xtract-sdk 0.0.7a4

A package used for downloading and processing files from multiple habitats.

Xtract SDK v0.0.7a3

Login: Creating an XtractClient object

First, we import the XtractClient class from the Xtract SDK

from xtract_sdk.client import XtractClient

Here we create an XtractClient object to request tokens from Globus Auth.

xtr = XtractClient(auth_scopes=[scope_1, ..., scope_n], force=False)

While additional auth scopes may be added with the auth_scopes argument, there are a number of default scopes automatically requested within the system. These are:

• openid: provides username for identity.
• search: interact with Globus Search
• funcx_scope: orchestrate the metadata exraction at the given funcX endpoint.
• petrel: read or write data on Petrel. Not needed if no data going to Petrel.
• transfer: needed to crawl the Globus endpoint and transfer metadata to its final location.
• funcx_scope: needed to orchestrate the metadata exraction at the given funcX endpoint.

Additional auth scopes can be added with the auth_scopes argument.

When true, dev makes you go through the full authorization flow again.

Defining endpoints: Creating an XtractEndpoint object

** Endpoints ** in Xtract are the computing fabric that enable us to move files and apply extractors to files. To this end, an Xtract endpoint is the combination of the following two software endpoints:

• Globus endpoints [required] enable us to access all file system metadata about files stored on an endpoint, and enables us to transfer files between machines for more-efficient processing.
• FuncX endpoints [optional] are capable of remotely receiving extraction functions that can be applied to files on the Globus endpoint. Note that the absence of a funcX endpoint on an Xtract endpoint means that a file must be transferred to an endpoint with a valid funcX endpoint in able to have its metadata extracted.

In order to create an Xtract endpoint, we first import the XtractEndpoint class from the Xtract SDK

from xtract_sdk.endpoint import XtractEndpoint

Here we create two XtractEndpoint objects to be used later in a crawl, etc.

xep1 = XtractEndpoint(repo_type='globus',
                      globus_ep_id='aaaa-0000-3333',
                      funcx_ep_id='aaaa-0000-3333',
                      dirs=['str1', 'str2', ..., 'strn'], 
                      grouper='file_is_group')

xep2 = XtractEndpoint(repo_type='globus',
                      globus_ep_id='aaaa-0000-3333',
                      dirs=['str1', 'str2', ..., 'strn'], 
                      grouper='file_is_group')

The arguments are as follow:

• repo_type: (str) at this point, only Globus is accepted. Google Drive and others will be made available at a later date.
• globus_ep_id: (uuid) the Globus endpoint ID.
• funcx_ep_id: (uuid) optional funcX endpoint ID.
• dirs: (listof(str)) directory paths on Globus endpoint for where the data reside.
• grouper: (str) grouping strategy for files.

The XtractEndpoint can also be given a funcx_ep_id.

Crawling

xtr.crawl([xep_1, ..., xep_n])

Where [xep_1, ..., xep_n] is a list of XtractEndpoint objects.

The crawl ID for each endpoint will be stored in the XtractClient object as a list xtr.crawl_ids.

Behind the scenes, this will scan a Globus directory breadth-first (using globus_ls), first extracting physical metadata such as path, size, and extension. Next, since the grouper we selected is 'file_is_group', the crawler will simply create n single-file groups.

The crawl is non-blocking, and the crawl_id here will be used to execute and monitor downstream extraction processes.

Getting Crawl status

crawl_statuses = xtr.get_crawl_status(crawl_ids=None)

We can get crawl status, seeing how many groups have been identified in the crawl. If xtr.crawl() has already been run, then xtr.get_crawl_status() will get the status of the IDs stored in xtr.crawl_ids. Otherwise, a list of crawl_ids may be given to xtr.get_crawl_status().

This will return a dictionary resembling:

{‘crawl_id’: String,
 ‘status’: String, 
 ‘message’: “OK” if everything is fine otherwise describes error,
 ‘data’: {'bytes_crawled': Integer, ..., 'files_crawled': Integer}}

Note that measuring the total files yet to crawl is impossible, as the BFS may not have discovered all files yet, and Globus does not yet have a file counting feature for all directories and subdirectories. I.e., we know when we're done, but we do not know until we get there.

Warning: it currently takes up to 30 seconds for a crawl to start. Why? Container warming time.

Flushing Crawl metadata

xtr.flush_crawl_metadata(crawl_ids=None)

After running a crawl, we can use xtr.flush_crawl_metadata() to return a list of all metadata from the crawl.

Similarly with .get_crawl_status(), if xtr.crawl() has already been run, then xtr.flush_crawl_metadata() will get the status of the IDs stored in xtr.crawl_ids. Otherwise, a list of crawl_ids may be given to xtr.flush_crawl_metadata().

Flushing crawl metadata will return a dictionary resembling:

{"crawl_id": String,
 "file_ls": List,
 "num_files": Integer,
 "queue_empty": Boolean}

Packaging: Creating file group objects for downstream data packaging.

Here we have a number of data elements that need to be defined.

Group: a collection of files and extractor to be executed on the collection of files. Groups need not have mutually exclusive sets of files. A group is also intitialized with an empty metadata dictionary.

from xtract_sdk.packagers import Group, Family, FamilyBatch

group1 = Group(files=[{'path': 12388550508, 'mdata': {}}], parser='image'}

group2 = Group(files=[{'path': 14546663235, 'mdata': {}}], parser='image'}

Family: a collection of groups such that all files in the group are mutually exclusive. This grouping is useful for parallelizing the download and extraction of separate groups that may contain the same files. You can easily add Group objects to a family as follows:

fam = Family()

fam.add_group(group1) fam.add_group(group2)

FamilyBatch: If the number of groups in a family is generally small, you may want to batch the families to increase application performance. In order to do this, you may add any number of families to a FamilyBatch that enables users to iterate over families. The family_batch can be directly read by the downloader that will batch-fetch all files therein.

fam_batch=FamilyBatch()

fam_batch.add_family(fam)

Downloaders: coming soon!