Work in progress sofree.us 2025Q3 Ceph handson

Timeframe

Per the title, Q3 of 2025. Some Saturday. Probably August or September.

Goals

Learners will install/configure a working Ceph cluster and gain some knowledge of its components. The hard way. No automatic container deployment or similar.

• Generate a cluster UUID
• Select server IP addresses
  • A quick diversion into public and cluster (private) networks as used by Ceph
• Install Ceph application packages
• Create a very basic cluster config file
• use monmaptool to create a single entry cluster monitor map (monmaptool --create --fsid UUID-FROM-ABOVE
• deploy a single monitor for the cluster (ceph-mon --mkfs ...)
• Start the monitor up, check on cluster state
• Add additional monitors on other nodes
• Install manager components
• Add OSD components (At last, some actual storage!)
• Create some storage pools.
• On another (client) system, make use of some of that storage. RBD is the easiest to demonstrate. RGW needs additional components.
• Choose your own adventure:
  • RGW, the Ceph S3 object gateway (and also OpenStack Swift)
    • Install radosgw
    • Create a user (radosgw-admin)
    • Make sure it is available over the internet (this is going to be pre-work on Aaron's part)
    • Install a client program
  • CephFS
    • Install MDS components
    • Create a filesystem
    • Create some client identities
    • Put client keyring on a client server
    • Mount the filesystem and start writing files
• Stretch goals
  • Replicate RBD data from one cluster to another
  • Replicate RGW data from one cluster to another
  • Replicate CephFS data from one cluster to another

Hardware available

The following is at Aaron's home data center, currently unused:

• 6x Dell R630 servers, each with:
  • 2x Intel Xeon E5-2630v4 processors
  • 64 Gibytes of RAM
  • 2x 800Gbyte SAS SSDs (for OS, applications, and Ceph Bluestore WAL/DB devices)
  • 4x 1.8Tbyte SAS HDDs (Ceph cluster data disks)
  • 1x 2-port Mellanox 56Gbits/sec Ethernet card (and sufficient switch ports and cabling for these)
  • 2x Intel 10GbaseT twisted pair Ethernet ports (I do not have switch ports for 10GbaseT, but have plenty of 1000baseT)
  • 2x Intel 1000baseT twisted pair Ethernet ports (and sufficient switch ports and cabling)
  • iDRAC remote console, power control, etc
• 3x Dell R720XD servers, each with:
  • 2x (1x?) Intel Xeon E5-26?? processors
  • 32 Gibytes of RAM
  • 2x 120Gbyte SATA SSDs (rear-mounted) for the OS and applications
  • 6 or 7x 3Tbyte SAS HDDs (Ceph cluster data disks)
  • 2x 400Gbyte SAS SSDs (Ceph Bluestore WAL/DB devices)
  • 2x 10Gbits/sec SFP+ (fiber optic) Ethernet adapters (and sufficient switch ports and cabling for these)
  • 2X 1000baseT twisted pair Ethernet ports (and sufficient switch ports and cabling)
  • iDRAC remote console, power control, etc

Potential Cluster end states

• Option 1: A single cluster with all 9 servers as members. This would probably make most sense with only 2 or 3 learners.
• Option 2: Two clusters. One with mixed hardware consisting of 3x R630 servers and 3x R720 servers. And a second cluster of 3x R630 servers. Makes sense with 4-6 learners in groups of 2 or 3. And demonstrates Ceph's flexibility in building from non-identical servers.
• Option 3: Two clusters. But this time all nodes in each cluster are similar. 6x R630s in one cluster. And 3x R720s in the second cluster. Makes sense with 4-6 learners in groups of 2 or 3.
• Option 4: Three clusters. Each cluster with consistent hardware. Makes most sense with a third group of learners.