the teuthology VM, which is often memory/CPU starved.

Developer playgrounds should be able to build the main branch. It is okay to use ./install-deps.sh top-level script from the ceph source tree to update dependencies. Do not run that script from an older release of Ceph as it may break other developer's work. If you need to build an older release, lock a throwaway node like smithi and build there. Or, use a container to do the build/testing!

Using the developer machines to look at teuthology QA artifacts is encouraged. Try to avoid using a text editor to look at large (1GB+) debug logs as this can be RAM intensive/disruptive. Instead, prefer less or use tail -c xM | $EDITOR - to look at portions of the log in a text editor.

Many of the developer playground nodes have extra disks for testing Ceph. It's okay to use these for vstart clusters but it may be more flexible to build LVM volumes on top of these devices so others may use them too.

Please use this MOTD for these playground machines:

*******************************************************************************

                                Welcome!

      This machine is a Ceph Developer Playground for shared use.  Please
          see the following wiki document for guidelines and a list of
                              available machines.

             https://wiki.sepia.ceph.com/doku.php?id=devplayground

                           Create support tickets at:

                     https://tracker.ceph.com/projects/lab

                                    Thanks!

*******************************************************************************

The following script can be run to setup CephFS mounts on a new developer playground machines:

#!/bin/bash

HOST="$1"

function run {
  printf '%s\n' "$*" >&2
  "$@"
}

function mssh {
  run ssh "$HOST" -- "$*"
}

mssh sudo mkdir -p -m 755 /etc/ceph
ssh reesi001.front.sepia.ceph.com 'env CEPH_KEYRING=/etc/ceph/client.sepian.keyring ceph --id sepian config generate-minimal-conf' | mssh sudo tee /etc/ceph/ceph.conf
mssh sudo chmod 644 /etc/ceph/ceph.conf
ssh reesi001.front.sepia.ceph.com 'cat /etc/ceph/client.sepian.keyring' | mssh sudo tee /etc/ceph/client.sepian.keyring
mssh sudo chmod 644 /etc/ceph/client.sepian.keyring
ssh reesi001.front.sepia.ceph.com 'ceph-authtool /etc/ceph/client.sepian.keyring -n client.sepian -p' | mssh sudo tee /etc/ceph/client.sepian.secret
mssh sudo chmod 600 /etc/ceph/client.sepian.secret


function genmount {
  local secret=$(mssh sudo cat /etc/ceph/client.sepian.secret)
  # create mountpoint
  mssh sudo mkdir -p -- "$2"
  # make the mountpoint directory (shadowed) unwriteable to prevent accidental modification
  mssh sudo chmod 000 -- "$2"
  # set it immutable to enforce that even for root
  mssh sudo chattr +i -- "$2"
  printf '172.21.2.201,172.21.2.202,172.21.2.203:%s\t%s\tceph\tname=sepian,secret=%s,mds_namespace=%s,_netdev\t0\t2\n' "$1" "$2" "$secret" "$3" | mssh sudo tee -a /etc/fstab
}

genmount /teuthology-archive /teuthology teuthology
genmount / /scratch scratch
genmount / /postfile postfile

mssh sudo systemctl daemon-reload
mssh sudo mount /teuthology
mssh sudo mount /scratch
mssh sudo mount /postfile
mssh sudo ln -s /teuthology /a

Configure the dev playground node to schedule jobs:

sudo tee /etc/teuthology.yaml <<EOF
default_machine_type: smithi
queue_host: teuthology.front.sepia.ceph.com
queue_port: 11300
active_machine_types:
  - smithi
EOF

Note: killing a run is (generally) still necessary on teuthology VM. This is because teuthology-kill requires killing the test processes running there.

When setting up a fresh Developer Playground machine, configure an LVM VolumeGroup for use by users. Volumes can be provisioned for a build directory, OSD block device, or anything else needed.

Note: no redundancy is configured below (i.e. RAID). If a disk is lost, all volumes will be affected.

sudo pvcreate /dev/$DISK

Do this for every disk. This is an ad-hoc process because all nodes are different. Also, some disks may have been used in the past so they will need wiped first:

sudo wipefs -a /dev/$DISK

Once all disks are added as physical volumes, it's then possible to add to a VolumeGroup:

sudo vgcreate DevPlayground $DISKS

Finally make a volume for yourself:

sudo lvcreate -L 256G DevPlayground -n $(whoami)-build
sudo mkfs.xfs /dev/DevPlayground/$(whoami)-build
mkdir $HOME/build
chmod 000 $HOME/build
sudo chattr +i $HOME/build
echo "/dev/DevPlayground/$(whoami)-build $HOME/build xfs defaults 1 1" | sudo tee -a /etc/fstab
sudo systemctl daemon-reload
sudo mount $HOME/build

and some OSD block devices:

for i in `seq 0 8`; do sudo lvcreate -L 16G DevPlayground -n $(whoami)-osd.$i ; done

Make those OSDs owned by you:

printf 'ENV{DM_VG_NAME}=="DevPlayground" ENV{DM_LV_NAME}=="%s-*" OWNER="%s" GROUP="users"\n' $(whoami) $(whoami) | sudo tee -a /etc/udev/rules.d/99-lvmowner.rules
sudo udevadm control --reload-rules
sudo udevadm trigger

Then you can use those devices with vstart.sh:

wipefs -a /dev/DevPlayground/$(whoami)-osd.*
shred -v -n 0 -z -s 16M /dev/DevPlayground/$(whoami)-osd.*
env OSD=8 ~/ceph/src/vstart.sh \
    --bluestore-devs $(echo /dev/DevPlayground/$(whoami)-osd.* | tr ' ' ',')

Feel free to make any other volumes that you require.