cubicweb/server/checkintegrity.py
author Laurent Peuch <cortex@worlddomination.be>
Fri, 12 Apr 2019 13:52:17 +0200
changeset 12574 6ccf8fda063f
parent 12567 26744ad37953
permissions -rw-r--r--
[mod] remove backward compatible code for passlib and force modern version

# copyright 2003-2016 LOGILAB S.A. (Paris, FRANCE), all rights reserved.
# contact http://www.logilab.fr/ -- mailto:contact@logilab.fr
#
# This file is part of CubicWeb.
#
# CubicWeb is free software: you can redistribute it and/or modify it under the
# terms of the GNU Lesser General Public License as published by the Free
# Software Foundation, either version 2.1 of the License, or (at your option)
# any later version.
#
# CubicWeb is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE.  See the GNU Lesser General Public License for more
# details.
#
# You should have received a copy of the GNU Lesser General Public License along
# with CubicWeb.  If not, see <http://www.gnu.org/licenses/>.
"""Integrity checking tool for instances:

* integrity of a CubicWeb repository. Hum actually only the system database is
  checked.
"""

import sys
from datetime import datetime

from logilab.common.shellutils import ProgressBar

from yams.constraints import UniqueConstraint

from cubicweb.toolsutils import underline_title
from cubicweb.schema import PURE_VIRTUAL_RTYPES, VIRTUAL_RTYPES, UNIQUE_CONSTRAINTS
from cubicweb.server.sqlutils import SQL_PREFIX
from cubicweb.server.schema2sql import iter_unique_index_names, build_index_name


def notify_fixed(fix):
    if fix:
        sys.stderr.write(' [FIXED]')
    sys.stderr.write('\n')


def has_eid(cnx, sqlcursor, eid, eids):
    """return true if the eid is a valid eid"""
    if eid in eids:
        return eids[eid]
    sqlcursor.execute('SELECT type FROM entities WHERE eid=%s' % eid)
    try:
        etype = sqlcursor.fetchone()[0]
    except Exception:
        eids[eid] = False
        return False
    if etype not in cnx.vreg.schema:
        eids[eid] = False
        return False
    sqlcursor.execute('SELECT * FROM %s%s WHERE %seid=%s' % (SQL_PREFIX, etype,
                                                             SQL_PREFIX, eid))
    result = sqlcursor.fetchall()
    if len(result) == 0:
        eids[eid] = False
        return False
    elif len(result) > 1:
        msg = ('  More than one entity with eid %s exists in source!\n'
               '  WARNING : Unable to fix this, do it yourself!\n')
        sys.stderr.write(msg % eid)
    eids[eid] = True
    return True


# XXX move to yams?
def etype_fti_containers(eschema, _done=None):
    if _done is None:
        _done = set()
    _done.add(eschema)
    containers = tuple(eschema.fulltext_containers())
    if containers:
        for rschema, target in containers:
            if target == 'object':
                targets = rschema.objects(eschema)
            else:
                targets = rschema.subjects(eschema)
            for targeteschema in targets:
                if targeteschema in _done:
                    continue
                _done.add(targeteschema)
                for container in etype_fti_containers(targeteschema, _done):
                    yield container
    else:
        yield eschema


def reindex_entities(schema, cnx, withpb=True, etypes=None):
    """reindex all entities in the repository"""
    # deactivate modification_date hook since we don't want them
    # to be updated due to the reindexation
    repo = cnx.repo
    dbhelper = repo.system_source.dbhelper
    cursor = cnx.cnxset.cu
    if not dbhelper.has_fti_table(cursor):
        print('no text index table')
        dbhelper.init_fti(cursor)
    repo.system_source.do_fti = True  # ensure full-text indexation is activated
    if etypes is None:
        print('Reindexing entities')
        etypes = set()
        for eschema in schema.entities():
            if eschema.final:
                continue
            indexable_attrs = tuple(eschema.indexable_attributes()) # generator
            if not indexable_attrs:
                continue
            for container in etype_fti_containers(eschema):
                etypes.add(container)
        # clear fti table first
        cnx.system_sql('DELETE FROM %s' % dbhelper.fti_table)
    else:
        print('Reindexing entities of type %s' % \
              ', '.join(sorted(str(e) for e in etypes)))
        # clear fti table first. Use subquery for sql compatibility
        cnx.system_sql("DELETE FROM %s WHERE EXISTS(SELECT 1 FROM ENTITIES "
                       "WHERE eid=%s AND type IN (%s))" % (
                           dbhelper.fti_table, dbhelper.fti_uid_attr,
                           ','.join("'%s'" % etype for etype in etypes)))
    if withpb:
        pb = ProgressBar(len(etypes) + 1)
        pb.update()
    # reindex entities by generating rql queries which set all indexable
    # attribute to their current value
    source = repo.system_source
    for eschema in etypes:
        etype_class = cnx.vreg['etypes'].etype_class(str(eschema))
        for rset in etype_class.cw_fti_index_rql_limit(cnx):
            source.fti_index_entities(cnx, rset.entities())
            # clear entity cache to avoid high memory consumption on big tables
            cnx.drop_entity_cache()
        if withpb:
            pb.update()
    if withpb:
        pb.finish()


_CHECKERS = {}


def _checker(func):
    """Decorator to register a function as a checker for check()."""
    fname = func.__name__
    prefix = 'check_'
    assert fname.startswith(prefix), 'cannot register %s as a checker' % func
    _CHECKERS[fname[len(prefix):]] = func
    return func


@_checker
def check_schema(schema, cnx, eids, fix=1):
    """check serialized schema"""
    print('Checking serialized schema')
    rql = ('Any COUNT(X),RN,SN,ON,CTN GROUPBY RN,SN,ON,CTN ORDERBY 1 '
           'WHERE X is CWConstraint, R constrained_by X, '
           'R relation_type RT, RT name RN, R from_entity ST, ST name SN, '
           'R to_entity OT, OT name ON, X cstrtype CT, CT name CTN')
    for count, rn, sn, on, cstrname in cnx.execute(rql):
        if count == 1:
            continue
        if cstrname in UNIQUE_CONSTRAINTS:
            print("ERROR: got %s %r constraints on relation %s.%s.%s" % (
                count, cstrname, sn, rn, on))
            if fix:
                print('dunno how to fix, do it yourself')


@_checker
def check_text_index(schema, cnx, eids, fix=1):
    """check all entities registered in the text index"""
    print('Checking text index')
    msg = ('  Entity with eid %s exists in the text index but not in any '
           'entity type table (autofix will remove from text index)')
    cursor = cnx.system_sql('SELECT uid FROM appears;')
    for row in cursor.fetchall():
        eid = row[0]
        if not has_eid(cnx, cursor, eid, eids):
            sys.stderr.write(msg % eid)
            if fix:
                cnx.system_sql('DELETE FROM appears WHERE uid=%s;' % eid)
            notify_fixed(fix)


@_checker
def check_entities(schema, cnx, eids, fix=1):
    """check all entities registered in the repo system table"""
    print('Checking entities system table')
    # system table but no source
    msg = ('  Entity %s with eid %s exists in "entities" table but not in any '
           'entity type table (autofix will delete the entity)')
    cursor = cnx.system_sql('SELECT eid,type FROM entities;')
    for row in cursor.fetchall():
        eid, etype = row
        if not has_eid(cnx, cursor, eid, eids):
            sys.stderr.write(msg % (etype, eid))
            if fix:
                cnx.system_sql('DELETE FROM entities WHERE eid=%s;' % eid)
            notify_fixed(fix)
    # source in entities, but no relation cw_source
    # XXX this (get_versions) requires a second connection to the db when we already have one open
    cursor = cnx.system_sql('SELECT e.eid FROM entities as e, cw_CWSource as s '
                            'WHERE NOT EXISTS(SELECT 1 FROM cw_source_relation as cs '
                            '  WHERE cs.eid_from=e.eid) '
                            'ORDER BY e.eid')
    msg = ('  Entity with eid %s is missing relation cw_source (autofix will create the relation)\n')
    for row in cursor.fetchall():
        sys.stderr.write(msg % row[0])
    if fix:
        cnx.system_sql('INSERT INTO cw_source_relation (eid_from, eid_to) '
                       'SELECT e.eid, s.cw_eid FROM entities as e, cw_CWSource as s '
                       "WHERE s.cw_name='system' AND NOT EXISTS(SELECT 1 FROM cw_source_relation as cs "
                       '  WHERE cs.eid_from=e.eid)')
        notify_fixed(True)
    # inconsistencies for 'is'
    msg = '  %s #%s is missing relation "is" (autofix will create the relation)\n'
    cursor = cnx.system_sql('SELECT e.type, e.eid FROM entities as e, cw_CWEType as s '
                                'WHERE s.cw_name=e.type AND NOT EXISTS(SELECT 1 FROM is_relation as cs '
                                '  WHERE cs.eid_from=e.eid AND cs.eid_to=s.cw_eid) '
                                'ORDER BY e.eid')
    for row in cursor.fetchall():
        sys.stderr.write(msg % tuple(row))
    if fix:
        cnx.system_sql('INSERT INTO is_relation (eid_from, eid_to) '
                           'SELECT e.eid, s.cw_eid FROM entities as e, cw_CWEType as s '
                           'WHERE s.cw_name=e.type AND NOT EXISTS(SELECT 1 FROM is_relation as cs '
                           '  WHERE cs.eid_from=e.eid AND cs.eid_to=s.cw_eid)')
        notify_fixed(True)
    # inconsistencies for 'is_instance_of'
    msg = '  %s #%s is missing relation "is_instance_of" (autofix will create the relation)\n'
    cursor = cnx.system_sql('SELECT e.type, e.eid FROM entities as e, cw_CWEType as s '
                                'WHERE s.cw_name=e.type AND NOT EXISTS(SELECT 1 FROM is_instance_of_relation as cs '
                                '  WHERE cs.eid_from=e.eid AND cs.eid_to=s.cw_eid) '
                                'ORDER BY e.eid')
    for row in cursor.fetchall():
        sys.stderr.write(msg % tuple(row))
    if fix:
        cnx.system_sql('INSERT INTO is_instance_of_relation (eid_from, eid_to) '
                           'SELECT e.eid, s.cw_eid FROM entities as e, cw_CWEType as s '
                           'WHERE s.cw_name=e.type AND NOT EXISTS(SELECT 1 FROM is_instance_of_relation as cs '
                           '  WHERE cs.eid_from=e.eid AND cs.eid_to=s.cw_eid)')
        notify_fixed(True)
    print('Checking entities tables')
    msg = ('  Entity with eid %s exists in the %s table but not in "entities" '
           'table (autofix will delete the entity)')
    for eschema in schema.entities():
        if eschema.final:
            continue
        table = SQL_PREFIX + eschema.type
        column = SQL_PREFIX +  'eid'
        cursor = cnx.system_sql('SELECT %s FROM %s;' % (column, table))
        for row in cursor.fetchall():
            eid = row[0]
            # eids is full since we have fetched everything from the entities table,
            # no need to call has_eid
            if not eid in eids or not eids[eid]:
                sys.stderr.write(msg % (eid, eschema.type))
                if fix:
                    cnx.system_sql('DELETE FROM %s WHERE %s=%s;' % (table, column, eid))
                notify_fixed(fix)


def bad_related_msg(rtype, target, eid, fix):
    msg = ('  A relation %(rtype)s with %(target)s eid %(eid)d exists but '
           'entity #(eid)d does not exist')
    sys.stderr.write(msg % {'rtype': rtype, 'target': target, 'eid': eid})
    notify_fixed(fix)


def bad_inlined_msg(rtype, parent_eid, eid, fix):
    msg = ('  An inlined relation %s from %s to %s exists but the latter '
           'entity does not exist')
    sys.stderr.write(msg % (rtype, parent_eid, eid))
    notify_fixed(fix)


@_checker
def check_relations(schema, cnx, eids, fix=1):
    """check that eids referenced by relations are registered in the repo system
    table
    """
    print('Checking relations')
    for rschema in schema.relations():
        if rschema.final or rschema.rule or rschema.type in PURE_VIRTUAL_RTYPES:
            continue
        if rschema.inlined:
            for subjtype in rschema.subjects():
                table = SQL_PREFIX + str(subjtype)
                column = SQL_PREFIX +  str(rschema)
                sql = 'SELECT cw_eid,%s FROM %s WHERE %s IS NOT NULL;' % (
                    column, table, column)
                cursor = cnx.system_sql(sql)
                for row in cursor.fetchall():
                    parent_eid, eid = row
                    if not has_eid(cnx, cursor, eid, eids):
                        bad_inlined_msg(rschema, parent_eid, eid, fix)
                        if fix:
                            sql = 'UPDATE %s SET %s=NULL WHERE %s=%s;' % (
                                table, column, column, eid)
                            cnx.system_sql(sql)
            continue
        try:
            cursor = cnx.system_sql('SELECT eid_from FROM %s_relation;' % rschema)
        except Exception as ex:
            # usually because table doesn't exist
            print('ERROR', ex)
            continue
        for row in cursor.fetchall():
            eid = row[0]
            if not has_eid(cnx, cursor, eid, eids):
                bad_related_msg(rschema, 'subject', eid, fix)
                if fix:
                    sql = 'DELETE FROM %s_relation WHERE eid_from=%s;' % (
                        rschema, eid)
                    cnx.system_sql(sql)
        cursor = cnx.system_sql('SELECT eid_to FROM %s_relation;' % rschema)
        for row in cursor.fetchall():
            eid = row[0]
            if not has_eid(cnx, cursor, eid, eids):
                bad_related_msg(rschema, 'object', eid, fix)
                if fix:
                    sql = 'DELETE FROM %s_relation WHERE eid_to=%s;' % (
                        rschema, eid)
                    cnx.system_sql(sql)


@_checker
def check_mandatory_relations(schema, cnx, eids, fix=1):
    """check entities missing some mandatory relation"""
    print('Checking mandatory relations')
    msg = '%s #%s is missing mandatory %s relation %s (autofix will delete the entity)'
    for rschema in schema.relations():
        if rschema.final or rschema in PURE_VIRTUAL_RTYPES or rschema in ('is', 'is_instance_of'):
            continue
        smandatory = set()
        omandatory = set()
        for rdef in rschema.rdefs.values():
            if rdef.cardinality[0] in '1+':
                smandatory.add(rdef.subject)
            if rdef.cardinality[1] in '1+':
                omandatory.add(rdef.object)
        for role, etypes in (('subject', smandatory), ('object', omandatory)):
            for etype in etypes:
                if role == 'subject':
                    rql = 'Any X WHERE NOT X %s Y, X is %s' % (rschema, etype)
                else:
                    rql = 'Any X WHERE NOT Y %s X, X is %s' % (rschema, etype)
                for entity in cnx.execute(rql).entities():
                    sys.stderr.write(msg % (entity.cw_etype, entity.eid, role, rschema))
                    if fix:
                        entity.cw_delete() # XXX this is BRUTAL!
                    notify_fixed(fix)


@_checker
def check_mandatory_attributes(schema, cnx, eids, fix=1):
    """check for entities stored in the system source missing some mandatory
    attribute
    """
    print('Checking mandatory attributes')
    msg = '%s #%s is missing mandatory attribute %s (autofix will delete the entity)'
    for rschema in schema.relations():
        if not rschema.final or rschema in VIRTUAL_RTYPES:
            continue
        for rdef in rschema.rdefs.values():
            if rdef.cardinality[0] in '1+':
                rql = 'Any X WHERE X %s NULL, X is %s, X cw_source S, S name "system"' % (
                    rschema, rdef.subject)
                for entity in cnx.execute(rql).entities():
                    sys.stderr.write(msg % (entity.cw_etype, entity.eid, rschema))
                    if fix:
                        entity.cw_delete()
                    notify_fixed(fix)


@_checker
def check_metadata(schema, cnx, eids, fix=1):
    """check entities has required metadata

    FIXME: rewrite using RQL queries ?
    """
    print('Checking metadata')
    cursor = cnx.system_sql("SELECT DISTINCT type FROM entities;")
    eidcolumn = SQL_PREFIX + 'eid'
    msg = '  %s with eid %s has no %s (autofix will set it to now)'
    for etype, in cursor.fetchall():
        if etype not in cnx.vreg.schema:
            sys.stderr.write('entities table references unknown type %s\n' %
                             etype)
            if fix:
                cnx.system_sql("DELETE FROM entities WHERE type = %(type)s",
                                   {'type': etype})
            continue
        table = SQL_PREFIX + etype
        for rel, default in ( ('creation_date', datetime.utcnow()),
                              ('modification_date', datetime.utcnow()), ):
            column = SQL_PREFIX + rel
            cursor = cnx.system_sql("SELECT %s FROM %s WHERE %s is NULL"
                                        % (eidcolumn, table, column))
            for eid, in cursor.fetchall():
                sys.stderr.write(msg % (etype, eid, rel))
                if fix:
                    cnx.system_sql("UPDATE %s SET %s=%%(v)s WHERE %s=%s ;"
                                       % (table, column, eidcolumn, eid),
                                       {'v': default})
                notify_fixed(fix)


def check(repo, cnx, checks, reindex, fix, withpb=True):
    """check integrity of instance's repository,
    using given user and password to locally connect to the repository
    (no running cubicweb server needed)
    """
    # yo, launch checks
    if checks:
        eids_cache = {}
        with cnx.security_enabled(read=False, write=False): # ensure no read security
            for check in checks:
                check_func = _CHECKERS[check]
                check_func(repo.schema, cnx, eids_cache, fix=fix)
        if fix:
            cnx.commit()
        else:
            print()
        if not fix:
            print('WARNING: Diagnostic run, nothing has been corrected')
    if reindex:
        cnx.rollback()
        reindex_entities(repo.schema, cnx, withpb=withpb)
        cnx.commit()


SYSTEM_INDEXES = {
    # see cw/server/sources/native.py
    'transactions_tx_time_idx': ('transactions', 'tx_time'),
    'transactions_tx_user_idx': ('transactions', 'tx_user'),
    'tx_entity_actions_txa_action_idx': ('tx_entity_actions', 'txa_action'),
    'tx_entity_actions_txa_public_idx': ('tx_entity_actions', 'txa_public'),
    'tx_entity_actions_eid_idx': ('tx_entity_actions', 'txa_eid'),
    'tx_entity_actions_etype_idx': ('tx_entity_actions', 'txa_etype'),
    'tx_entity_actions_tx_uuid_idx': ('tx_entity_actions', 'tx_uuid'),
    'tx_relation_actions_txa_action_idx': ('tx_relation_actions', 'txa_action'),
    'tx_relation_actions_txa_public_idx': ('tx_relation_actions', 'txa_public'),
    'tx_relation_actions_eid_from_idx': ('tx_relation_actions', 'eid_from'),
    'tx_relation_actions_eid_to_idx': ('tx_relation_actions', 'eid_to'),
    'tx_relation_actions_tx_uuid_idx': ('tx_relation_actions', 'tx_uuid'),
}


def expected_indexes(cnx):
    """Return a dictionary describing indexes expected by the schema {index name: (table, column)}.

    This doesn't include primary key indexes.
    """
    source = cnx.repo.system_source
    dbh = source.dbhelper
    schema = cnx.repo.schema
    schema_indexes = SYSTEM_INDEXES.copy()
    if source.dbdriver == 'postgres':
        schema_indexes.update({'appears_words_idx': ('appears', 'words')})
    else:
        schema_indexes.update({'appears_uid': ('appears', 'uid'),
                               'appears_word_id': ('appears', 'word_id')})
    for rschema in schema.relations():
        if rschema.rule or rschema in PURE_VIRTUAL_RTYPES:
            continue  # computed relation
        if rschema.final or rschema.inlined:
            for rdef in rschema.rdefs.values():
                table = 'cw_{0}'.format(rdef.subject)
                column = 'cw_{0}'.format(rdef.rtype)
                if any(isinstance(cstr, UniqueConstraint) for cstr in rdef.constraints):
                    schema_indexes[dbh._index_name(table, column, unique=True)] = (
                        table, [column])
                if rschema.inlined or rdef.indexed:
                    schema_indexes[dbh._index_name(table, column)] = (table, [column])
        else:
            table = '{0}_relation'.format(rschema)
            if source.dbdriver == 'postgres':
                # index built after the primary key constraint
                schema_indexes[build_index_name(table, ['eid_from', 'eid_to'], 'key_')] = (
                    table, ['eid_from', 'eid_to'])
            schema_indexes[build_index_name(table, ['eid_from'], 'idx_')] = (
                table, ['eid_from'])
            schema_indexes[build_index_name(table, ['eid_to'], 'idx_')] = (
                table, ['eid_to'])
    for eschema in schema.entities():
        if eschema.final:
            continue
        table = 'cw_{0}'.format(eschema)
        for columns, index_name in iter_unique_index_names(eschema):
            schema_indexes[index_name] = (table, columns)

    return schema_indexes


def database_indexes(cnx):
    """Return a set of indexes found in the database, excluding primary key indexes."""
    source = cnx.repo.system_source
    dbh = source.dbhelper
    if source.dbdriver == 'postgres':

        def index_filter(idx):
            return not (idx.startswith('pg_') or '_pkey' in idx or '_p_key' in idx
                        or idx.endswith('_key'))
    else:

        def index_filter(idx):
            return not idx.startswith('sqlite_')

    return set(idx for idx in dbh.list_indices(cnx.cnxset.cu)
               if index_filter(idx))


def check_indexes(cnx):
    """Check indexes of a system database: output missing expected indexes as well as unexpected ones.

    Return 0 if there is no differences, else 1.
    """
    schema_indexes = expected_indexes(cnx)
    db_indexes = database_indexes(cnx)

    missing_indexes = set(schema_indexes) - db_indexes
    if missing_indexes:
        print(underline_title('Missing indexes'))
        print('index expected by the schema but not found in the database:\n')
        missing = ['{0} ON {1[0]} {1[1]}'.format(idx, schema_indexes[idx])
                   for idx in missing_indexes]
        print('\n'.join(sorted(missing)))
        print()
        status = 1

    additional_indexes = db_indexes - set(schema_indexes)
    if additional_indexes:
        print(underline_title('Additional indexes'))
        print('index in the database but not expected by the schema:\n')
        print('\n'.join(sorted(additional_indexes)))
        print()
        status = 1

    if not (missing_indexes or additional_indexes):
        print('Everything is Ok')
        status = 0

    return status