5 include::attributes.txt[]
10 ha-manager - Proxmox VE HA Manager
15 include::ha-manager.1-synopsis.adoc[]
24 include::attributes.txt[]
28 Our modern society depends heavily on information provided by
29 computers over the network. Mobile devices amplified that dependency,
30 because people can access the network any time from anywhere. If you
31 provide such services, it is very important that they are available
34 We can mathematically define the availability as the ratio of (A) the
35 total time a service is capable of being used during a given interval
36 to (B) the length of the interval. It is normally expressed as a
37 percentage of uptime in a given year.
39 .Availability - Downtime per Year
40 [width="60%",cols="<d,d",options="header"]
41 |===========================================================
42 |Availability % |Downtime per year
47 |99.9999 |31.5 seconds
48 |99.99999 |3.15 seconds
49 |===========================================================
51 There are several ways to increase availability. The most elegant
52 solution is to rewrite your software, so that you can run it on
53 several host at the same time. The software itself need to have a way
54 to detect erors and do failover. This is relatively easy if you just
55 want to serve read-only web pages. But in general this is complex, and
56 sometimes impossible because you cannot modify the software
57 yourself. The following solutions works without modifying the
60 * Use reliable "server" components
62 NOTE: Computer components with same functionality can have varying
63 reliability numbers, depending on the component quality. Most verdors
64 sell components with higher reliability as "server" components -
65 usually at higher price.
67 * Eliminate single point of failure (redundant components)
69 - use an uniteruptable power supply (UPS)
70 - use redundant power supplies on the main boards
72 - use redundant network hardware
73 - use RAID for local storage
74 - use distributed, redundant storage for VM data
78 - rapidly accessible adminstrators (24/7)
79 - availability of spare parts (other nodes is a {pve} cluster)
80 - automatic error detection ('ha-manager')
81 - automatic failover ('ha-manager')
83 Virtualization environments like {pve} makes it much easier to reach
84 high availability because they remove the "hardware" dependency. They
85 also support to setup and use redundant storage and network
86 devices. So if one host fail, you can simply start those services on
87 another host within your cluster.
89 Even better, {pve} provides a software stack called 'ha-manager',
90 which can do that automatically for you. It is able to automatically
91 detect errors and do automatic failover.
93 {pve} 'ha-manager' works like an "automated" administrator. First, you
94 configure what resources (VMs, containers, ...) it should
95 manage. 'ha-manager' then observes correct functionality, and handles
96 service failover to another node in case of errors. 'ha-manager' can
97 also handle normal user requests which may start, stop, relocate and
100 But high availability comes at a price. High quality components are
101 more expensive, and making them redundant duplicates the costs at
102 least. Additional spare parts increase costs further. So you should
103 carefully calculate the benefits, and compare with those additional
106 TIP: Increasing availability from 99% to 99.9% is relatively
107 simply. But increasing availability from 99.9999% to 99.99999% is very
108 hard and costly. 'ha-manager' has typical error detection and failover
109 times of about 2 minutes, so you can get no more than 99.999%
116 A resource (sometimes also called service) is uniquely identified by a
117 service ID (SID) which consists of the service type and an type
118 specific id, e.g.: 'vm:100'. That example would be a service of type
119 vm (Virtual machine) with the VMID 100.
125 * at least three nodes
129 * hardware redundancy
131 * hardware watchdog - if not available we fall back to the
132 linux kernel soft dog
137 This section provides an in detail description of the {PVE} HA-manager
138 internals. It describes how the CRM and the LRM work together.
140 To provide High Availability two daemons run on each node:
144 The local resource manager (LRM), it controls the services running on
146 It reads the requested states for its services from the current manager
147 status file and executes the respective commands.
151 The cluster resource manager (CRM), it controls the cluster wide
152 actions of the services, processes the LRM result includes the state
153 machine which controls the state of each service.
155 .Locks in the LRM & CRM
157 Locks are provided by our distributed configuration file system (pmxcfs).
158 They are used to guarantee that each LRM is active and working as a
159 LRM only executes actions when he has its lock we can mark a failed node
160 as fenced if we get its lock. This lets us then recover the failed HA services
161 securely without the failed (but maybe still running) LRM interfering.
162 This all gets supervised by the CRM which holds currently the manager master
165 Local Resource Manager
166 ~~~~~~~~~~~~~~~~~~~~~~
168 The local resource manager ('pve-ha-lrm') is started as a daemon on
169 boot and waits until the HA cluster is quorate and thus cluster wide
172 It can be in three states:
174 * *wait for agent lock*: the LRM waits for our exclusive lock. This is
175 also used as idle sate if no service is configured
176 * *active*: the LRM holds its exclusive lock and has services configured
177 * *lost agent lock*: the LRM lost its lock, this means a failure happened
180 After the LRM gets in the active state it reads the manager status
181 file in '/etc/pve/ha/manager_status' and determines the commands it
182 has to execute for the service it owns.
183 For each command a worker gets started, this workers are running in
184 parallel and are limited to maximal 4 by default. This default setting
185 may be changed through the datacenter configuration key "max_worker".
187 .Maximal Concurrent Worker Adjustment Tips
189 The default value of 4 maximal concurrent Workers may be unsuited for
190 a specific setup. For example may 4 live migrations happen at the same
191 time, which can lead to network congestions with slower networks and/or
192 big (memory wise) services. Ensure that also in the worst case no congestion
193 happens and lower the "max_worker" value if needed. In the contrary, if you
194 have a particularly powerful high end setup you may also want to increase it.
196 Each command requested by the CRM is uniquely identifiable by an UID, when
197 the worker finished its result will be processed and written in the LRM
198 status file '/etc/pve/nodes/<nodename>/lrm_status'. There the CRM may collect
199 it and let its state machine - respective the commands output - act on it.
201 The actions on each service between CRM and LRM are normally always synced.
202 This means that the CRM requests a state uniquely marked by an UID, the LRM
203 then executes this action *one time* and writes back the result, also
204 identifiable by the same UID. This is needed so that the LRM does not
205 executes an outdated command.
206 With the exception of the 'stop' and the 'error' command,
207 those two do not depend on the result produce and are executed
208 always in the case of the stopped state and once in the case of
213 The HA Stack logs every action it makes. This helps to understand what
214 and also why something happens in the cluster. Here its important to see
215 what both daemons, the LRM and the CRM, did. You may use
216 `journalctl -u pve-ha-lrm` on the node(s) where the service is and
217 the same command for the pve-ha-crm on the node which is the current master.
219 Cluster Resource Manager
220 ~~~~~~~~~~~~~~~~~~~~~~~~
222 The cluster resource manager ('pve-ha-crm') starts on each node and
223 waits there for the manager lock, which can only be held by one node
224 at a time. The node which successfully acquires the manager lock gets
225 promoted to the CRM master.
227 It can be in three states: TODO
229 * *wait for agent lock*: the LRM waits for our exclusive lock. This is
230 also used as idle sate if no service is configured
231 * *active*: the LRM holds its exclusive lock and has services configured
232 * *lost agent lock*: the LRM lost its lock, this means a failure happened
235 It main task is to manage the services which are configured to be highly
236 available and try to get always bring them in the wanted state, e.g.: a
237 enabled service will be started if its not running, if it crashes it will
238 be started again. Thus it dictates the LRM the wanted actions.
240 When an node leaves the cluster quorum, its state changes to unknown.
241 If the current CRM then can secure the failed nodes lock, the services
242 will be 'stolen' and restarted on another node.
244 When a cluster member determines that it is no longer in the cluster
245 quorum, the LRM waits for a new quorum to form. As long as there is no
246 quorum the node cannot reset the watchdog. This will trigger a reboot
252 The HA stack is well integrated int the Proxmox VE API2. So, for
253 example, HA can be configured via 'ha-manager' or the PVE web
254 interface, which both provide an easy to use tool.
256 The resource configuration file can be located at
257 '/etc/pve/ha/resources.cfg' and the group configuration file at
258 '/etc/pve/ha/groups.cfg'. Use the provided tools to make changes,
259 there shouldn't be any need to edit them manually.
264 If a node needs maintenance you should migrate and or relocate all
265 services which are required to run always on another node first.
266 After that you can stop the LRM and CRM services. But note that the
267 watchdog triggers if you stop it with active services.
275 Fencing secures that on a node failure the dangerous node gets will be rendered
276 unable to do any damage and that no resource runs twice when it gets recovered
277 from the failed node.
279 Configure Hardware Watchdog
280 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
281 By default all watchdog modules are blocked for security reasons as they are
282 like a loaded gun if not correctly initialized.
283 If you have a hardware watchdog available remove its module from the blacklist
284 and restart 'the watchdog-mux' service.
287 Resource/Service Agents
288 -------------------------
290 A resource or also called service can be managed by the
291 ha-manager. Currently we support virtual machines and container.
296 A group is a collection of cluster nodes which a service may be bound to.
303 list of group node members
307 resources bound to this group may only run on nodes defined by the
308 group. If no group node member is available the resource will be
309 placed in the stopped state.
313 the resource won't automatically fail back when a more preferred node
314 (re)joins the cluster.
320 There are two service recover policy settings which can be configured
321 specific for each resource.
325 maximal number of tries to restart an failed service on the actual
326 node. The default is set to one.
330 maximal number of tries to relocate the service to a different node.
331 A relocate only happens after the max_restart value is exceeded on the
332 actual node. The default is set to one.
334 Note that the relocate count state will only reset to zero when the
335 service had at least one successful start. That means if a service is
336 re-enabled without fixing the error only the restart policy gets
342 If after all tries the service state could not be recovered it gets
343 placed in an error state. In this state the service won't get touched
344 by the HA stack anymore. To recover from this state you should follow
347 * bring the resource back into an safe and consistent state (e.g:
350 * disable the ha resource to place it in an stopped state
352 * fix the error which led to this failures
354 * *after* you fixed all errors you may enable the service again
360 This are how the basic user-initiated service operations (via
365 the service will be started by the LRM if not already running.
369 the service will be stopped by the LRM if running.
373 the service will be relocated (live) to another node.
377 the service will be removed from the HA managed resource list. Its
378 current state will not be touched.
382 start and stop commands can be issued to the resource specific tools
383 (like 'qm' or 'pct'), they will forward the request to the
384 'ha-manager' which then will execute the action and set the resulting
385 service state (enabled, disabled).
393 Service is stopped (confirmed by LRM)
397 Service should be stopped. Waiting for confirmation from LRM.
401 Service is active an LRM should start it ASAP if not already running.
405 Wait for node fencing (service node is not inside quorate cluster
410 Do not touch the service state. We use this state while we reboot a
411 node, or when we restart the LRM daemon.
415 Migrate service (live) to other node.
419 Service disabled because of LRM errors. Needs manual intervention.
423 include::pve-copyright.adoc[]