1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <!DOCTYPE refentry PUBLIC "-//Samba-Team//DTD DocBook V4.2-Based Variant V1.0//EN" "http://www.samba.org/samba/DTD/samba-doc">
3 <refentry id="ctdbd.1">
6 <refentrytitle>ctdbd</refentrytitle>
7 <manvolnum>1</manvolnum>
12 <refname>ctdbd</refname>
13 <refpurpose>The CTDB cluster daemon</refpurpose>
18 <command>ctdbd</command>
22 <command>ctdbd</command>
23 <arg choice="opt">-? --help</arg>
24 <arg choice="opt">-d --debug=<INTEGER></arg>
25 <arg choice="req">--dbdir=<directory></arg>
26 <arg choice="req">--dbdir-persistent=<directory></arg>
27 <arg choice="opt">--event-script-dir=<directory></arg>
28 <arg choice="opt">-i --interactive</arg>
29 <arg choice="opt">--listen=<address></arg>
30 <arg choice="opt">--logfile=<filename></arg>
31 <arg choice="opt">--lvs</arg>
32 <arg choice="req">--nlist=<filename></arg>
33 <arg choice="opt">--no-lmaster</arg>
34 <arg choice="opt">--no-recmaster</arg>
35 <arg choice="opt">--nosetsched</arg>
36 <arg choice="opt">--public-addresses=<filename></arg>
37 <arg choice="opt">--public-interface=<interface></arg>
38 <arg choice="req">--reclock=<filename></arg>
39 <arg choice="opt">--single-public-ip=<address></arg>
40 <arg choice="opt">--socket=<filename></arg>
41 <arg choice="opt">--start-as-disabled</arg>
42 <arg choice="opt">--syslog</arg>
43 <arg choice="opt">--torture</arg>
44 <arg choice="opt">--transport=<STRING></arg>
45 <arg choice="opt">--usage</arg>
50 <refsect1><title>DESCRIPTION</title>
52 ctdbd is the main ctdb daemon.
55 ctdbd provides a clustered version of the TDB database with automatic rebuild/recovery of the databases upon nodefailures.
58 Combined with a cluster filesystem ctdbd provides a full HA environment for services such as clustered Samba and NFS as well as other services.
61 ctdbd provides monitoring of all nodes in the cluster and automatically reconfigures the cluster and recovers upon node failures.
64 ctdbd is the main component in clustered Samba that provides a high-awailability load-sharing CIFS server cluster.
70 <title>OPTIONS</title>
73 <varlistentry><term>-? --help</term>
76 Print some help text to the screen.
81 <varlistentry><term>-d --debug=<DEBUGLEVEL></term>
84 This option sets the debuglevel on the ctdbd daemon which controls what will be written to the logfile. The default is 0 which will only log important events and errors. A larger number will provide additional logging.
89 <varlistentry><term>--dbdir=<directory></term>
92 This is the directory on local storage where ctdbd keeps the local
93 copy of the TDB databases. This directory is local for each node and should not be stored on the shared cluster filesystem.
96 This directory would usually be /var/ctdb .
101 <varlistentry><term>--dbdir-persistent=<directory></term>
104 This is the directory on local storage where ctdbd keeps the local
105 copy of the persistent TDB databases. This directory is local for each node and should not be stored on the shared cluster filesystem.
108 This directory would usually be /etc/ctdb/persistent .
113 <varlistentry><term>--event-script-dir=<directory></term>
116 This option is used to specify the directory where the CTDB event
120 This will normally be /etc/ctdb/events.d which is part of the ctdb distribution.
125 <varlistentry><term>-i --interactive</term>
128 By default ctdbd will detach itself from the shell and run in
129 the background as a daemon. This option makes ctdbd to start in interactive mode.
134 <varlistentry><term>--listen=<address></term>
137 This specifies which ip address ctdb will bind to. By default ctdbd will bind to the first address it finds in the /etc/ctdb/nodes file and which is also present on the local system in which case you do not need to provide this option.
140 This option is only required when you want to run multiple ctdbd daemons/nodes on the same physical host in which case there would be multiple entries in /etc/ctdb/nodes what would match a local interface.
145 <varlistentry><term>--logfile=<filename></term>
148 This is the file where ctdbd will write its log. This is usually /var/log/log.ctdb .
153 <varlistentry><term>--lvs</term>
156 This option is used to activate the LVS capability on a CTDB node.
157 Please see the LVS section.
162 <varlistentry><term>--nlist=<filename></term>
165 This file contains a list of the private ip addresses of every node in the cluster. There is one line/ip address for each node. This file must be the same for all nodes in the cluster.
168 This file is usually /etc/ctdb/nodes .
173 <varlistentry><term>--no-lmaster</term>
176 This argument specifies that this node can NOT become an lmaster
177 for records in the database. This means that it will never show up
178 in the vnnmap. This feature is primarily used for making a cluster
179 span across a WAN link and use CTDB as a WAN-accelerator.
182 Please see the "remote cluster nodes" section for more information.
187 <varlistentry><term>--no-recmaster</term>
190 This argument specifies that this node can NOT become a recmaster
191 for the database. This feature is primarily used for making a cluster
192 span across a WAN link and use CTDB as a WAN-accelerator.
195 Please see the "remote cluster nodes" section for more information.
200 <varlistentry><term>--nosetsched</term>
203 This is a ctdbd debugging option. this option is only used when
207 Normally ctdb will change its scheduler to run as a real-time
208 process. This is the default mode for a normal ctdbd operation
209 to gurarantee that ctdbd always gets the cpu cycles that it needs.
212 This option is used to tell ctdbd to NOT run as a real-time process
213 and instead run ctdbd as a normal userspace process.
214 This is useful for debugging and when you want to run ctdbd under
215 valgrind or gdb. (You dont want to attach valgrind or gdb to a
221 <varlistentry><term>--public_addresses=<filename></term>
224 When used with IP takeover this specifies a file containing the public ip addresses to use on the cluster. This file contains a list of ip addresses netmasks and interfaces. When ctdb is operational it will distribute these public ip addresses evenly across the available nodes.
227 This is usually the file /etc/ctdb/public_addresses
232 <varlistentry><term>--public-interface=<interface></term>
235 This option tells ctdb which interface to attach public-addresses
236 to and also where to attach the single-public-ip when used.
239 This is only required when using public ip addresses and only when
240 you dont specify the interface explicitely on in /etc/ctdb/public_addresses or when you are using --single-public-ip.
243 If you omit this argument when using public addresses or single public ip, ctdb will not be able to send out Gratious ARPs correctly or be able to kill tcp connections correctly which will lead to application failures.
248 <varlistentry><term>--reclock=<filename></term>
251 This is the name of the lock file stored of the shared cluster filesystem that ctdbd uses to arbitrate which node has the role of recovery-master.
252 This file must be stored on shared storage.
257 <varlistentry><term>--single-public-ip=<address></term>
260 This option is used to activate the "ipmux" or the "lvs"
261 functionality of ctdb where the cluster provides a single
262 public ip address for the entire cluster. When using this option
263 you must also use the --public-interface option.
266 In this mode, all nodes of the cluster will expose a single
267 ip address from all nodes with all incoming traffic to the cluster
268 being passed through the current recmaster. This functionality
269 is similar to using a load-balancing switch.
272 All incoming packets are sent to the recmaster which will multiplex
273 the clients across all available nodes and pass the packets on to
274 a different node in the cluster to manage the connection based
275 on the clients ip address. Outgoing packets however are sent
276 directly from the node that was choosen back to the client.
277 Since all incoming packets are sent through the recmaster this will
278 have a throughput and performance impact when used. This impact
279 in performance primarily affects write-performance while
280 read-performance should be mainly unaffected.
281 Only use this feature if your environment is mostly-read
282 (i.e. most traffic is from the nodes back to the clients) or
283 if it is not important to get maximum write-performance to the
287 This feature is completely controlled from the eventscripts and
288 does not require any CTDBD involvement. However, the CTDBD daemon
289 does need to know that the "single public ip" exists so that the
290 CTDBD daemon will allow clients to set up killtcp to work on this
294 CTDBD only allows clients to use killtcp to kill off (RST) tcp
295 connections to/from an ip address that is either a normal public
296 address or to/from the ip address specified by --single-public-ip.
297 No other tcp connections are allowed to be specified with killtcp.
300 Please note that ipmux is obsolete. Use LVS, not ipmux.
301 Please see the LVS section in this manpage for instructions on
302 how to configure and use CTDB with LVS.
307 <varlistentry><term>--socket=<filename></term>
310 This specifies the name of the domain socket that ctdbd will create. This socket is used for local clients to attach to and communicate with the ctdbd daemon.
313 The default is /tmp/ctdb.socket . You only need to use this option if you plan to run multiple ctdbd daemons on the same physical host.
318 <varlistentry><term>--start-as-disabled</term>
321 This makes the ctdb daemon to be DISABLED when it starts up.
324 As it is DISABLED it will not get any of the public ip addresses
325 allocated to it, and thus this allow you to start ctdb on a node
326 without causing any ip address to failover from other nodes onto
330 When used, the administrator must keep track of when nodes start and
331 manually enable them again using the "ctdb enable" command, or else
332 the node will not host any services.
335 A node that is DISABLED will not host any services and will not be
336 reachable/used by any clients.
341 <varlistentry><term>--syslog</term>
344 Send all log messages to syslog instead of to the ctdb logfile.
349 <varlistentry><term>--torture</term>
352 This option is only used for development and testing of ctdbd. It adds artificial errors and failures to the common codepaths in ctdbd to verify that ctdbd can recover correctly for failures.
355 You do NOT want to use this option unless you are developing and testing new functionality in ctdbd.
360 <varlistentry><term>--transport=<STRING></term>
363 This option specifies which transport to use for ctdbd internode communications. The default is "tcp".
366 Currently only "tcp" is supported but "infiniband" might be
367 implemented in the future.
372 <varlistentry><term>--usage</term>
375 Print useage information to the screen.
384 <refsect1><title>Private vs Public addresses</title>
386 When used for ip takeover in a HA environment, each node in a ctdb
387 cluster has multiple ip addresses assigned to it. One private and one or more public.
390 <refsect2><title>Private address</title>
392 This is the physical ip address of the node which is configured in
393 linux and attached to a physical interface. This address uniquely
394 identifies a physical node in the cluster and is the ip addresses
395 that ctdbd will use to communicate with the ctdbd daemons on the
396 other nodes in the cluster.
399 The private addresses are configured in /etc/ctdb/nodes
400 (unless the --nlist option is used) and contain one line for each
401 node in the cluster. Each line contains the private ip address for one
402 node in the cluster. This file must be the same on all nodes in the
406 Since the private addresses are only available to the network when the
407 corresponding node is up and running you should not use these addresses
408 for clients to connect to services provided by the cluster. Instead
409 client applications should only attach to the public addresses since
410 these are guaranteed to always be available.
413 When using ip takeover, it is strongly recommended that the private
414 addresses are configured on a private network physically separated
415 from the rest of the network and that this private network is dedicated
418 Example /etc/ctdb/nodes for a four node cluster:
419 <screen format="linespecific">
426 <refsect2><title>Public address</title>
428 A public address on the other hand is not attached to an interface.
429 This address is managed by ctdbd itself and is attached/detached to
430 a physical node at runtime.
433 The ctdb cluster will assign/reassign these public addresses across the
434 available healthy nodes in the cluster. When one node fails, its public address
435 will be migrated to and taken over by a different node in the cluster
436 to ensure that all public addresses are always available to clients as
437 long as there are still nodes available capable of hosting this address.
440 These addresses are not physically attached to a specific node.
441 The 'ctdb ip' command can be used to view the current assignment of
442 public addresses and which physical node is currently serving it.
445 On each node this file contains a list of the public addresses that
446 this node is capable of hosting.
447 The list also contain the netmask and the
448 interface where this address should be attached for the case where you
449 may want to serve data out through multiple different interfaces.
451 Example /etc/ctdb/public_addresses for a node that can host 4 public addresses:
452 <screen format="linespecific">
460 In most cases this file would be the same on all nodes in a cluster but
461 there are exceptions when one may want to use different files
464 Example: 4 nodes partitioned into two subgroups :
465 <screen format="linespecific">
466 Node 0:/etc/ctdb/public_addresses
470 Node 1:/etc/ctdb/public_addresses
474 Node 2:/etc/ctdb/public_addresses
478 Node 3:/etc/ctdb/public_addresses
483 In this example nodes 0 and 1 host two public addresses on the
484 10.1.1.x network while nodes 2 and 3 host two public addresses for the
488 Ip address 10.1.1.1 can be hosted by either of nodes 0 or 1 and will be
489 available to clients as long as at least one of these two nodes are
491 If both nodes 0 and node 1 become unavailable 10.1.1.1 also becomes
492 unavailable. 10.1.1.1 can not be failed over to node 2 or node 3 since
493 these nodes do not have this ip address listed in their public
500 <refsect1><title>Node status</title>
502 The current status of each node in the cluster can be viewed by the
503 'ctdb status' command.
506 There are five possible for a node.
510 OK - This node is fully functional.
514 DISCONNECTED - This node could not be connected through the network
515 and is currently not particpating in the cluster. If there is a
516 public IP address associated with this node it should have been taken
517 over by a different node. No services are running on this node.
521 DISABLED - This node has been administratively disabled. This node is
522 still functional and participates in the CTDB cluster but its IP
523 addresses have been taken over by a different node and no services are
524 currently being hosted.
528 UNHEALTHY - A service provided by this node is malfunctioning and should
529 be investigated. The CTDB daemon itself is operational and participates
530 in the cluster. Its public IP address has been taken over by a different
531 node and no services are currently being hosted. All unhealthy nodes
532 should be investigated and require an administrative action to rectify.
536 BANNED - This node failed too many recovery attempts and has been banned
537 from participating in the cluster for a period of RecoveryBanPeriod
538 seconds. Any public IP address has been taken over by other nodes. This
539 node does not provide any services. All banned nodes should be
540 investigated and require an administrative action to rectify. This node
541 does not perticipate in the CTDB cluster but can still be communicated
542 with. I.e. ctdb commands can be sent to it.
547 <title>PUBLIC TUNABLES</title>
549 These are the public tuneables that can be used to control how ctdb behaves.
552 <refsect2><title>KeepaliveInterval</title>
553 <para>Default: 1</para>
555 How often should the nodes send keepalives to eachother.
558 <refsect2><title>KeepaliveLimit</title>
559 <para>Default: 5</para>
561 After how many keepalive intervals without any traffic should a node
562 wait until marking the peer as DISCONNECTED.
565 <refsect2><title>MonitorInterval</title>
566 <para>Default: 15</para>
568 How often should ctdb run the event scripts to check for a nodes health.
571 <refsect2><title>TickleUpdateInterval</title>
572 <para>Default: 20</para>
574 How often will ctdb record and store the "tickle" information used to
575 kickstart stalled tcp connections after a recovery.
578 <refsect2><title>EventScriptTimeout</title>
579 <para>Default: 20</para>
581 How long should ctdb let an event script run before aborting it and
582 marking the node unhealthy.
585 <refsect2><title>RecoveryBanPeriod</title>
586 <para>Default: 300</para>
588 If a node becomes banned causing repetitive recovery failures. The node will
589 eventually become banned from the cluster.
590 This controls how long the culprit node will be banned from the cluster
591 before it is allowed to try to join the cluster again.
592 Dont set to small. A node gets banned for a reason and it is usually due
593 to real problems with the node.
596 <refsect2><title>DatabaseHashSize</title>
597 <para>Default: 100000</para>
599 Size of the hash chains for the local store of the tdbs that ctdb manages.
602 <refsect2><title>RerecoveryTimeout</title>
603 <para>Default: 10</para>
605 Once a recovery has completed, no additional recoveries are permitted until this timeout has expired.
608 <refsect2><title>EnableBans</title>
609 <para>Default: 1</para>
611 When set to 0, this disables BANNING completely in the cluster and thus nodes can not get banned, even it they break. Dont set to 0.
614 <refsect2><title>DeterministicIPs</title>
615 <para>Default: 1</para>
617 When enabled, this tunable makes ctdb try to keep public ip addresses locked to specific nodes as far as possible. This makes it easier for debugging since you can know that as long as all nodes are healthy public ip X will always be hosted by node Y.
620 The cost of using deterministic ip address assignment is that it disables part of the logic where ctdb tries to reduce the number of public ip assignment changes in the cluster. This tunable may increase the number of ip failover/failbacks that are performed on the cluster by a small margin.
623 <refsect2><title>DisableWhenUnhealthy</title>
624 <para>Default: 0</para>
626 When set, As soon as a node becomes unhealthy, that node will also automatically become permanently DISABLED. Once a node is DISABLED, the only way to make it participate in the cluster again and host services is by manually enabling the node again using 'ctdb enable'.
629 This disables parts of the resilience and robustness of the cluster and should ONLY be used when the system administrator is actively monitoring the cluster, so that nodes can be enabled again.
632 <refsect2><title>NoIPFailback</title>
633 <para>Default: 0</para>
635 When set to 1, ctdb will not perform failback of ip addresses when a node becomes healthy. Ctdb WILL perform failover of public ip addresses when a node becomes UNHEALTHY, but when the node becomes HEALTHY again, ctdb will not fail the addresses back.
638 Use with caution! Normally when a node becomes available to the cluster
639 ctdb will try to reassign public ip addresses onto the new node as a way to distribute the workload evenly across the clusternode. Ctdb tries to make sure that all running nodes have approximately the same number of public addresses it hosts.
642 When you enable this tunable, CTDB will no longer attempt to rebalance the cluster by failing ip addresses back to the new nodes. An unbalanced cluster will therefore remain unbalanced until there is manual intervention from the administrator. When this parameter is set, you can manually fail public ip addresses over to the new node(s) using the 'ctdb moveip' command.
647 <refsect1><title>LVS</title>
649 LVS is a mode where CTDB presents one single ip address for the entire
650 cluster. This is an alternative to using public ip addresses and round-robin
651 DNS to loadbalance clients across the cluster.
655 This is similar to using a layer-4 loadbalancing switch but with some restrictions.
659 In this mode the cluster select a set of nodes in the cluster and loadbalance
660 all client access to the LVS address across this set of nodes. This set of nodes are all LVS capable nodes that are HEALTHY, or if no HEALTHY nodes exists
661 all LVS capable nodes regardless of health status.
662 LVS will however never loadbalance traffic to nodes that are BANNED,
663 DISABLED or DISCONNECTED. The "ctdb lvs" command is used to show
664 which nodes are currently load-balanced across.
668 One of the these nodes are elected as the LVSMASTER. This node receives all
669 traffic from clients coming in to the LVS address and multiplexes it
670 across the internal network to one of the nodes that LVS is using.
671 When responding to the client, that node will send the data back
672 directly to the client, bypassing the LVSMASTER node.
673 The command "ctdb lvsmaster" will show which node is the current
678 The path used for a client i/o is thus :
679 <screen format="linespecific">
680 (1) Client sends request packet to LVSMASTER
681 (2) LVSMASTER passes the request on to one node across the internal network.
682 (3) Selected node processes the request.
683 (4) Node responds back to client.
688 This means that all incoming traffic to the cluster will pass through
689 one physical node, which limits scalability. You can send more data to the
690 LVS address that one physical node can multiplex. This means that you
691 should not use LVS if your I/O pattern is write-intensive since you will be
692 limited in the available network bandwidth that node can handle.
693 LVS does work wery well for read-intensive workloads where only smallish
694 READ requests are going through the LVSMASTER bottleneck and the majority
695 of the traffic volume (the data in the read replies) goes straight from
696 the processing node back to the clients. For read-intensive i/o patterns you can acheive very high throughput rates in this mode.
700 Note: you can use LVS and public addresses at the same time.
703 <refsect2><title>Configuration</title>
705 To activate LVS on a CTDB node you must specify CTDB_PUBLIC_INTERFACE and
706 CTDB_LVS_PUBLIC_ADDRESS in /etc/sysconfig/ctdb.
710 You must also specify the "--lvs" command line argument to ctdbd to activete LVS as a capability of the node. This should be done automatically for you by the /etc/init.d/ctdb script.
715 <screen format="linespecific">
716 CTDB_PUBLIC_INTERFACE=eth0
717 CTDB_LVS_PUBLIC_IP=10.0.0.237
724 If you use LVS, you must still have a real/permanent address configured
725 for the public interface on each node. This address must be routable
726 and the cluster nodes must be configured so that all traffic back to client
727 hosts are routed through this interface. This is also required in order
728 to allow samba/winbind on the node to talk to the domain controller.
729 (we can not use the lvs ip address to initiate outgoing traffic)
732 I.e. make sure that you can "ping" both the domain controller and also
733 all of the clients from the node BEFORE you enable LVS. Also make sure
734 that when you ping these hosts that the traffic is routed out through the
740 <refsect1><title>REMOTE CLUSTER NODES</title>
742 It is possible to have a CTDB cluster that spans across a WAN link.
743 For example where you have a CTDB cluster in your datacentre but you also
744 want to have one additional CTDB node located at a remote branch site.
745 This is similar to how a WAN accelerator works but with the difference
746 that while a WAN-accelerator often acts as a Proxy or a MitM, in
747 the ctdb remote cluster node configuration the Samba instance at the remote site
748 IS the genuine server, not a proxy and not a MitM, and thus provides 100%
749 correct CIFS semantics to clients.
753 See the cluster as one single multihomed samba server where one of
754 the NICs (the remote node) is very far away.
758 NOTE: This does require that the cluster filesystem you use can cope
759 with WAN-link latencies. Not all cluster filesystems can handle
760 WAN-link latencies! Whether this will provide very good WAN-accelerator
761 performance or it will perform very poorly depends entirely
762 on how optimized your cluster filesystem is in handling high latency
763 for data and metadata operations.
767 To activate a node as being a remote cluster node you need to set
768 the following two parameters in /etc/sysconfig/ctdb for the remote node:
769 <screen format="linespecific">
770 CTDB_CAPABILITY_LMASTER=no
771 CTDB_CAPABILITY_RECMASTER=no
776 Verify with the command "ctdb getcapabilities" that that node no longer
777 has the recmaster or the lmaster capabilities.
784 <refsect1><title>SEE ALSO</title>
787 <ulink url="http://ctdb.samba.org/"/>
791 <refsect1><title>COPYRIGHT/LICENSE</title>
793 Copyright (C) Andrew Tridgell 2007
794 Copyright (C) Ronnie sahlberg 2007
796 This program is free software; you can redistribute it and/or modify
797 it under the terms of the GNU General Public License as published by
798 the Free Software Foundation; either version 3 of the License, or (at
799 your option) any later version.
801 This program is distributed in the hope that it will be useful, but
802 WITHOUT ANY WARRANTY; without even the implied warranty of
803 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
804 General Public License for more details.
806 You should have received a copy of the GNU General Public License
807 along with this program; if not, see http://www.gnu.org/licenses/.