1
0
mirror of https://github.com/ilri/dspace-statistics-api.git synced 2024-11-24 15:20:18 +01:00
dspace-statistics-api/dspace_statistics_api/stats.py
2021-03-22 13:42:42 +02:00

127 lines
4.7 KiB
Python

# SPDX-License-Identifier: GPL-3.0-only
import requests
from .config import SOLR_SERVER
from .util import get_statistics_shards
def get_views(solr_date_string: str, elements: list, facetField: str):
"""
Get view statistics for a list of elements from Solr. Depending on the req-
uest this could be items, communities, or collections.
:parameter solr_date_string (str): Solr date string, for example "[* TO *]"
:parameter elements (list): a list of IDs
:parameter facetField (str): Solr field to facet by, for example "id"
:returns: A dict of IDs and views
"""
shards = get_statistics_shards()
# Join the UUIDs with "OR" and escape the hyphens for Solr
solr_elements_string: str = " OR ".join(elements).replace("-", r"\-")
solr_query_params = {
"q": f"{facetField}:({solr_elements_string})",
"fq": f"type:2 AND -isBot:true AND statistics_type:view AND time:{solr_date_string}",
"fl": facetField,
"facet": "true",
"facet.field": facetField,
"facet.mincount": 1,
"shards": shards,
"rows": 0,
"wt": "json",
"json.nl": "map", # return facets as a dict instead of a flat list
}
solr_url = SOLR_SERVER + "/statistics/select"
res = requests.get(solr_url, params=solr_query_params)
# Create an empty dict to store views
data = {}
# Solr returns facets as a dict of dicts (see the json.nl parameter)
views = res.json()["facet_counts"]["facet_fields"]
# iterate over the facetField dict and ids and views
for id_, views in views[facetField].items():
# For items we can rely on Solr returning facets for the *only* the ids
# in our query, but for communities and collections, the owningComm and
# owningColl fields are multi-value so Solr will return facets with the
# values in our query as well as *any others* that happen to be present
# in the field (which looks like Solr returning unrelated results until
# you realize that the field is multi-value and this is correct).
#
# To work around this I make sure that each id in the returned dict are
# present in the elements list POSTed by the user.
if id_ in elements:
data[id_] = views
# Check if any ids have missing stats so we can set them to 0
if len(data) < len(elements):
# List comprehension to get a list of ids (keys) in the data
data_ids = [k for k, v in data.items()]
for element_id in elements:
if element_id not in data_ids:
data[element_id] = 0
continue
return data
def get_downloads(solr_date_string: str, elements: list, facetField: str):
"""
Get download statistics for a list of items from Solr. Depending on the req-
uest this could be items, communities, or collections.
:parameter solr_date_string (str): Solr date string, for example "[* TO *]"
:parameter elements (list): a list of IDs
:parameter facetField (str): Solr field to facet by, for example "id"
:returns: A dict of IDs and downloads
"""
shards = get_statistics_shards()
# Join the UUIDs with "OR" and escape the hyphens for Solr
solr_elements_string: str = " OR ".join(elements).replace("-", r"\-")
solr_query_params = {
"q": f"{facetField}:({solr_elements_string})",
"fq": f"type:0 AND -isBot:true AND statistics_type:view AND bundleName:ORIGINAL AND time:{solr_date_string}",
"fl": facetField,
"facet": "true",
"facet.field": facetField,
"facet.mincount": 1,
"shards": shards,
"rows": 0,
"wt": "json",
"json.nl": "map", # return facets as a dict instead of a flat list
}
solr_url = SOLR_SERVER + "/statistics/select"
res = requests.get(solr_url, params=solr_query_params)
# Create an empty dict to store downloads
data = {}
# Solr returns facets as a dict of dicts (see the json.nl parameter)
downloads = res.json()["facet_counts"]["facet_fields"]
# Iterate over the facetField dict and get the ids and downloads
for id_, downloads in downloads[facetField].items():
# Make sure that each id in the returned dict are present in the
# elements list POSTed by the user.
if id_ in elements:
data[id_] = downloads
# Check if any elements have missing stats so we can set them to 0
if len(data) < len(elements):
# List comprehension to get a list of ids (keys) in the data
data_ids = [k for k, v in data.items()]
for element_id in elements:
if element_id not in data_ids:
data[element_id] = 0
continue
return data
# vim: set sw=4 ts=4 expandtab: