Tune DS replication settings

Tune 389-DS replication settings to improve performance and avoid
timeouts. During installation of a replica, the value of
nsDS5ReplicaBindDnGroupCheckInterval is reduced to 2 seconds. At the end
of the installation, the value is increased sensible production
settings. This avoids long delays during replication.

See: https://pagure.io/freeipa/issue/7617
Signed-off-by: Christian Heimes <cheimes@redhat.com>
Reviewed-By: Tibor Dudlak <tdudlak@redhat.com>

ipaserver/install/cainstance.py

@@ -419,6 +419,9 @@ class CAInstance(DogtagInstance):
                           self.teardown_admin)
             self.step("starting certificate server instance",
                       self.start_instance)
+            if promote:
+                self.step("Finalize replication settings",
+                          self.finalize_replica_config)
         # Step 1 of external is getting a CSR so we don't need to do these
         # steps until we get a cert back from the external CA.
         if self.external != 1:
@@ -1253,13 +1256,16 @@ class CAInstance(DogtagInstance):
         api.Backend.ldap2.add_entry(entry)
 
     def __setup_replication(self):
-
         repl = replication.CAReplicationManager(self.realm, self.fqdn)
         repl.setup_cs_replication(self.master_host)
 
         # Activate Topology for o=ipaca segments
         self.__update_topology()
 
+    def finalize_replica_config(self):
+        repl = replication.CAReplicationManager(self.realm, self.fqdn)
+        repl.finalize_replica_config(self.master_host)
+
     def __enable_instance(self):
         basedn = ipautil.realm_to_suffix(self.realm)
         if not self.clone:

ipaserver/install/dsinstance.py

@@ -418,6 +418,20 @@ class DsInstance(service.Service):
 
         self.start_creation(runtime=30)
 
+    def _get_replication_manager(self):
+        # Always connect to self over ldapi
+        ldap_uri = ipaldap.get_ldap_uri(protocol='ldapi', realm=self.realm)
+        conn = ipaldap.LDAPClient(ldap_uri)
+        conn.external_bind()
+        repl = replication.ReplicationManager(
+            self.realm, self.fqdn, self.dm_password, conn=conn
+        )
+        if self.dm_password is not None and not self.promote:
+            bind_dn = DN(('cn', 'Directory Manager'))
+            bind_pw = self.dm_password
+        else:
+            bind_dn = bind_pw = None
+        return repl, bind_dn, bind_pw
 
     def __setup_replica(self):
         """
@@ -434,26 +448,24 @@ class DsInstance(service.Service):
             self.realm,
             self.dm_password)
 
-        # Always connect to self over ldapi
+        repl, bind_dn, bind_pw = self._get_replication_manager()
-        ldap_uri = ipaldap.get_ldap_uri(protocol='ldapi', realm=self.realm)
+        repl.setup_promote_replication(
-        conn = ipaldap.LDAPClient(ldap_uri)
+            self.master_fqdn,
-        conn.external_bind()
+            r_binddn=bind_dn,
-        repl = replication.ReplicationManager(self.realm,
+            r_bindpw=bind_pw,
-                                              self.fqdn,
+            cacert=self.ca_file
-                                              self.dm_password, conn=conn)
+        )
-
-        if self.dm_password is not None and not self.promote:
-            bind_dn = DN(('cn', 'Directory Manager'))
-            bind_pw = self.dm_password
-        else:
-            bind_dn = bind_pw = None
-
-        repl.setup_promote_replication(self.master_fqdn,
-                                       r_binddn=bind_dn,
-                                       r_bindpw=bind_pw,
-                                       cacert=self.ca_file)
         self.run_init_memberof = repl.needs_memberof_fixup()
 
+    def finalize_replica_config(self):
+        repl, bind_dn, bind_pw = self._get_replication_manager()
+        repl.finalize_replica_config(
+            self.master_fqdn,
+            r_binddn=bind_dn,
+            r_bindpw=bind_pw,
+            cacert=self.ca_file
+        )
+
     def __configure_sasl_mappings(self):
         # we need to remove any existing SASL mappings in the directory as otherwise they
         # they may conflict.

ipaserver/install/replication.py

@@ -75,6 +75,20 @@ STRIP_ATTRS = ('modifiersName',
                'internalModifiersName',
                'internalModifyTimestamp')
 
+# settings for cn=replica,cn=$DB,cn=mapping tree,cn=config
+# during replica installation
+REPLICA_CREATION_SETTINGS = {
+    "nsds5ReplicaReleaseTimeout": ["20"],
+    "nsds5ReplicaBackoffMax": ["3"],
+    "nsDS5ReplicaBindDnGroupCheckInterval": ["2"]
+}
+# after replica installation
+REPLICA_FINAL_SETTINGS = {
+    "nsds5ReplicaReleaseTimeout": ["60"],
+    "nsds5ReplicaBackoffMax": ["300"],  # default
+    "nsDS5ReplicaBindDnGroupCheckInterval": ["60"]
+}
+
 
 def replica_conn_check(master_host, host_name, realm, check_ca,
                        dogtag_master_ds_port, admin_password=None,
@@ -201,9 +215,13 @@ def wait_for_entry(connection, dn, timeout, attr=None, attrvalue='*',
 
 
 class ReplicationManager(object):
-    """Manage replication agreements between DS servers, and sync
+    """Manage replication agreements
-    agreements with Windows servers"""
+
-    def __init__(self, realm, hostname, dirman_passwd, port=PORT, starttls=False, conn=None):
+    between DS servers, and sync  agreements with Windows servers
+    """
+
+    def __init__(self, realm, hostname, dirman_passwd=None, port=PORT,
+                 starttls=False, conn=None):
         self.hostname = hostname
         self.port = port
         self.dirman_passwd = dirman_passwd
@@ -481,22 +499,16 @@ class ReplicationManager(object):
         except errors.NotFound:
             pass
         else:
-            managers = {DN(m) for m in entry.get('nsDS5ReplicaBindDN', [])}
+            binddns = entry.setdefault('nsDS5ReplicaBindDN', [])
-
+            if replica_binddn not in {DN(m) for m in binddns}:
-            mods = []
-            if replica_binddn not in managers:
                 # Add the new replication manager
-                mods.append(
+                binddns.append(replica_binddn)
-                    (ldap.MOD_ADD, 'nsDS5ReplicaBindDN', replica_binddn)
+            for key, value in REPLICA_CREATION_SETTINGS.items():
-                )
+                entry[key] = value
-            if 'nsds5replicareleasetimeout' not in entry:
+            try:
-                # See https://pagure.io/freeipa/issue/7488
+                conn.update_entry(entry)
-                mods.append(
+            except errors.EmptyModlist:
-                    (ldap.MOD_ADD, 'nsds5replicareleasetimeout', ['60'])
+                pass
-                )
-
-            if mods:
-                conn.modify_s(dn, mods)
 
             self.set_replica_binddngroup(conn, entry)
 
@@ -515,9 +527,8 @@ class ReplicationManager(object):
             nsds5flags=["1"],
             nsds5replicabinddn=[replica_binddn],
             nsds5replicabinddngroup=[self.repl_man_group_dn],
-            nsds5replicabinddngroupcheckinterval=["60"],
-            nsds5replicareleasetimeout=["60"],
             nsds5replicalegacyconsumer=["off"],
+            **REPLICA_CREATION_SETTINGS
         )
         conn.add_entry(entry)
 
@@ -543,6 +554,47 @@ class ReplicationManager(object):
         except errors.DuplicateEntry:
             return
 
+    def _finalize_replica_settings(self, conn):
+        """Change replica settings to final values
+
+        During replica installation, some settings are configured for faster
+        replication.
+        """
+        dn = self.replica_dn()
+        entry = conn.get_entry(dn)
+        for key, value in REPLICA_FINAL_SETTINGS.items():
+            entry[key] = value
+        try:
+            conn.update_entry(entry)
+        except errors.EmptyModlist:
+            pass
+
+    def finalize_replica_config(self, r_hostname, r_binddn=None,
+                                r_bindpw=None, cacert=paths.IPA_CA_CRT):
+        """Apply final cn=replica settings
+
+        replica_config() sets several attribute to fast cache invalidation
+        and fast reconnects to optimize replicat installation. For
+        production, longer timeouts and less aggressive cache invalidation
+        is sufficient. finalize_replica_config() sets the values on new
+        replica and the master.
+
+        When installing multiple replicas in parallel, one replica may
+        finalize the values while another is still installing.
+
+        See https://pagure.io/freeipa/issue/7617
+        """
+        self._finalize_replica_settings(self.conn)
+
+        ldap_uri = ipaldap.get_ldap_uri(r_hostname)
+        r_conn = ipaldap.LDAPClient(ldap_uri, cacert=cacert)
+        if r_bindpw:
+            r_conn.simple_bind(r_binddn, r_bindpw)
+        else:
+            r_conn.gssapi_bind()
+        self._finalize_replica_settings(r_conn)
+        r_conn.close()
+
     def setup_chaining_backend(self, conn):
         chaindn = DN(('cn', 'chaining database'), ('cn', 'plugins'), ('cn', 'config'))
         benamebase = "chaindb"

ipaserver/install/server/replicainstall.py

@@ -1505,6 +1505,8 @@ def install(installer):
     # Apply any LDAP updates. Needs to be done after the replica is synced-up
     service.print_msg("Applying LDAP updates")
     ds.apply_updates()
+    service.print_msg("Finalize replication settings")
+    ds.finalize_replica_config()
 
     if kra_enabled:
         kra.install(api, config, options, custodia=custodia)

ipaserver/install/server/upgrade.py

@@ -48,6 +48,7 @@ from ipaserver.install import dnskeysyncinstance
 from ipaserver.install import dogtaginstance
 from ipaserver.install import krbinstance
 from ipaserver.install import adtrustinstance
+from ipaserver.install import replication
 from ipaserver.install.upgradeinstance import IPAUpgrade
 from ipaserver.install.ldapupdate import BadSyntax
 
@@ -1680,11 +1681,14 @@ def update_replica_config(db_suffix):
     except ipalib.errors.NotFound:
         return  # entry does not exist until a replica is installed
 
-    if 'nsds5replicareleasetimeout' not in entry:
+    for key, value in replication.REPLICA_FINAL_SETTINGS.items():
-        # See https://pagure.io/freeipa/issue/7488
+        entry[key] = value
-        logger.info("Adding nsds5replicaReleaseTimeout=60 to %s", dn)
+    try:
-        entry['nsds5replicareleasetimeout'] = '60'
         api.Backend.ldap2.update_entry(entry)
+    except ipalib.errors.EmptyModlist:
+        pass
+    else:
+        logger.info("Updated entry %s", dn)
 
 
 def migrate_to_authselect():

ipatests/test_integration/test_external_ca.py

@@ -148,7 +148,11 @@ class TestExternalCA(IntegrationTest):
             '-LLL',
             '-o',
             'ldif-wrap=no'])
-        assert 'nsds5ReplicaReleaseTimeout: 60' in result.stdout_text
+        # case insensitive match
+        text = result.stdout_text.lower()
+        # see ipaserver.install.replication.REPLICA_FINAL_SETTINGS
+        assert 'nsds5ReplicaReleaseTimeout: 60'.lower() in text
+        assert 'nsDS5ReplicaBindDnGroupCheckInterval: 60'.lower() in text
 
     def test_client_installation_with_otp(self):
         # Test for issue 7526: client installation fails with one-time