From 587cbf1c555a0c953e89bbb384f8dda5406e5c20 Mon Sep 17 00:00:00 2001 From: Frederick Muriuki Muriithi Date: Mon, 15 Jul 2024 10:15:53 -0500 Subject: Rename script. --- scripts/assign_data_to_default_admin.py | 434 ++++++++++++++++++++++++++++++++ scripts/migrate_existing_data.py | 434 -------------------------------- 2 files changed, 434 insertions(+), 434 deletions(-) create mode 100644 scripts/assign_data_to_default_admin.py delete mode 100644 scripts/migrate_existing_data.py (limited to 'scripts') diff --git a/scripts/assign_data_to_default_admin.py b/scripts/assign_data_to_default_admin.py new file mode 100644 index 0000000..0ae209a --- /dev/null +++ b/scripts/assign_data_to_default_admin.py @@ -0,0 +1,434 @@ +""" +Assign any existing data (that is not currently assigned to any group) to the +default sys-admin group for accessibility purposes. +""" +import sys +import json +import time +import random +import logging +from pathlib import Path +from uuid import UUID, uuid4 + +import click +from MySQLdb.cursors import DictCursor + +from gn_auth.auth.db import mariadb as biodb + +import gn_auth.auth.db.sqlite3 as authdb +from gn_auth.auth.authentication.users import User +from gn_auth.auth.authorisation.roles.models import ( + revoke_user_role_by_name, assign_user_role_by_name) + +from gn_auth.auth.authorisation.resources.groups.models import ( + Group, save_group, add_resources_to_group) +from gn_auth.auth.authorisation.resources.models import ( + Resource, ResourceCategory, __assign_resource_owner_role__) + + +class DataNotFound(Exception): + """Raise if no admin user exists.""" + + +def sys_admins(conn: authdb.DbConnection) -> tuple[User, ...]: + """Retrieve all the existing system admins.""" + with authdb.cursor(conn) as cursor: + cursor.execute( + "SELECT u.* FROM users AS u " + "INNER JOIN user_roles AS ur ON u.user_id=ur.user_id " + "INNER JOIN roles AS r ON ur.role_id=r.role_id " + "WHERE r.role_name='system-administrator'") + return tuple(User.from_sqlite3_row(row) for row in cursor.fetchall()) + return tuple() + + +def choose_admin(enum_admins: dict[int, User]) -> int: + """Prompt and read user choice.""" + while True: + try: + print("\n===========================\n") + print("We found the following system administrators:") + for idx, admin in enum_admins.items(): + print(f"\t{idx}: {admin.name} ({admin.email})") + choice = input(f"Choose [1 .. {len(enum_admins)}]: ") + return int(choice) + except ValueError as _verr: + if choice.lower() == "quit": + print("Goodbye!") + sys.exit(0) + print(f"\nERROR: Invalid choice '{choice}'!") + + +def select_sys_admin(admins: tuple[User, ...]) -> User: + """Pick one admin out of list.""" + if len(admins) > 0: + if len(admins) == 1: + print(f"-> Found Admin: {admins[0].name} ({admins[0].email})") + return admins[0] + enum_admins = dict(enumerate(admins, start=1)) + chosen = enum_admins[choose_admin(enum_admins)] + print(f"-> Chosen Admin: {chosen.name} ({chosen.email})") + return chosen + raise DataNotFound( + "No administrator user found. Create an administrator user first.") + + +def admin_group(conn: authdb.DbConnection, admin: User) -> Group: + """Retrieve the admin's user group. If none exist, create one.""" + with authdb.cursor(conn) as cursor: + cursor.execute( + "SELECT g.* FROM users AS u " + "INNER JOIN group_users AS gu ON u.user_id=gu.user_id " + "INNER JOIN groups AS g on gu.group_id=g.group_id " + "WHERE u.user_id = ?", + (str(admin.user_id),)) + row = cursor.fetchone() + if row: + return Group(UUID(row["group_id"]), + row["group_name"], + json.loads(row["group_metadata"])) + new_group = save_group(cursor, "AutoAdminGroup", { + "group_description": ( + "Created by script for existing data visibility. " + "Existing data was migrated into this group and assigned " + "to publicly visible resources according to type.") + }) + + cursor.execute( + "SELECT * FROM resource_categories WHERE " + "resource_category_key='group'") + res_cat_id = cursor.fetchone()["resource_category_id"] + grp_res = { + "group_id": str(new_group.group_id), + "resource_id": str(uuid4()), + "resource_name": new_group.group_name, + "resource_category_id": res_cat_id, + "public": 0 + } + cursor.execute( + "INSERT INTO resources VALUES " + "(:resource_id, :resource_name, :resource_category_id, :public)", + grp_res) + cursor.execute( + "INSERT INTO group_resources(resource_id, group_id) " + "VALUES(:resource_id, :group_id)", + grp_res) + cursor.execute("INSERT INTO group_users VALUES (?, ?)", + (str(new_group.group_id), str(admin.user_id))) + revoke_user_role_by_name(cursor, admin, "group-creator") + assign_user_role_by_name( + cursor, admin, UUID(grp_res["resource_id"]), "group-leader") + return new_group + + +def __resource_category_by_key__( + cursor: authdb.DbCursor, category_key: str) -> ResourceCategory: + """Retrieve a resource category by its ID.""" + cursor.execute( + "SELECT * FROM resource_categories WHERE resource_category_key = ?", + (category_key,)) + row = cursor.fetchone() + if not bool(row): + raise DataNotFound( + f"Could not find resource category with key {category_key}") + return ResourceCategory(UUID(row["resource_category_id"]), + row["resource_category_key"], + row["resource_category_description"]) + + +def __create_resources__(cursor: authdb.DbCursor) -> tuple[Resource, ...]: + """Create default resources.""" + resources = tuple(Resource( + uuid4(), name, __resource_category_by_key__(cursor, catkey), + True, tuple() + ) for name, catkey in ( + ("mRNA-euhrin", "mrna"), + ("pheno-xboecp", "phenotype"), + ("geno-welphd", "genotype"))) + cursor.executemany( + "INSERT INTO resources VALUES (:rid, :rname, :rcid, :pub)", + tuple({ + "rid": str(res.resource_id), + "rname": res.resource_name, + "rcid": str(res.resource_category.resource_category_id), + "pub": 1 + } for res in resources)) + return resources + + +def default_resources(conn: authdb.DbConnection, group: Group) -> tuple[ + Resource, ...]: + """Create default resources, or return them if they exist.""" + with authdb.cursor(conn) as cursor: + cursor.execute( + "SELECT r.resource_id, r.resource_name, r.public, rc.* " + "FROM resource_ownership AS ro INNER JOIN resources AS r " + "ON ro.resource_id=r.resource_id " + "INNER JOIN resource_categories AS rc " + "ON r.resource_category_id=rc.resource_category_id " + "WHERE ro.group_id=? AND r.resource_name IN " + "('mRNA-euhrin', 'pheno-xboecp', 'geno-welphd')", + (str(group.group_id),)) + rows = cursor.fetchall() + if len(rows) == 0: + return __create_resources__(cursor) + + return tuple(Resource( + UUID(row["resource_id"]), + row["resource_name"], + ResourceCategory( + UUID(row["resource_category_id"]), + row["resource_category_key"], + row["resource_category_description"]), + bool(row["public"]), + tuple() + ) for row in rows) + + +def delay(): + """Delay a while: anything from 2 seconds to 15 seconds.""" + time.sleep(random.choice(range(2,16))) + + +def __assigned_mrna__(authconn): + """Retrieve assigned mRNA items.""" + with authdb.cursor(authconn) as cursor: + cursor.execute( + "SELECT SpeciesId, InbredSetId, ProbeFreezeId, ProbeSetFreezeId " + "FROM linked_mrna_data") + return tuple( + (row["SpeciesId"], row["InbredSetId"], row["ProbeFreezeId"], + row["ProbeSetFreezeId"]) for row in cursor.fetchall()) + + +def __unassigned_mrna__(bioconn, assigned): + """Retrieve unassigned mRNA data items.""" + query = ( + "SELECT s.SpeciesId, iset.InbredSetId, pf.ProbeFreezeId, " + "psf.Id AS ProbeSetFreezeId, psf.Name AS dataset_name, " + "psf.FullName AS dataset_fullname, psf.ShortName AS dataset_shortname " + "FROM Species AS s INNER JOIN InbredSet AS iset " + "ON s.SpeciesId=iset.SpeciesId INNER JOIN ProbeFreeze AS pf " + "ON iset.InbredSetId=pf.InbredSetId INNER JOIN ProbeSetFreeze AS psf " + "ON pf.ProbeFreezeId=psf.ProbeFreezeId " + "WHERE s.Name != 'human' ") + if len(assigned) > 0: + paramstr = ", ".join(["(%s, %s, %s, %s)"] * len(assigned)) + query = query + ( + "AND (s.SpeciesId, iset.InbredSetId, pf.ProbeFreezeId, psf.Id) " + f"NOT IN ({paramstr}) ") + + query = query + "LIMIT 100000" + with bioconn.cursor(DictCursor) as cursor: + cursor.execute(query, tuple(item for row in assigned for item in row)) + return (row for row in cursor.fetchall()) + + +def __assign_mrna__(authconn, bioconn, resource, group): + "Assign any unassigned mRNA data to resource." + while True: + unassigned = tuple({ + "data_link_id": str(uuid4()), + "group_id": str(group.group_id), + "resource_id": str(resource.resource_id), + **row + } for row in __unassigned_mrna__( + bioconn, __assigned_mrna__(authconn))) + + if len(unassigned) <= 0: + print("-> mRNA: Completed!") + break + with authdb.cursor(authconn) as cursor: + cursor.executemany( + "INSERT INTO linked_mrna_data VALUES " + "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " + ":ProbeFreezeId, :ProbeSetFreezeId, :dataset_name, " + ":dataset_fullname, :dataset_shortname)", + unassigned) + cursor.executemany( + "INSERT INTO mrna_resources VALUES " + "(:resource_id, :data_link_id)", + unassigned) + print(f"-> mRNA: Linked {len(unassigned)}") + delay() + + +def __assigned_geno__(authconn): + """Retrieve assigned genotype data.""" + with authdb.cursor(authconn) as cursor: + cursor.execute( + "SELECT SpeciesId, InbredSetId, GenoFreezeId " + "FROM linked_genotype_data") + return tuple((row["SpeciesId"], row["InbredSetId"], row["GenoFreezeId"]) + for row in cursor.fetchall()) + +def __unassigned_geno__(bioconn, assigned): + """Fetch unassigned genotype data.""" + query = ( + "SELECT s.SpeciesId, iset.InbredSetId, iset.InbredSetName, " + "gf.Id AS GenoFreezeId, gf.Name AS dataset_name, " + "gf.FullName AS dataset_fullname, " + "gf.ShortName AS dataset_shortname " + "FROM Species AS s INNER JOIN InbredSet AS iset " + "ON s.SpeciesId=iset.SpeciesId INNER JOIN GenoFreeze AS gf " + "ON iset.InbredSetId=gf.InbredSetId " + "WHERE s.Name != 'human' ") + if len(assigned) > 0: + paramstr = ", ".join(["(%s, %s, %s)"] * len(assigned)) + query = query + ( + "AND (s.SpeciesId, iset.InbredSetId, gf.Id) " + f"NOT IN ({paramstr}) ") + + query = query + "LIMIT 100000" + with bioconn.cursor(DictCursor) as cursor: + cursor.execute(query, tuple(item for row in assigned for item in row)) + return (row for row in cursor.fetchall()) + + +def __assign_geno__(authconn, bioconn, resource, group): + "Assign any unassigned Genotype data to resource." + while True: + unassigned = tuple({ + "data_link_id": str(uuid4()), + "group_id": str(group.group_id), + "resource_id": str(resource.resource_id), + **row + } for row in __unassigned_geno__( + bioconn, __assigned_geno__(authconn))) + + if len(unassigned) <= 0: + print("-> Genotype: Completed!") + break + with authdb.cursor(authconn) as cursor: + cursor.executemany( + "INSERT INTO linked_genotype_data VALUES " + "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " + ":GenoFreezeId, :dataset_name, :dataset_fullname, " + ":dataset_shortname)", + unassigned) + cursor.executemany( + "INSERT INTO genotype_resources VALUES " + "(:resource_id, :data_link_id)", + unassigned) + print(f"-> Genotype: Linked {len(unassigned)}") + delay() + + +def __assigned_pheno__(authconn): + """Retrieve assigned phenotype data.""" + with authdb.cursor(authconn) as cursor: + cursor.execute( + "SELECT SpeciesId, InbredSetId, PublishFreezeId, PublishXRefId " + "FROM linked_phenotype_data") + return tuple(( + row["SpeciesId"], row["InbredSetId"], row["PublishFreezeId"], + row["PublishXRefId"]) for row in cursor.fetchall()) + + +def __unassigned_pheno__(bioconn, assigned): + """Retrieve all unassigned Phenotype data.""" + query = ( + "SELECT spc.SpeciesId, iset.InbredSetId, " + "pf.Id AS PublishFreezeId, pf.Name AS dataset_name, " + "pf.FullName AS dataset_fullname, " + "pf.ShortName AS dataset_shortname, pxr.Id AS PublishXRefId " + "FROM " + "Species AS spc " + "INNER JOIN InbredSet AS iset " + "ON spc.SpeciesId=iset.SpeciesId " + "INNER JOIN PublishFreeze AS pf " + "ON iset.InbredSetId=pf.InbredSetId " + "INNER JOIN PublishXRef AS pxr " + "ON pf.InbredSetId=pxr.InbredSetId " + "WHERE spc.Name != 'human' ") + if len(assigned) > 0: + paramstr = ", ".join(["(%s, %s, %s, %s)"] * len(assigned)) + query = query + ( + "AND (spc.SpeciesId, iset.InbredSetId, pf.Id, pxr.Id) " + f"NOT IN ({paramstr}) ") + + query = query + "LIMIT 100000" + with bioconn.cursor(DictCursor) as cursor: + cursor.execute(query, tuple(item for row in assigned for item in row)) + return (row for row in cursor.fetchall()) + + +def __assign_pheno__(authconn, bioconn, resource, group): + """Assign any unassigned Phenotype data to resource.""" + while True: + unassigned = tuple({ + "data_link_id": str(uuid4()), + "group_id": str(group.group_id), + "resource_id": str(resource.resource_id), + **row + } for row in __unassigned_pheno__( + bioconn, __assigned_pheno__(authconn))) + + if len(unassigned) <= 0: + print("-> Phenotype: Completed!") + break + with authdb.cursor(authconn) as cursor: + cursor.executemany( + "INSERT INTO linked_phenotype_data VALUES " + "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " + ":PublishFreezeId, :dataset_name, :dataset_fullname, " + ":dataset_shortname, :PublishXRefId)", + unassigned) + cursor.executemany( + "INSERT INTO phenotype_resources VALUES " + "(:resource_id, :data_link_id)", + unassigned) + print(f"-> Phenotype: Linked {len(unassigned)}") + delay() + + +def assign_data_to_resource( + authconn, bioconn, resource: Resource, group: Group): + """Assign existing data, not linked to any group to the resource.""" + assigner_fns = { + "mrna": __assign_mrna__, + "genotype": __assign_geno__, + "phenotype": __assign_pheno__ + } + return assigner_fns[resource.resource_category.resource_category_key]( + authconn, bioconn, resource, group) + + +def entry(authdbpath, mysqldburi): + """Entry-point for data migration.""" + if not Path(authdbpath).exists(): + print( + f"ERROR: Auth db file `{authdbpath}` does not exist.", + file=sys.stderr) + sys.exit(2) + try: + with (authdb.connection(authdbpath) as authconn, + biodb.database_connection(mysqldburi) as bioconn): + admin = select_sys_admin(sys_admins(authconn)) + the_admin_group = admin_group(authconn, admin) + resources = default_resources(authconn, the_admin_group) + add_resources_to_group(authconn, resources, the_admin_group) + for resource in resources: + assign_data_to_resource( + authconn, bioconn, resource, the_admin_group) + with authdb.cursor(authconn) as cursor: + __assign_resource_owner_role__(cursor, resource, admin) + except DataNotFound as dnf: + print(dnf.args[0], file=sys.stderr) + sys.exit(1) + + +@click.command() +@click.argument("authdbpath") # "Path to the Auth(entic|oris)ation database" +@click.argument("mysqldburi") # "URI to the MySQL database with the biology data" +@click.option("--loglevel", default="WARNING", show_default=True, + type=click.Choice(["CRITICAL", "ERROR", "WARNING", "INFO", "DEBUG"])) +def run(authdbpath, mysqldburi, loglevel): + """Setup command-line arguments.""" + globallogger = logging.getLogger() + globallogger.setLevel(loglevel) + entry(authdbpath, mysqldburi) + + +if __name__ == "__main__": + run() # pylint: disable=[no-value-for-parameter] diff --git a/scripts/migrate_existing_data.py b/scripts/migrate_existing_data.py deleted file mode 100644 index 198d37d..0000000 --- a/scripts/migrate_existing_data.py +++ /dev/null @@ -1,434 +0,0 @@ -""" -Migrate existing data that is not assigned to any group to the default sys-admin -group for accessibility purposes. -""" -import sys -import json -import time -import random -import logging -from pathlib import Path -from uuid import UUID, uuid4 - -import click -from MySQLdb.cursors import DictCursor - -from gn_auth.auth.db import mariadb as biodb - -import gn_auth.auth.db.sqlite3 as authdb -from gn_auth.auth.authentication.users import User -from gn_auth.auth.authorisation.roles.models import ( - revoke_user_role_by_name, assign_user_role_by_name) - -from gn_auth.auth.authorisation.resources.groups.models import ( - Group, save_group, add_resources_to_group) -from gn_auth.auth.authorisation.resources.models import ( - Resource, ResourceCategory, __assign_resource_owner_role__) - - -class DataNotFound(Exception): - """Raise if no admin user exists.""" - - -def sys_admins(conn: authdb.DbConnection) -> tuple[User, ...]: - """Retrieve all the existing system admins.""" - with authdb.cursor(conn) as cursor: - cursor.execute( - "SELECT u.* FROM users AS u " - "INNER JOIN user_roles AS ur ON u.user_id=ur.user_id " - "INNER JOIN roles AS r ON ur.role_id=r.role_id " - "WHERE r.role_name='system-administrator'") - return tuple(User.from_sqlite3_row(row) for row in cursor.fetchall()) - return tuple() - - -def choose_admin(enum_admins: dict[int, User]) -> int: - """Prompt and read user choice.""" - while True: - try: - print("\n===========================\n") - print("We found the following system administrators:") - for idx, admin in enum_admins.items(): - print(f"\t{idx}: {admin.name} ({admin.email})") - choice = input(f"Choose [1 .. {len(enum_admins)}]: ") - return int(choice) - except ValueError as _verr: - if choice.lower() == "quit": - print("Goodbye!") - sys.exit(0) - print(f"\nERROR: Invalid choice '{choice}'!") - - -def select_sys_admin(admins: tuple[User, ...]) -> User: - """Pick one admin out of list.""" - if len(admins) > 0: - if len(admins) == 1: - print(f"-> Found Admin: {admins[0].name} ({admins[0].email})") - return admins[0] - enum_admins = dict(enumerate(admins, start=1)) - chosen = enum_admins[choose_admin(enum_admins)] - print(f"-> Chosen Admin: {chosen.name} ({chosen.email})") - return chosen - raise DataNotFound( - "No administrator user found. Create an administrator user first.") - - -def admin_group(conn: authdb.DbConnection, admin: User) -> Group: - """Retrieve the admin's user group. If none exist, create one.""" - with authdb.cursor(conn) as cursor: - cursor.execute( - "SELECT g.* FROM users AS u " - "INNER JOIN group_users AS gu ON u.user_id=gu.user_id " - "INNER JOIN groups AS g on gu.group_id=g.group_id " - "WHERE u.user_id = ?", - (str(admin.user_id),)) - row = cursor.fetchone() - if row: - return Group(UUID(row["group_id"]), - row["group_name"], - json.loads(row["group_metadata"])) - new_group = save_group(cursor, "AutoAdminGroup", { - "group_description": ( - "Created by script for existing data visibility. " - "Existing data was migrated into this group and assigned " - "to publicly visible resources according to type.") - }) - - cursor.execute( - "SELECT * FROM resource_categories WHERE " - "resource_category_key='group'") - res_cat_id = cursor.fetchone()["resource_category_id"] - grp_res = { - "group_id": str(new_group.group_id), - "resource_id": str(uuid4()), - "resource_name": new_group.group_name, - "resource_category_id": res_cat_id, - "public": 0 - } - cursor.execute( - "INSERT INTO resources VALUES " - "(:resource_id, :resource_name, :resource_category_id, :public)", - grp_res) - cursor.execute( - "INSERT INTO group_resources(resource_id, group_id) " - "VALUES(:resource_id, :group_id)", - grp_res) - cursor.execute("INSERT INTO group_users VALUES (?, ?)", - (str(new_group.group_id), str(admin.user_id))) - revoke_user_role_by_name(cursor, admin, "group-creator") - assign_user_role_by_name( - cursor, admin, UUID(grp_res["resource_id"]), "group-leader") - return new_group - - -def __resource_category_by_key__( - cursor: authdb.DbCursor, category_key: str) -> ResourceCategory: - """Retrieve a resource category by its ID.""" - cursor.execute( - "SELECT * FROM resource_categories WHERE resource_category_key = ?", - (category_key,)) - row = cursor.fetchone() - if not bool(row): - raise DataNotFound( - f"Could not find resource category with key {category_key}") - return ResourceCategory(UUID(row["resource_category_id"]), - row["resource_category_key"], - row["resource_category_description"]) - - -def __create_resources__(cursor: authdb.DbCursor) -> tuple[Resource, ...]: - """Create default resources.""" - resources = tuple(Resource( - uuid4(), name, __resource_category_by_key__(cursor, catkey), - True, tuple() - ) for name, catkey in ( - ("mRNA-euhrin", "mrna"), - ("pheno-xboecp", "phenotype"), - ("geno-welphd", "genotype"))) - cursor.executemany( - "INSERT INTO resources VALUES (:rid, :rname, :rcid, :pub)", - tuple({ - "rid": str(res.resource_id), - "rname": res.resource_name, - "rcid": str(res.resource_category.resource_category_id), - "pub": 1 - } for res in resources)) - return resources - - -def default_resources(conn: authdb.DbConnection, group: Group) -> tuple[ - Resource, ...]: - """Create default resources, or return them if they exist.""" - with authdb.cursor(conn) as cursor: - cursor.execute( - "SELECT r.resource_id, r.resource_name, r.public, rc.* " - "FROM resource_ownership AS ro INNER JOIN resources AS r " - "ON ro.resource_id=r.resource_id " - "INNER JOIN resource_categories AS rc " - "ON r.resource_category_id=rc.resource_category_id " - "WHERE ro.group_id=? AND r.resource_name IN " - "('mRNA-euhrin', 'pheno-xboecp', 'geno-welphd')", - (str(group.group_id),)) - rows = cursor.fetchall() - if len(rows) == 0: - return __create_resources__(cursor) - - return tuple(Resource( - UUID(row["resource_id"]), - row["resource_name"], - ResourceCategory( - UUID(row["resource_category_id"]), - row["resource_category_key"], - row["resource_category_description"]), - bool(row["public"]), - tuple() - ) for row in rows) - - -def delay(): - """Delay a while: anything from 2 seconds to 15 seconds.""" - time.sleep(random.choice(range(2,16))) - - -def __assigned_mrna__(authconn): - """Retrieve assigned mRNA items.""" - with authdb.cursor(authconn) as cursor: - cursor.execute( - "SELECT SpeciesId, InbredSetId, ProbeFreezeId, ProbeSetFreezeId " - "FROM linked_mrna_data") - return tuple( - (row["SpeciesId"], row["InbredSetId"], row["ProbeFreezeId"], - row["ProbeSetFreezeId"]) for row in cursor.fetchall()) - - -def __unassigned_mrna__(bioconn, assigned): - """Retrieve unassigned mRNA data items.""" - query = ( - "SELECT s.SpeciesId, iset.InbredSetId, pf.ProbeFreezeId, " - "psf.Id AS ProbeSetFreezeId, psf.Name AS dataset_name, " - "psf.FullName AS dataset_fullname, psf.ShortName AS dataset_shortname " - "FROM Species AS s INNER JOIN InbredSet AS iset " - "ON s.SpeciesId=iset.SpeciesId INNER JOIN ProbeFreeze AS pf " - "ON iset.InbredSetId=pf.InbredSetId INNER JOIN ProbeSetFreeze AS psf " - "ON pf.ProbeFreezeId=psf.ProbeFreezeId " - "WHERE s.Name != 'human' ") - if len(assigned) > 0: - paramstr = ", ".join(["(%s, %s, %s, %s)"] * len(assigned)) - query = query + ( - "AND (s.SpeciesId, iset.InbredSetId, pf.ProbeFreezeId, psf.Id) " - f"NOT IN ({paramstr}) ") - - query = query + "LIMIT 100000" - with bioconn.cursor(DictCursor) as cursor: - cursor.execute(query, tuple(item for row in assigned for item in row)) - return (row for row in cursor.fetchall()) - - -def __assign_mrna__(authconn, bioconn, resource, group): - "Assign any unassigned mRNA data to resource." - while True: - unassigned = tuple({ - "data_link_id": str(uuid4()), - "group_id": str(group.group_id), - "resource_id": str(resource.resource_id), - **row - } for row in __unassigned_mrna__( - bioconn, __assigned_mrna__(authconn))) - - if len(unassigned) <= 0: - print("-> mRNA: Completed!") - break - with authdb.cursor(authconn) as cursor: - cursor.executemany( - "INSERT INTO linked_mrna_data VALUES " - "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " - ":ProbeFreezeId, :ProbeSetFreezeId, :dataset_name, " - ":dataset_fullname, :dataset_shortname)", - unassigned) - cursor.executemany( - "INSERT INTO mrna_resources VALUES " - "(:resource_id, :data_link_id)", - unassigned) - print(f"-> mRNA: Linked {len(unassigned)}") - delay() - - -def __assigned_geno__(authconn): - """Retrieve assigned genotype data.""" - with authdb.cursor(authconn) as cursor: - cursor.execute( - "SELECT SpeciesId, InbredSetId, GenoFreezeId " - "FROM linked_genotype_data") - return tuple((row["SpeciesId"], row["InbredSetId"], row["GenoFreezeId"]) - for row in cursor.fetchall()) - -def __unassigned_geno__(bioconn, assigned): - """Fetch unassigned genotype data.""" - query = ( - "SELECT s.SpeciesId, iset.InbredSetId, iset.InbredSetName, " - "gf.Id AS GenoFreezeId, gf.Name AS dataset_name, " - "gf.FullName AS dataset_fullname, " - "gf.ShortName AS dataset_shortname " - "FROM Species AS s INNER JOIN InbredSet AS iset " - "ON s.SpeciesId=iset.SpeciesId INNER JOIN GenoFreeze AS gf " - "ON iset.InbredSetId=gf.InbredSetId " - "WHERE s.Name != 'human' ") - if len(assigned) > 0: - paramstr = ", ".join(["(%s, %s, %s)"] * len(assigned)) - query = query + ( - "AND (s.SpeciesId, iset.InbredSetId, gf.Id) " - f"NOT IN ({paramstr}) ") - - query = query + "LIMIT 100000" - with bioconn.cursor(DictCursor) as cursor: - cursor.execute(query, tuple(item for row in assigned for item in row)) - return (row for row in cursor.fetchall()) - - -def __assign_geno__(authconn, bioconn, resource, group): - "Assign any unassigned Genotype data to resource." - while True: - unassigned = tuple({ - "data_link_id": str(uuid4()), - "group_id": str(group.group_id), - "resource_id": str(resource.resource_id), - **row - } for row in __unassigned_geno__( - bioconn, __assigned_geno__(authconn))) - - if len(unassigned) <= 0: - print("-> Genotype: Completed!") - break - with authdb.cursor(authconn) as cursor: - cursor.executemany( - "INSERT INTO linked_genotype_data VALUES " - "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " - ":GenoFreezeId, :dataset_name, :dataset_fullname, " - ":dataset_shortname)", - unassigned) - cursor.executemany( - "INSERT INTO genotype_resources VALUES " - "(:resource_id, :data_link_id)", - unassigned) - print(f"-> Genotype: Linked {len(unassigned)}") - delay() - - -def __assigned_pheno__(authconn): - """Retrieve assigned phenotype data.""" - with authdb.cursor(authconn) as cursor: - cursor.execute( - "SELECT SpeciesId, InbredSetId, PublishFreezeId, PublishXRefId " - "FROM linked_phenotype_data") - return tuple(( - row["SpeciesId"], row["InbredSetId"], row["PublishFreezeId"], - row["PublishXRefId"]) for row in cursor.fetchall()) - - -def __unassigned_pheno__(bioconn, assigned): - """Retrieve all unassigned Phenotype data.""" - query = ( - "SELECT spc.SpeciesId, iset.InbredSetId, " - "pf.Id AS PublishFreezeId, pf.Name AS dataset_name, " - "pf.FullName AS dataset_fullname, " - "pf.ShortName AS dataset_shortname, pxr.Id AS PublishXRefId " - "FROM " - "Species AS spc " - "INNER JOIN InbredSet AS iset " - "ON spc.SpeciesId=iset.SpeciesId " - "INNER JOIN PublishFreeze AS pf " - "ON iset.InbredSetId=pf.InbredSetId " - "INNER JOIN PublishXRef AS pxr " - "ON pf.InbredSetId=pxr.InbredSetId " - "WHERE spc.Name != 'human' ") - if len(assigned) > 0: - paramstr = ", ".join(["(%s, %s, %s, %s)"] * len(assigned)) - query = query + ( - "AND (spc.SpeciesId, iset.InbredSetId, pf.Id, pxr.Id) " - f"NOT IN ({paramstr}) ") - - query = query + "LIMIT 100000" - with bioconn.cursor(DictCursor) as cursor: - cursor.execute(query, tuple(item for row in assigned for item in row)) - return (row for row in cursor.fetchall()) - - -def __assign_pheno__(authconn, bioconn, resource, group): - """Assign any unassigned Phenotype data to resource.""" - while True: - unassigned = tuple({ - "data_link_id": str(uuid4()), - "group_id": str(group.group_id), - "resource_id": str(resource.resource_id), - **row - } for row in __unassigned_pheno__( - bioconn, __assigned_pheno__(authconn))) - - if len(unassigned) <= 0: - print("-> Phenotype: Completed!") - break - with authdb.cursor(authconn) as cursor: - cursor.executemany( - "INSERT INTO linked_phenotype_data VALUES " - "(:data_link_id, :group_id, :SpeciesId, :InbredSetId, " - ":PublishFreezeId, :dataset_name, :dataset_fullname, " - ":dataset_shortname, :PublishXRefId)", - unassigned) - cursor.executemany( - "INSERT INTO phenotype_resources VALUES " - "(:resource_id, :data_link_id)", - unassigned) - print(f"-> Phenotype: Linked {len(unassigned)}") - delay() - - -def assign_data_to_resource( - authconn, bioconn, resource: Resource, group: Group): - """Assign existing data, not linked to any group to the resource.""" - assigner_fns = { - "mrna": __assign_mrna__, - "genotype": __assign_geno__, - "phenotype": __assign_pheno__ - } - return assigner_fns[resource.resource_category.resource_category_key]( - authconn, bioconn, resource, group) - - -def entry(authdbpath, mysqldburi): - """Entry-point for data migration.""" - if not Path(authdbpath).exists(): - print( - f"ERROR: Auth db file `{authdbpath}` does not exist.", - file=sys.stderr) - sys.exit(2) - try: - with (authdb.connection(authdbpath) as authconn, - biodb.database_connection(mysqldburi) as bioconn): - admin = select_sys_admin(sys_admins(authconn)) - the_admin_group = admin_group(authconn, admin) - resources = default_resources(authconn, the_admin_group) - add_resources_to_group(authconn, resources, the_admin_group) - for resource in resources: - assign_data_to_resource( - authconn, bioconn, resource, the_admin_group) - with authdb.cursor(authconn) as cursor: - __assign_resource_owner_role__(cursor, resource, admin) - except DataNotFound as dnf: - print(dnf.args[0], file=sys.stderr) - sys.exit(1) - - -@click.command() -@click.argument("authdbpath") # "Path to the Auth(entic|oris)ation database" -@click.argument("mysqldburi") # "URI to the MySQL database with the biology data" -@click.option("--loglevel", default="WARNING", show_default=True, - type=click.Choice(["CRITICAL", "ERROR", "WARNING", "INFO", "DEBUG"])) -def run(authdbpath, mysqldburi, loglevel): - """Setup command-line arguments.""" - globallogger = logging.getLogger() - globallogger.setLevel(loglevel) - entry(authdbpath, mysqldburi) - - -if __name__ == "__main__": - run() # pylint: disable=[no-value-for-parameter] -- cgit v1.2.3