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:
53 * Eliminate single point of failure (redundant components)
55 - use an uniteruptable power supply (UPS)
56 - use redundant power supplies on the main boards
58 - use redundant network hardware
59 - use distributed, redundant storage
63 - automatic error detection
66 Virtualization environments like {pve} makes it much easier to reach
67 high availability because they remove the "hardware" dependency. It is
68 also easy to setup and use redundant storage and network devices. So
69 if one host fail, you can simply start those services on another host
70 within your cluster. Even better, 'ha-manager' is able to
71 automatically detect errors and do automatic failover.
73 'ha-manager' handles management of user-defined cluster services. This
74 includes handling of user requests which may start, stop, relocate,
76 The cluster resource manager daemon also handles restarting and relocating
77 services to another node in the event of failures.
79 A service (also called resource) is uniquely identified by a service ID
80 (SID) which consists of the service type and an type specific id, e.g.:
81 'vm:100'. That example would be a service of type vm (Virtual machine)
87 * at least three nodes
93 * hardware watchdog - if not available we fall back to the
99 This section provides an in detail description of the {PVE} HA-manager
100 internals. It describes how the CRM and the LRM work together.
102 To provide High Availability two daemons run on each node:
106 The local resource manager (LRM), it controls the services running on
108 It reads the requested states for its services from the current manager
109 status file and executes the respective commands.
113 The cluster resource manager (CRM), it controls the cluster wide
114 actions of the services, processes the LRM result includes the state
115 machine which controls the state of each service.
117 .Locks in the LRM & CRM
119 Locks are provided by our distributed configuration file system (pmxcfs).
120 They are used to guarantee that each LRM is active and working as a
121 LRM only executes actions when he has its lock we can mark a failed node
122 as fenced if we get its lock. This lets us then recover the failed HA services
123 securely without the failed (but maybe still running) LRM interfering.
124 This all gets supervised by the CRM which holds currently the manager master
127 Local Resource Manager
128 ~~~~~~~~~~~~~~~~~~~~~~
130 The local resource manager ('pve-ha-lrm') is started as a daemon on
131 boot and waits until the HA cluster is quorate and thus cluster wide
134 It can be in three states:
136 * *wait for agent lock*: the LRM waits for our exclusive lock. This is
137 also used as idle sate if no service is configured
138 * *active*: the LRM holds its exclusive lock and has services configured
139 * *lost agent lock*: the LRM lost its lock, this means a failure happened
142 After the LRM gets in the active state it reads the manager status
143 file in '/etc/pve/ha/manager_status' and determines the commands it
144 has to execute for the service it owns.
145 For each command a worker gets started, this workers are running in
146 parallel and are limited to maximal 4 by default. This default setting
147 may be changed through the datacenter configuration key "max_worker".
149 .Maximal Concurrent Worker Adjustment Tips
151 The default value of 4 maximal concurrent Workers may be unsuited for
152 a specific setup. For example may 4 live migrations happen at the same
153 time, which can lead to network congestions with slower networks and/or
154 big (memory wise) services. Ensure that also in the worst case no congestion
155 happens and lower the "max_worker" value if needed. In the contrary, if you
156 have a particularly powerful high end setup you may also want to increase it.
158 Each command requested by the CRM is uniquely identifiable by an UID, when
159 the worker finished its result will be processed and written in the LRM
160 status file '/etc/pve/nodes/<nodename>/lrm_status'. There the CRM may collect
161 it and let its state machine - respective the commands output - act on it.
163 The actions on each service between CRM and LRM are normally always synced.
164 This means that the CRM requests a state uniquely marked by an UID, the LRM
165 then executes this action *one time* and writes back the result, also
166 identifiable by the same UID. This is needed so that the LRM does not
167 executes an outdated command.
168 With the exception of the 'stop' and the 'error' command,
169 those two do not depend on the result produce and are executed
170 always in the case of the stopped state and once in the case of
175 The HA Stack logs every action it makes. This helps to understand what
176 and also why something happens in the cluster. Here its important to see
177 what both daemons, the LRM and the CRM, did. You may use
178 `journalctl -u pve-ha-lrm` on the node(s) where the service is and
179 the same command for the pve-ha-crm on the node which is the current master.
181 Cluster Resource Manager
182 ~~~~~~~~~~~~~~~~~~~~~~~~
184 The cluster resource manager ('pve-ha-crm') starts on each node and
185 waits there for the manager lock, which can only be held by one node
186 at a time. The node which successfully acquires the manager lock gets
187 promoted to the CRM master.
189 It can be in three states: TODO
191 * *wait for agent lock*: the LRM waits for our exclusive lock. This is
192 also used as idle sate if no service is configured
193 * *active*: the LRM holds its exclusive lock and has services configured
194 * *lost agent lock*: the LRM lost its lock, this means a failure happened
197 It main task is to manage the services which are configured to be highly
198 available and try to get always bring them in the wanted state, e.g.: a
199 enabled service will be started if its not running, if it crashes it will
200 be started again. Thus it dictates the LRM the wanted actions.
202 When an node leaves the cluster quorum, its state changes to unknown.
203 If the current CRM then can secure the failed nodes lock, the services
204 will be 'stolen' and restarted on another node.
206 When a cluster member determines that it is no longer in the cluster
207 quorum, the LRM waits for a new quorum to form. As long as there is no
208 quorum the node cannot reset the watchdog. This will trigger a reboot
214 The HA stack is well integrated int the Proxmox VE API2. So, for
215 example, HA can be configured via 'ha-manager' or the PVE web
216 interface, which both provide an easy to use tool.
218 The resource configuration file can be located at
219 '/etc/pve/ha/resources.cfg' and the group configuration file at
220 '/etc/pve/ha/groups.cfg'. Use the provided tools to make changes,
221 there shouldn't be any need to edit them manually.
226 If a node needs maintenance you should migrate and or relocate all
227 services which are required to run always on another node first.
228 After that you can stop the LRM and CRM services. But note that the
229 watchdog triggers if you stop it with active services.
237 Fencing secures that on a node failure the dangerous node gets will be rendered
238 unable to do any damage and that no resource runs twice when it gets recovered
239 from the failed node.
241 Configure Hardware Watchdog
242 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
243 By default all watchdog modules are blocked for security reasons as they are
244 like a loaded gun if not correctly initialized.
245 If you have a hardware watchdog available remove its module from the blacklist
246 and restart 'the watchdog-mux' service.
249 Resource/Service Agents
250 -------------------------
252 A resource or also called service can be managed by the
253 ha-manager. Currently we support virtual machines and container.
258 A group is a collection of cluster nodes which a service may be bound to.
265 list of group node members
269 resources bound to this group may only run on nodes defined by the
270 group. If no group node member is available the resource will be
271 placed in the stopped state.
275 the resource won't automatically fail back when a more preferred node
276 (re)joins the cluster.
282 There are two service recover policy settings which can be configured
283 specific for each resource.
287 maximal number of tries to restart an failed service on the actual
288 node. The default is set to one.
292 maximal number of tries to relocate the service to a different node.
293 A relocate only happens after the max_restart value is exceeded on the
294 actual node. The default is set to one.
296 Note that the relocate count state will only reset to zero when the
297 service had at least one successful start. That means if a service is
298 re-enabled without fixing the error only the restart policy gets
304 If after all tries the service state could not be recovered it gets
305 placed in an error state. In this state the service won't get touched
306 by the HA stack anymore. To recover from this state you should follow
309 * bring the resource back into an safe and consistent state (e.g:
312 * disable the ha resource to place it in an stopped state
314 * fix the error which led to this failures
316 * *after* you fixed all errors you may enable the service again
322 This are how the basic user-initiated service operations (via
327 the service will be started by the LRM if not already running.
331 the service will be stopped by the LRM if running.
335 the service will be relocated (live) to another node.
339 the service will be removed from the HA managed resource list. Its
340 current state will not be touched.
344 start and stop commands can be issued to the resource specific tools
345 (like 'qm' or 'pct'), they will forward the request to the
346 'ha-manager' which then will execute the action and set the resulting
347 service state (enabled, disabled).
355 Service is stopped (confirmed by LRM)
359 Service should be stopped. Waiting for confirmation from LRM.
363 Service is active an LRM should start it ASAP if not already running.
367 Wait for node fencing (service node is not inside quorate cluster
372 Do not touch the service state. We use this state while we reboot a
373 node, or when we restart the LRM daemon.
377 Migrate service (live) to other node.
381 Service disabled because of LRM errors. Needs manual intervention.
385 include::pve-copyright.adoc[]