misc/scripts/repair_splitbrain_ldapuser_source.py
author Sylvain Thénault <sylvain.thenault@logilab.fr>
Mon, 21 Jan 2013 18:01:25 +0100
changeset 8669 62213a34726e
parent 8568 56d9eb013feb
child 10589 7c23b7de2b8d
permissions -rw-r--r--
[db-api/configuration] simplify db-api and configuration so that all the connection information is in the repository url, closes #2521848 eg no more specific option of pyro ns host, group, etc. This also fixes broken ZMQ sources. Changes: * dropped pyro-ns-host, pyro-instance-id, pyro-ns-group from client side config, in favor of repository-uri. No migration done, supposing there is **no web-only config** in the wild. Also stop discovering the connection method through the repo_method class attribute of the configuration, varying according to the configuration class. This is a first step on the way to a simpler configuration handling. Notice those pyro options are still available for repository only / all-in-one configurations as they are needed to configure the pyro server. * stop telling connection method using ConnectionProperties, this is so boring. Also, drop _cnxtype from Connection and cnxtype from Session. The former is replaced by a is_repo_in_memory property and the later is totaly useless. * deprecate in_memory_cnx which becomes useless, use _repo_connect instead

"""
CAUTION: READ THIS CAREFULLY

Sometimes it happens that ldap (specifically ldapuser type) source
yield "ghost" users. The reasons may vary (server upgrade while some
instances are still running & syncing with the ldap source, unmanaged
updates to the upstream ldap, etc.).

This script was written and refined enough times that we are confident
in that it does something reasonnable (at least it did for the
target application).

However you should really REALLY understand what it does before
deciding to apply it for you. And then ADAPT it tou your needs.

"""

import base64
from collections import defaultdict

from cubicweb.server.session import hooks_control

try:
    source_name, = __args__
    source = repo.sources_by_uri[source_name]
except ValueError:
    print('you should specify the source name as script argument (i.e. after --'
          ' on the command line)')
    sys.exit(1)
except KeyError:
    print '%s is not an active source' % source_name
    sys.exit(1)

# check source is reachable before doing anything
if not source.get_connection().cnx:
    print '%s is not reachable. Fix this before running this script' % source_name
    sys.exit(1)

def find_dupes():
    # XXX this retrieves entities from a source name "ldap"
    #     you will want to adjust
    rset = sql("SELECT eid, extid FROM entities WHERE source='%s'" % source_name)
    extid2eids = defaultdict(list)
    for eid, extid in rset:
        extid2eids[extid].append(eid)
    return dict((base64.b64decode(extid).lower(), eids)
                for extid, eids in extid2eids.items()
                if len(eids) > 1)

def merge_dupes(dupes, docommit=False):
    gone_eids = []
    CWUser = schema['CWUser']
    for extid, eids in dupes.items():
        newest = eids.pop() # we merge everything on the newest
        print 'merging ghosts of', extid, 'into', newest
        # now we merge pairwise into the newest
        for old in eids:
            subst = {'old': old, 'new': newest}
            print '  merging', old
            gone_eids.append(old)
            for rschema in CWUser.subject_relations():
                if rschema.final or rschema == 'identity':
                    continue
                if CWUser.rdef(rschema, 'subject').composite == 'subject':
                    # old 'composite' property is wiped ...
                    # think about email addresses, excel preferences
                    for eschema in rschema.objects():
                        rql('DELETE %s X WHERE U %s X, U eid %%(old)s' % (eschema, rschema), subst)
                else:
                    # relink the new user to its old relations
                    rql('SET NU %s X WHERE NU eid %%(new)s, NOT NU %s X, OU %s X, OU eid %%(old)s' %
                        (rschema, rschema, rschema), subst)
                    # delete the old relations
                    rql('DELETE U %s X WHERE U eid %%(old)s' % rschema, subst)
            # same thing ...
            for rschema in CWUser.object_relations():
                if rschema.final or rschema == 'identity':
                    continue
                rql('SET X %s NU WHERE NU eid %%(new)s, NOT X %s NU, X %s OU, OU eid %%(old)s' %
                    (rschema, rschema, rschema), subst)
                rql('DELETE X %s U WHERE U eid %%(old)s' % rschema, subst)
    if not docommit:
        rollback()
        return
    commit() # XXX flushing operations is wanted rather than really committing
    print 'clean up entities table'
    sql('DELETE FROM entities WHERE eid IN (%s)' % (', '.join(str(x) for x in gone_eids)))
    commit()

def main():
    dupes = find_dupes()
    if not dupes:
        print 'No duplicate user'
        return

    print 'Found %s duplicate user instances' % len(dupes)

    while True:
        print 'Fix or dry-run? (f/d)  ... or Ctrl-C to break out'
        answer = raw_input('> ')
        if answer.lower() not in 'fd':
            continue
        print 'Please STOP THE APPLICATION INSTANCES (service or interactive), and press Return when done.'
        raw_input('<I swear all running instances and workers of the application are stopped>')
        with hooks_control(session, session.HOOKS_DENY_ALL):
            merge_dupes(dupes, docommit=answer=='f')

main()