1
0
forked from ScoDoc/ScoDoc
ScoDoc-Front/tools/import_scodoc7_dept.py

523 lines
17 KiB
Python
Raw Normal View History

# -*- mode: python -*-
# -*- coding: utf-8 -*-
import inspect
import logging
import pdb
2021-08-18 18:53:45 +02:00
import time
import psycopg2
import sqlalchemy
from sqlalchemy import func
from flask import current_app
from app import db
from app.auth.models import User, get_super_admin
import app
2021-08-15 09:33:18 +02:00
from app import clear_scodoc_cache
from app import models
from app.models import APO_CODE_STR_LEN, SHORT_STR_LEN, GROUPNAME_STR_LEN
from app.scodoc import notesdb as ndb
def truncate_field(table_name, field, max_len):
"renvoie une fonction de troncation"
def troncator(value):
"Si la chaine est trop longue pour la nouvelle base, émet un warning et tronque"
if value and len(value) > max_len:
logging.warning(
"Chaine trop longue tronquée: %s.%s=%s", table_name, field, value
)
return value[:max_len]
return value
return troncator
# Attributs dont le nom change entre les bases ScoDoc 7 et 9:
# (None indique que l'attribut est supprimé, "nouveau_nom" qu'il change de nom)
ATTRIBUTES_MAPPING = {
"admissions": {
"debouche": None,
},
"adresse": {
"entreprise_id": None,
},
"etud_annotations": {
"zope_authenticated_user": "author",
"zope_remote_addr": None,
},
"identite": {
"foto": None,
},
"notes_formsemestre": {
"etape_apo2": None, # => suppressed
"etape_apo3": None,
"etape_apo4": None,
# préférences, plus dans formsemestre:
# (inutilisés depuis ScoDoc 6 environ)
"bul_show_decision": None,
"bul_show_uevalid": None,
"nomgroupetd": None,
"nomgroupetp": None,
"nomgroupeta": None,
"gestion_absence": None,
"bul_show_codemodules": None,
"bul_show_rangs": None,
"bul_show_ue_rangs": None,
"bul_show_mod_rangs": None,
},
"partition": {
"compute_ranks": None,
},
"notes_appreciations": {
"zope_authenticated_user": "author",
"zope_remote_addr": None,
},
"scolog": {
"remote_addr": None,
"remote_host": None,
},
}
# Attributs à transformer pour passer de ScoDoc 7 à 9
# la fonction est appliquée au nouvel attribut
ATTRIBUTES_TRANSFORM = {
"notes_formsemestre": {
# la modalité CP est devenue CPRO
"modalite": lambda x: x if x != "CP" else "CPRO",
"bul_bgcolor": truncate_field(
"notes_formsemestre", "bul_bgcolor", SHORT_STR_LEN
),
},
# tronque les codes trop longs pour être honnêtes...
"notes_formations": {
"formation_code": truncate_field(
"notes_formations", "formation_code", SHORT_STR_LEN
),
"code_specialite": truncate_field(
"notes_formations", "code_specialite", SHORT_STR_LEN
),
},
"notes_ue": {
"ue_code": truncate_field("notes_ue", "ue_code", SHORT_STR_LEN),
"code_apogee": truncate_field("notes_ue", "code_apogee", APO_CODE_STR_LEN),
},
"notes_modules": {
"code_apogee": truncate_field("notes_modules", "code_apogee", APO_CODE_STR_LEN),
},
"notes_formsemestre_etapes": {
"etape_apo": truncate_field(
"notes_formsemestre_etapes", "etape_apo", APO_CODE_STR_LEN
),
},
"notes_form_modalites": {
"modalite": truncate_field("notes_form_modalites", "modalite", SHORT_STR_LEN),
},
"notes_formsemestre_inscription": {
"etape": truncate_field(
"notes_formsemestre_inscription", "etape", APO_CODE_STR_LEN
),
},
"partition": {
"partition_name": truncate_field("partition", "partition_name", SHORT_STR_LEN),
},
"group_descr": {
"group_name": truncate_field("group_descr", "group_name", GROUPNAME_STR_LEN),
},
"scolar_autorisation_inscription": {
"formation_code": truncate_field(
"scolar_autorisation_inscription", "formation_code", SHORT_STR_LEN
),
},
}
def setup_log(dept_acronym: str):
2021-08-26 14:24:25 +02:00
"""log to console (stderr) and /opt/scodoc-data/log/migration79.log"""
log_formatter = logging.Formatter(
"%(asctime)s %(levelname)s (" + dept_acronym + ") %(message)s"
)
# Log to file:
logger = logging.getLogger()
2021-08-26 14:24:25 +02:00
file_handler = logging.FileHandler("/opt/scodoc-data/log/migration79.log")
file_handler.setFormatter(log_formatter)
logger.addHandler(file_handler)
# Log to stderr:
console_handler = logging.StreamHandler() # stderr
console_handler.setFormatter(log_formatter)
logger.addHandler(console_handler)
# Caution:
logger.setLevel(logging.DEBUG)
def import_scodoc7_dept(dept_id: str, dept_db_uri=None):
"""Importe un département ScoDoc7 dans ScoDoc >= 8.1
(base de donnée unique)
Args:
dept_id: acronyme du département ("RT")
dept_db_uri: URI de la base ScoDoc7eg "postgresql:///SCORT"
si None, utilise postgresql:///SCO{dept_id}
"""
dept = models.Departement.query.filter_by(acronym=dept_id).first()
if dept:
raise ValueError(f"le département {dept_id} existe déjà !")
if dept_db_uri is None:
dept_db_uri = f"postgresql:///SCO{dept_id}"
setup_log(dept_id)
2021-08-15 09:33:18 +02:00
logging.info(f"connecting to database {dept_db_uri}")
cnx = psycopg2.connect(dept_db_uri)
cursor = cnx.cursor(cursor_factory=ndb.ScoDocCursor)
# Create dept:
dept = models.Departement(acronym=dept_id, description="migré de ScoDoc7")
db.session.add(dept)
db.session.commit()
#
id_from_scodoc7 = {} # { scodoc7id (str) : scodoc8 id (int)}
# Utilisateur de rattachement par défaut:
default_user = get_super_admin()
#
2021-08-18 18:53:45 +02:00
t0 = time.time()
for (table, id_name) in SCO7_TABLES_ORDONNEES:
2021-08-15 09:33:18 +02:00
logging.info(f"{dept.acronym}: converting {table}...")
klass = get_class_for_table(table)
2021-08-18 18:53:45 +02:00
t1 = time.time()
n = convert_table(dept, cursor, id_from_scodoc7, klass, id_name, default_user)
2021-08-18 19:36:30 +02:00
logging.info(f" inserted {n} objects in {time.time()-t1:3.2f}s.")
2021-08-18 18:53:45 +02:00
2021-08-18 19:36:30 +02:00
logging.info(f"All table imported in {time.time()-t0:3.2f}s")
2021-08-18 18:53:45 +02:00
logging.info(f"clearing app caches...")
2021-08-15 09:33:18 +02:00
clear_scodoc_cache()
logging.info(f"Done.")
logging.warning(f"Un redémarrage du serveur postgresql est conseillé.")
def get_class_for_table(table):
"""Return ScoDoc orm class for the given SQL table: search in our models"""
for name in dir(models):
item = getattr(models, name)
if inspect.isclass(item):
if issubclass(item, db.Model):
if item.__tablename__ == table:
return item
try: # pour les db.Table qui ne sont pas des classes (isclass est faux !)
if item.name == table:
return item
except:
pass
raise ValueError(f"No model for table {table}")
def get_boolean_columns(klass):
"return list of names of boolean attributes in this model"
boolean_columns = []
column_names = sqlalchemy.inspect(klass).columns.keys()
for column_name in column_names:
column = getattr(klass, column_name)
if isinstance(column.expression.type, sqlalchemy.sql.sqltypes.Boolean):
boolean_columns.append(column_name)
return boolean_columns
def get_table_max_id(klass):
"return max id in this Table (or -1 if no id)"
if not id in sqlalchemy.inspect(klass).columns.keys():
return -1
sql_table = str(klass.description)
cnx = db.engine.connect()
r = cnx.execute("SELECT max(id) FROM " + sql_table)
r.fetchone()
if r:
return r[0]
else: # empty table
return 0
def update_table_sequence(table_name):
"""After filling the table, we need to update the serial
so that the next insertions will use new ids
"""
with db.engine.connect() as cnx:
cnx.execute(
f"""SELECT
setval('{table_name}_id_seq',
(SELECT MAX(id) FROM {table_name}))
"""
)
def convert_table(
dept, cursor, id_from_scodoc7: dict, klass=None, id_name=None, default_user=None
):
"converti les élements d'une table scodoc7"
# Est-ce une Table ou un Model dans l'ORM ?
if isinstance(klass, sqlalchemy.sql.schema.Table):
is_table = True
current_id = get_table_max_id(klass)
has_id = current_id != -1
table_name = str(klass.description)
boolean_columns = []
else:
is_table = False
has_id = True
table_name = klass.__tablename__
# Colonnes booléennes (valeurs à convertir depuis int)
boolean_columns = get_boolean_columns(klass)
# Part de l'id le plus haut actuellement présent
# (évidemment, nous sommes les seuls connectés à la base destination !)
current_id = db.session.query(func.max(klass.id)).first()
if (current_id is None) or (current_id[0] is None):
current_id = 0
else:
current_id = current_id[0]
2021-08-18 18:53:45 +02:00
cnx = db.engine.connect()
# mapping: login (scodoc7) : user id (scodoc8)
login2id = {u.user_name: u.id for u in User.query}
# les tables ont le même nom dans les deux versions de ScoDoc:
cursor.execute(f"SELECT * FROM {table_name}")
objects = cursor.dictfetchall()
2021-08-18 18:53:45 +02:00
n = 0
for obj in objects:
current_id += 1
convert_object(
current_id,
dept,
obj,
has_id,
id_from_scodoc7,
klass,
is_table,
id_name,
boolean_columns,
login2id,
default_user,
cnx,
)
2021-08-18 18:53:45 +02:00
# commit progressif pour ne pas consommer trop de mémoire:
n += 1
if (not n % 1000) and cnx:
db.session.commit()
if cnx:
cnx.close()
2021-08-18 19:36:30 +02:00
db.session.commit() # écrit la table
if has_id:
update_table_sequence(table_name)
return len(objects)
def convert_object(
new_id,
dept,
obj: dict,
has_id: bool = True,
id_from_scodoc7: dict = None,
klass=None,
is_table: bool = False,
id_name=None,
boolean_columns=None,
login2id=None,
default_user=None,
2021-08-18 18:53:45 +02:00
cnx=None, # cnx à la base destination
):
# Supprime l'id ScoDoc7 (eg "formsemestre_id") qui deviendra "id"
if id_name:
old_id = obj[id_name]
del obj[id_name]
if hasattr(klass, "scodoc7_id"):
obj["scodoc7_id"] = old_id
else:
old_id = None # tables ScoDoc7 sans id
if is_table:
table_name = str(klass.description)
else:
table_name = klass.__tablename__
# Les champs contant des id utilisateurs:
# chaine login en ScoDoc7, uid numérique en ScoDoc 8+
USER_REFS = {"responsable_id", "ens_id", "uid"}
if not is_table:
# Supprime les attributs obsoletes (très anciennes versions de ScoDoc):
attributs = ATTRIBUTES_MAPPING.get(table_name, {})
# renomme ou supprime les attributs
for k in attributs.keys() & obj.keys():
v = attributs[k]
if v is not None:
obj[v] = obj[k]
del obj[k]
# transforme les valeurs: obj[k] = transform(obj[k])
for k in ATTRIBUTES_TRANSFORM.get(table_name, {}):
obj[k] = ATTRIBUTES_TRANSFORM[table_name][k](obj[k])
# map les ids (foreign keys)
for k in obj:
if (k.endswith("id") or k == "object") and k not in USER_REFS | {
"semestre_id",
"sem_id",
"scodoc7_id",
}:
old_ref = obj[k]
if old_ref is not None:
if isinstance(old_ref, str):
old_ref = old_ref.strip()
elif k == "entreprise_id": # id numérique spécial
old_ref = f"entreprises.{old_ref}"
elif k == "entreprise_corresp_id":
old_ref = f"entreprise_correspondant.{old_ref}"
if old_ref == "NULL" or not old_ref: # buggy old entries
new_ref = None
elif old_ref in id_from_scodoc7:
new_ref = id_from_scodoc7[old_ref]
elif (not is_table) and table_name in {
"scolog",
"etud_annotations",
"notes_notes_log",
"scolar_news",
"absences",
"absences_notifications",
"itemsuivi", # etudid n'était pas une clé
2021-09-11 15:59:06 +02:00
"adresse", # etudid n'était pas une clé
"admissions", # idem
"scolar_events",
}:
# tables avec "fausses" clés
# (l'object référencé a pu disparaitre)
new_ref = None
elif is_table and table_name in {
"notes_semset_formsemestre",
}:
# pour anciennes installs où des relations n'avait pas été déclarées clés étrangères
# eg: notes_semset_formsemestre.semset_id n'était pas une clé
# Dans ce cas, mieux vaut supprimer la relation si l'un des objets n'existe pas
return
else:
raise ValueError(f"no new id for {table_name}.{k}='{obj[k]}' !")
obj[k] = new_ref
# Remape les utilisateur: user.id
# S'il n'existe pas, rattache à l'admin
for k in USER_REFS & obj.keys():
login_scodoc7 = obj[k]
uid = login2id.get(login_scodoc7)
if not uid:
uid = default_user.id
warning_user_dont_exist(
login_scodoc7,
f"non existent user: {login_scodoc7}: giving {table_name}({old_id}) to admin",
)
# raise ValueError(f"non existent user: {login_scodoc7}")
obj[k] = uid
# Converti les booléens
for k in boolean_columns:
obj[k] = bool(obj[k])
# Ajoute le département si besoin:
if hasattr(klass, "dept_id"):
obj["dept_id"] = dept.id
# Fixe l'id (ainsi nous évitons d'avoir à commit() après chaque entrée)
if has_id:
obj["id"] = new_id
if is_table:
statement = sqlalchemy.insert(klass).values(**obj)
_ = cnx.execute(statement)
else:
2021-08-18 19:36:30 +02:00
new_obj = klass(**obj) # ORM object
db.session.add(new_obj)
# insert_object(cnx, table_name, obj)
# Stocke l'id pour les références (foreign keys):
if id_name and has_id:
if isinstance(old_id, int):
# les id int étaient utilisés pour les "entreprises"
old_id = table_name + "." + str(old_id)
id_from_scodoc7[old_id] = new_id
MISSING_USERS = set() # login ScoDoc7 référencés mais non existants...
def warning_user_dont_exist(login_scodoc7, msg):
if login_scodoc7 not in MISSING_USERS:
return
MISSING_USERS.add(login_scodoc7)
logging.warning(msg)
2021-08-18 18:53:45 +02:00
def insert_object(cnx, table_name: str, vals: dict) -> str:
2021-08-18 19:36:30 +02:00
"""insert tuple in db
version manuelle => ne semble pas plus rapide
"""
2021-08-18 18:53:45 +02:00
cols = list(vals.keys())
colnames = ",".join(cols)
fmt = ",".join(["%%(%s)s" % col for col in cols])
cnx.execute("insert into %s (%s) values (%s)" % (table_name, colnames, fmt), vals)
# tables ordonnées topologiquement pour les clés étrangères:
# g = nx.read_adjlist("misc/model-scodoc7.csv", create_using=nx.DiGraph,delimiter=";")
# L = list(reversed(list(nx.topological_sort(g))))
SCO7_TABLES_ORDONNEES = [
# (table SQL, nom de l'id scodoc7)
("notes_formations", "formation_id"),
("notes_ue", "ue_id"),
("notes_matieres", "matiere_id"),
("notes_formsemestre", "formsemestre_id"),
("notes_modules", "module_id"),
("notes_moduleimpl", "moduleimpl_id"),
(
"notes_modules_enseignants",
"modules_enseignants_id",
), # (relation) avait un id modules_enseignants_id
("partition", "partition_id"),
("identite", "etudid"),
("entreprises", "entreprise_id"),
("notes_evaluation", "evaluation_id"),
("group_descr", "group_id"),
("group_membership", "group_membership_id"), # (relation, qui avait un id)
("notes_semset", "semset_id"),
("notes_tags", "tag_id"),
("itemsuivi", "itemsuivi_id"),
("itemsuivi_tags", "tag_id"),
("adresse", "adresse_id"),
("admissions", "adm_id"),
("absences", ""),
("scolar_news", "news_id"),
("scolog", ""),
("etud_annotations", "id"),
("billet_absence", "billet_id"),
("entreprise_correspondant", "entreprise_corresp_id"),
("entreprise_contact", "entreprise_contact_id"),
("absences_notifications", ""),
# ("notes_form_modalites", "form_modalite_id"), : déjà initialisées
("notes_appreciations", "id"),
("scolar_autorisation_inscription", "autorisation_inscription_id"),
("scolar_formsemestre_validation", "formsemestre_validation_id"),
("scolar_events", "event_id"),
("notes_notes_log", "id"),
("notes_notes", ""),
("notes_moduleimpl_inscription", "moduleimpl_inscription_id"),
("notes_formsemestre_inscription", "formsemestre_inscription_id"),
("notes_formsemestre_custommenu", "custommenu_id"),
(
"notes_formsemestre_ue_computation_expr",
"notes_formsemestre_ue_computation_expr_id",
),
("notes_formsemestre_uecoef", "formsemestre_uecoef_id"),
("notes_semset_formsemestre", ""), # (relation)
("notes_formsemestre_etapes", ""),
("notes_formsemestre_responsables", ""), # (relation)
("notes_modules_tags", ""),
("itemsuivi_tags_assoc", ""), # (relation)
("sco_prefs", "pref_id"),
]
"""
from tools.import_scodoc7_dept import *
import_scodoc7_dept( "RT", "SCORT" )
"""