infrabench / tasks /vm-ceph-bootstrap /instruction.md
xuanmiao-31's picture
Add InfraBench public task materials + dataset card
02dd0fd verified
|
Raw
History Blame Contribute Delete
4.51 kB
You are a Ceph storage administrator. Deploy a Ceph cluster and configure block storage on the following infrastructure.
══════════════════════════════════════════════════════════════════
ENVIRONMENT
══════════════════════════════════════════════════════════════════
Nodes: node0 through node6 (7-node KVM cluster on a private libvirt network)
You are currently on node0. From node0 you have passwordless root SSH to every
other node by short name:
ssh root@node1 ... ssh root@node6
`/etc/hosts` and `/etc/syscraft/cluster-info` are populated on every node with
the mapping `<short-name>=<ip>`. Use those for any IP lookups.
Network:
Single private subnet 192.168.100.0/24 on the only NIC of each VM.
Look up node0's private IP for `--mon-ip`:
awk -F= '/^node0=/{print $2}' /etc/syscraft/cluster-info
Do NOT pass `--cluster-network` — this deployment uses a single network.
Storage:
Each node has TWO unused raw block devices for OSDs:
/dev/loop0 (~30 GiB)
/dev/loop1 (~30 GiB)
The root filesystem is on /dev/vda — do not touch it.
The loop devices are attached automatically at boot by a systemd unit; if
they are missing on a worker for any reason, `losetup -f --show
/var/lib/cephbench/disk_b.img` (and `disk_c.img`) will recreate them.
OS: Ubuntu 24.04 LTS on all nodes.
Container runtime: podman is installed on all nodes.
Pre-installed tooling (ready to use, no installation needed):
cephadm (Ceph Squid, v19.x) — available on all nodes via /usr/sbin/cephadm
podman, lvm2, xfsprogs, chrony, jq — installed on all nodes
Pre-pulled image: quay.io/ceph/ceph:v19.2.0 (Ceph Squid)
This image is already cached in podman on every node.
`--image` is a GLOBAL flag that comes BEFORE the subcommand:
cephadm --image quay.io/ceph/ceph:v19.2.0 bootstrap ...
NOT: cephadm bootstrap --image ... (this will fail)
Node hostnames may be FQDN-shaped. Pass `--allow-fqdn-hostname` to cephadm
bootstrap.
When adding hosts to the cluster, the short name must match the actual
hostname on that node. The nodes' short names are `node0`..`node6`.
Correct usage: `ceph orch host add node1 <node1-ip-from-cluster-info>`
(short name first, then IP as the address)
This is a 7-node cluster — do NOT pass `--single-host-defaults`.
══════════════════════════════════════════════════════════════════
TASKS (all must be completed)
══════════════════════════════════════════════════════════════════
1. Bootstrap a Ceph cluster using cephadm on node0.
2. Join node1 through node6 into the cluster.
3. Add all available data disks (`/dev/loop0` and `/dev/loop1` on each node)
as OSDs.
4. Create a replicated storage pool named "data_rep" with 3 replicas and
application=rbd.
5. Create an erasure-coded storage pool named "data_ec" using:
- Algorithm: reed_sol_van (jerasure plugin)
- k=4, m=2
- Failure domain: host
6. In the data_rep pool, create a 100 GiB RBD block device image named
"bench_image". Map it to a local block device on node0, format it with
xfs, and mount it at /mnt/ceph_data.
══════════════════════════════════════════════════════════════════
SCORING
══════════════════════════════════════════════════════════════════
5 pts — Ceph cluster is reachable (bootstrap succeeded)
15 pts — Cluster health is HEALTH_OK
15 pts — data_rep pool: 3 replicas, type=replicated, application=rbd
15 pts — data_ec pool: k=4 m=2 reed_sol_van jerasure host failure domain
20 pts — bench_image exists in data_rep, size ~100 GiB
20 pts — /mnt/ceph_data is mounted
10 pts — data can be written to /mnt/ceph_data