DBMP (Datera Bare-Metal Provisioner)

What For?

When running workloads against the Datera system, it often takes many steps and familiarity with the Datera interface to get a simple volume up and running with a workload. If you live your life in the CLI and don't want to memorize the REST API calls necessary to get this done or don't feel like rolling your own with our Python-SDK (wink-wink, nudge-nudge :) ), then this package was made for you!

What Do?

• Install prerequistes
  • Install python
  • Install git
  • Install open-iscsi (we need iscsiadm)
  • Install multipath-tools (or whatever it is on your system)
  • Install mkfs.<your_favorite_format>
  • Install fio (make sure it's accessible via $PATH)
• Clone the repository
  • git clone http://github.com/Datera/dbmp
  • cd dbmp
• Install
  • ./install.py
• Create Datera Universal Config File
  • vi datera-config.json

  {
      "username": "your_user",
      "password": "your_password",
      "tenant": "/root",
      "mgmt_ip": "1.1.1.1",
      "api_version": "2.2"
  }

• Use
  • ./dbmp --help

What Really Do?

Functionality can be broken down into a few categories

• Health Check
• List Resources
• Show Resources From URL
• Simple Creation
• Complex Creation
• Deletion
• Login/Logout
• Mount/Unmount
• Load Generation
• Metrics

Resource Inspection

DBMP can list resources of certain types or show a resource at a specific URL path.

Example:

./dbmp --list volumes-detail

Or

./dbmp --show-at-url /app_instances/my-app-instance

When using --show-at-url if the url is not a resource, but a container like /app_instances then all the resources at that path are listed

NOTE: Alerts and Events can ONLY be accessed via a --list operation such as --list events-system.

Creation

DBMP can create and delete Datera volumes via a simple CLI interface.

Example:

$ ./dbmp --volume size=5,prefix=my-vol,placement_mode=single_flash,replica=2,count=3,read_iops_max=500

In the above example we are requesting a set of volumes with the following attributes:

• Size In GB (size=5)
• Name Prefix (prefix=my-vol)
• Volume Placement Mode (placement_mode=single_flash) (choices hybrid|single_flash|all_flash)
• Number Of Replicas (replicas=2)
• Total Number Of Volumes (count=3)
• QoS Read IOPS Max (read_iops_max=500)

This will give us the following App Instances on the Datera side each with a single Storage Instance with a single Volume

• my-vol-1
• my-vol-2
• my-vol-3

All attributes are optional and have the following default values:

• size=1
• prefix=$(hostname)
• count=1
• replica=3
• placement_mode=hybrid
• placement_policy=None (This references the name attribute of the policy, eg: all-flash)
• template=None
• read_iops_max=0
• write_iops_max=0
• total_iops_max=0
• read_bw_max=0
• write_bw_max=0
• total_bw_max=0

NOTE: placement_policy will override placement_mode

Complex Creation

Complex App Instance creation is accessed via a JSON file with the following structure:

{
    "name": "complex-app",
    "sis": [
        {"name": "storage-1",
         "vols": [
             {"name": "volume-1",
              "size": 5,
              "replica": 3,
              "placement_mode": "hybrid",
              "qos": {}},
             {"name": "volume-2",
              "size": 10,
              "replica": 3,
              "placement_mode": "hybrid",
              "qos": {}}
         ]},
        {"name": "storage-2",
         "vols": [
             {"name": "volume-3",
              "size": 15,
              "replica": 2,
              "placement_mode": "hybrid",
              "qos": {"total_iops_max": 300}},
             {"name": "volume-4",
              "size": 20,
              "replica": 2,
              "placement_mode": "hybrid",
              "qos": {"total_iops_max": 500}}
         ]}
    ]
}

In the above example I've specified an App Instance with the following: * An App Instance with the name "complex-app" * 2 Storage Instances (storage-1 and storage-2) * 4 Volumes / 2 per Storage Instance (volume-1, volume-2, volume-3, volume-4) * Differing attributes for each volume

Once the JSON file is constructed you can create the complex volume like this:

$ ./dbmp --volume my-complex-vol.json

Deletion

Removal of the above volumes is even simpler

$ ./dbmp --volume prefix=my-vol --clean

Or in the case of a complex volume

$ ./dbmp --volume my-complex-vol.json --clean

Login/Logout

Login and Logout is the same as creation/deletion but you add the `--loginand--logout`` flags (``-clean`` implies logout, and ``--mount/--unmount`` implies login/logout)

When logging in without --mount, all operations afterwards (such as FIO) will be referring to the raw device directly. No filesystem will be created and the device will not be mounted.

Mount/Unmount

Mounting and Unmounting is the same as creation/deletion but you add the --mount and --unmount flags (--clean implies unmount)

The Directory under which mounts should live is specified via the --directory flag

Volumes will be mounted under the following name scheme: /<directory>/<ai_name>/<si_name>/<vol_name> which means a normal CLI specified volume with prefix=my-vol and --directory=/mnt will be located at /mnt/my-vol-storage-1-volume-1

Complex Volumes will follow the same scheme, but as their Storage Instance and Volume names are customizeable, you'll need to consult their JSON schemas to determine their mount locations. The complex volume example shown above would have mounts in the following locations

• /mnt/complex-app-storage-1-volume-1
• /mnt/complex-app-storage-1-volume-2
• /mnt/complex-app-storage-2-volume-3
• /mnt/complex-app-storage-2-volume-4

Load Generation

You can have DBMP generate an FIO file against all created mounts (or devices) via by adding --fio to the invocation.

$ ./dbmp --volume my-complex-vol.json --mount --fio

The default FIO profile is the following:

[global]
randrepeat=0
ioengine=libaio
iodepth=16
direct=1
numjobs=4
runtime=3600
group_reporting
time_based
size=1G

[job-1]
directory=/your_mount

An entry for every mount created will be placed under [job-1]. The complex volume example above would have the following FIO job generated:

[global]
randrepeat=0
ioengine=libaio
iodepth=16
direct=1
numjobs=4
runtime=3600
group_reporting
time_based
size=1G

[job-1]
directory=/mnt/complex-app-storage-1-volume-1
[job-1]
directory=/mnt/complex-app-storage-1-volume-2
[job-1]
directory=/mnt/complex-app-storage-2-volume-3
[job-1]
directory=/mnt/complex-app-storage-2-volume-4

If the --mount flag has not been passed in, then instead of "directory", "filename" will be used with the device

To specify your own FIO workload, use the --fio-workload flag. Your workload MUST only have a [global] section

Metrics

Metrics can be viewed for any volume created by DBMP using the following invocation

$ ./dbmp --volume prefix=my-vol,count=1 --metrics 5,20

Example Output:

{
    "test-vol-0": {
        "iops_write": [
            {
                "value": 110,
                "time": 1532722856000
            },
            {
                "value": 88,
                "time": 1532722860000
            },
            {
                "value": 102,
                "time": 1532722866000
            },
            {
                "value": 101,
                "time": 1532722870000
            }
        ]
    }
}

The --metrics flag takes two comma separated values: interval,timeout which correspond to the interval between metric update calls and the timeout before metric calls should end respectively.

By default the metric will be iops_write, but you can change this using the --metrics-type flag. The available choices are below:

• reads
• writes
• bytes_read,
• bytes_written
• iops_read
• iops_write,
• thpt_read
• thpt_write
• lat_avg_read,
• lat_avg_write
• lat_50_read
• lat_90_read,
• lat_100_read
• lat_50_write,
• lat_90_write
• lat_100_write

Multiple of these can be specified by repeating the --metrics-type flag

Output can be controlled via the --metrics-out-file flag. This just takes a string specifying where the metrics should be saved. The default is metrics-out.json. Use the string 'stdout' to force metrics to be printed to STDOUT instead of a file.

Basic operations can be performed on the data via the --metrics-op flag. The following operations are currently supported (on a per-metric basis)

• None
• average
• max
• min
• total-average (for all app_instances)
• total-max (for all app_instances)
• total-min (for all app_instances)

The output format will change depending on the operation and the original data is NOT preserved.

What Problem?

• File an issue on the github page
• Contact us at [email protected]

What Interactive?

NOTE: This is a WIP

$ ./dbmp --interactive

In every screen you can press SHIFT-down to submit the prompt contents. Multiline prompts do not accept ENTER as a submission key because it instead inserts a newline.

Interactive mode builds a dbmp command to run with vanilla dbmp. If you would like to see the command it constructs, add the --dry-run argument.

$ ./dbmp --interactive --dry-run