cloud-diagnostics-xprof 0.0.6

Abstraction over profile session locations and infrastructure running the analysis.

xprofiler

The xprofiler SDK and CLI tool provides abstraction over profile session locations and infrastructure running the analysis.

This includes allowing users to create and manage VM instances for TensorBoard instances in regards to profiling workloads for GPU and TPU.

Quickstart

Install Dependencies

xprofiler relies on using gcloud.

The first step is to follow the documentation to install.

Running the initial gcloud setup will ensure things like your default project ID are set.

Create a VM Instance for TensorBoard

To create a TensorBoard instance, you must provide a path to a GCS bucket. It is also useful to define your specific zone.

ZONE=us-central1-a
GCS_PATH="gs://example-bucket/my-profile-data"

xprofiler create -z $ZONE -l $GCS_PATH

When the command completes, you will see it return information about the instance created, similar to below:

LOG_PATH                            NAME                                            ZONE
gs://example-bucket/my-profile-data xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef  us-central1-a

This will create a VM instance with TensorBoard installed. Note that this initial startup for TensorBoard will take up to a few minutes (typically less than 5 minutes) if you want to connect to the VM's TensorBoard.

List VM Instances

To list the TensorBoard instances created by xprofiler, you can simply run xprofiler list. However, it's recommended to specify the zone (though not required).

ZONE=us-central1-a

xprofiler list -z $ZONE

This will output something like the following if there are instances matching the list criteria:

LOG_PATH                                   NAME                                            ZONE
gs://example-bucket/my-other-profile-data  xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253  us-central1-a
gs://example-bucket/my-profile-data        xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef  us-central1-a

Note you can specify the GCS bucket to get just that one associated instance:

xprofiler list -l $GCS_PATH

Delete VM Instance

To delete an instance, you'll need to specify either the GCS bucket paths or the VM instances' names. Specifying the zone is required.

# Delete by associated GCS path
xprofiler delete -z $ZONE -l $GCS_PATH

# Delete by VM instance name
VM_NAME="xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253"
xprofiler delete -z $ZONE --vm-name $VM_NAME

Details on xprofiler

Main Command: xprofiler

The xprofiler command has additional subcommands that can be invoked to create VM instances, list VM instances, delete instances, etc. However, the main xprofiler command has some additional options without invoking a subcommand.

xprofiler --help

Gives additional information about using the command including flag options and available subcommands. Also can be called with xprofiler -h.

Note that each subcommand has a --help flag that can give information about that specific subcommand. For example: xprofiler list --help

xprofiler --abbrev ...

When invoking a subcommand, typically there is output related to VM instances involved with the subcommand, usually as a detailed table.

In some cases, a user may only want the relevant information (for example a log directory GCS path or VM name instance). This can be particularly useful in scripting with xprofiler by chaining with other commands.

To assist with this, the --abbrev (or equivalent -a) flag will simply print the relevant item (log directory path or VM instance name).

For example, calling xprofiler list might give the following output:

LOG_PATH                                   NAME                                            ZONE
gs://example-bucket/my-other-profile-data  xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253  us-central1-a
gs://example-bucket/my-profile-data        xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef  us-central1-a

But calling with xprofiler --abbrev list will instead print out an abbreviated form of the above output where each item is displayed on a new line:

xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253
xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef

Subcommand: xprofiler create

This command is used to create a new VM instance for TensorBoard to run with a given profile log directory GCS path.

Usage details:

xprofiler create
  [--help]
  --log-directory GS_PATH
  [--zone ZONE_NAME]
  [--vm-name VM_NAME]
  [--verbose]

At the successful completion of this command, the information regarding the newly created VM instances is printed out like the example below:

LOG_PATH                            NAME                                            ZONE
gs://example-bucket/my-profile-data xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef  us-central1-a

If the xprofiler abbreviation flag is used, then an abbreviated output is given like so:

xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef

xprofiler create --help

This provides the basic usage guide for the xprofiler create subcommand.

Creating a VM Instance

To create a new VM instance, a user must specify a profile log directory path (a GCS path) as in xprofiler create -l gs://example-bucket/my-profile-data. This will create a VM instance associated with the log directory. The instance will also have TensorBoard installed and setup ready for use.

Note that after the VM creation, it might take a few minutes for the VM instance to fully be ready (installing dependencies, launching TensorBoard, etc.)

It is recommended to also provide a zone with --zone or -z but it is optional.

By default, the VM instance's name will be uniquely created prepended with xprofiler-. However, this can be specified with the --vm-name or -n flag to give a specific name to the newly created VM.

Lastly, there is a --verbose or -v flag that will provide information as the xprofiler create subcommand runs.

Subcommand: xprofiler list

This command is used to list a VM instances created by the xprofiler tool.

Usage details:

xprofiler list
  [--help]
  [--zone ZONE_NAME]
  [--log-directory GS_PATH [GS_PATH ...]]
  [--filter FILTER_NAME [FILTER_NAME ...]]
  [--verbose]

At the successful completion of this command, the information of matching VM instances is printed out like the example below:

LOG_PATH                                   NAME                                            ZONE
gs://example-bucket/my-other-profile-data  xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253  us-central1-a
gs://example-bucket/my-profile-data        xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef  us-central1-a

If the xprofiler abbreviation flag is used, then an abbreviated output is given like so:

xprofiler-8187640b-e612-4c47-b4df-59a7fc86b253
xprofiler-ev86r7c5-3d09-xb9b-a8e5-a495f5996eef

xprofiler list --help

This provides the basic usage guide for the xprofiler list subcommand.

Listing Specific Subsets

Note that the xprofiler list command will default to listing all VM instances that have the prefix xprofiler.

However, a specific subset of VM instances can be returned using different options.

Providing Zone

xprofiler list -z $ZONE

Providing the zone is highly recommended since otherwise the command can take a while to search for all relevant VM instances.

Providing GCS Path (Profile Log Directory)

Since xprofiler list is meant to look for VM instances created with xprofiler, it is likely the VM instance of interest is associated with a profile log directory.

To filter for a specific VM instance with an associated log directory, simply use the command like so:

xprofiler list -l $GS_PATH

You can even use multiple log directory paths to find any VMs associated with any of these paths:

xprofiler list -l $GS_PATH_0 $GS_PATH_1 $GS_PATH_2