High Availability: Difference between revisions

From Proxmox VE
Jump to navigation Jump to search
No edit summary
No edit summary
 
(10 intermediate revisions by the same user not shown)
Line 9: Line 9:
provide such services, it is very important that they are available
provide such services, it is very important that they are available
most of the time.
most of the time.
We can mathematically define the availability as the ratio of (A) the
We can mathematically define the availability as the ratio of (A), the
total time a service is capable of being used during a given interval
total time a service is capable of being used during a given interval
to (B) the length of the interval. It is normally expressed as a
to (B), the length of the interval. It is normally expressed as a
percentage of uptime in a given year.
percentage of uptime in a given year.
Table 1. Availability - Downtime per Year
Table 1. Availability - Downtime per Year
Line 30: Line 30:
There are several ways to increase availability. The most elegant
There are several ways to increase availability. The most elegant
solution is to rewrite your software, so that you can run it on
solution is to rewrite your software, so that you can run it on
several host at the same time. The software itself need to have a way
several hosts at the same time. The software itself needs to have a way
to detect errors and do failover. This is relatively easy if you just
to detect errors and do failover. If you only want to serve read-only
want to serve read-only web pages. But in general this is complex, and
web pages, then this is relatively simple. However, this is generally complex
sometimes impossible because you cannot modify the software
and sometimes impossible, because you cannot modify the software yourself. The
yourself. The following solutions works without modifying the
following solutions works without modifying the software:
software:
Use reliable “server” components
Use reliable “server” components
Computer components with same functionality can have varying
Computer components with the same functionality can have varying
reliability numbers, depending on the component quality. Most vendors
reliability numbers, depending on the component quality. Most vendors
sell components with higher reliability as “server” components -
sell components with higher reliability as “server” components -
Line 43: Line 42:
Eliminate single point of failure (redundant components)
Eliminate single point of failure (redundant components)
use an uninterruptible power supply (UPS)
use an uninterruptible power supply (UPS)
use redundant power supplies on the main boards
use redundant power supplies in your servers
use ECC-RAM
use ECC-RAM
use redundant network hardware
use redundant network hardware
Line 55: Line 54:
Virtualization environments like Proxmox VE make it much easier to reach
Virtualization environments like Proxmox VE make it much easier to reach
high availability because they remove the “hardware” dependency. They
high availability because they remove the “hardware” dependency. They
also support to setup and use redundant storage and network
also support the setup and use of redundant storage and network
devices. So if one host fail, you can simply start those services on
devices, so if one host fails, you can simply start those services on
another host within your cluster.
another host within your cluster.
Even better, Proxmox VE provides a software stack called ha-manager,
Better still, Proxmox VE provides a software stack called ha-manager,
which can do that automatically for you. It is able to automatically
which can do that automatically for you. It is able to automatically
detect errors and do automatic failover.
detect errors and do automatic failover.
Proxmox VE ha-manager works like an “automated” administrator. First, you
Proxmox VE ha-manager works like an “automated” administrator. First, you
configure what resources (VMs, containers, …) it should
configure what resources (VMs, containers, …) it should
manage. ha-manager then observes correct functionality, and handles
manage. Then, ha-manager observes the correct functionality, and handles
service failover to another node in case of errors. ha-manager can
service failover to another node in case of errors. ha-manager can
also handle normal user requests which may start, stop, relocate and
also handle normal user requests which may start, stop, relocate and
migrate a service.
migrate a service.
But high availability comes at a price. High quality components are
But high availability comes at a price. High quality components are
more expensive, and making them redundant duplicates the costs at
more expensive, and making them redundant doubles the costs at
least. Additional spare parts increase costs further. So you should
least. Additional spare parts increase costs further. So you should
carefully calculate the benefits, and compare with those additional
carefully calculate the benefits, and compare with those additional
costs.
costs.
Increasing availability from 99% to 99.9% is relatively
Increasing availability from 99% to 99.9% is relatively
simply. But increasing availability from 99.9999% to 99.99999% is very
simple. But increasing availability from 99.9999% to 99.99999% is very
hard and costly. ha-manager has typical error detection and failover
hard and costly. ha-manager has typical error detection and failover
times of about 2 minutes, so you can get no more than 99.999%
times of about 2 minutes, so you can get no more than 99.999%
Line 90: Line 89:
resource. A resource (also called “service”) is uniquely
resource. A resource (also called “service”) is uniquely
identified by a service ID (SID), which consists of the resource type
identified by a service ID (SID), which consists of the resource type
and an type specific ID, e.g.: vm:100. That example would be a
and a type specific ID, for example vm:100. That example would be a
resource of type vm (virtual machine) with the ID 100.
resource of type vm (virtual machine) with the ID 100.
For now we have two important resources types - virtual machines and
For now we have two important resources types - virtual machines and
containers. One basic idea here is that we can bundle related software
containers. One basic idea here is that we can bundle related software
into such VM or container, so there is no need to compose one big
into such a VM or container, so there is no need to compose one big
service from other services, like it was done with rgmanager. In
service from other services, as was done with rgmanager. In
general, a HA managed resource should not depend on other resources.
general, a HA managed resource should not depend on other resources.
Management Tasks
Management Tasks
Line 101: Line 100:
first step is to enable HA for a resource. This is done by adding the
first step is to enable HA for a resource. This is done by adding the
resource to the HA resource configuration. You can do this using the
resource to the HA resource configuration. You can do this using the
GUI, or simply use the command line tool, for example:
GUI, or simply use the command-line tool, for example:
# ha-manager add vm:100
# ha-manager add vm:100
The HA stack now tries to start the resources and keeps it
The HA stack now tries to start the resources and keep them
running. Please note that you can configure the “requested”
running. Please note that you can configure the “requested”
resources state. For example you may want that the HA stack stops the
resources state. For example you may want the HA stack to stop the
resource:
resource:
# ha-manager set vm:100 --state stopped
# ha-manager set vm:100 --state stopped
Line 113: Line 112:
automatically forward the commands to the HA stack, so
automatically forward the commands to the HA stack, so
# qm start 100
# qm start 100
simply sets the requested state to started. Same applied to qm
simply sets the requested state to started. The same applies to qm
stop, which sets the requested state to stopped.
stop, which sets the requested state to stopped.
The HA stack works fully asynchronous and needs to communicate
The HA stack works fully asynchronous and needs to communicate
with other cluster members. So it takes some seconds until you see
with other cluster members. Therefore, it takes some seconds until you see
the result of such actions.
the result of such actions.
To view the current HA resource configuration use:
To view the current HA resource configuration use:
Line 132: Line 131:
This uses online migration and tries to keep the VM running. Online
This uses online migration and tries to keep the VM running. Online
migration needs to transfer all used memory over the network, so it is
migration needs to transfer all used memory over the network, so it is
sometimes faster to stop VM, then restart it on the new node. This can be
sometimes faster to stop the VM, then restart it on the new node. This can be
done using the relocate command:
done using the relocate command:
# ha-manager relocate vm:100 node2
# ha-manager relocate vm:100 node2
Line 139: Line 138:
# ha-manager remove vm:100
# ha-manager remove vm:100
This does not start or stop the resource.
This does not start or stop the resource.
But all HA related task can be done on the GUI, so there is no need to
But all HA related tasks can be done in the GUI, so there is no need to
use the command line at all.
use the command line at all.
How It Works
How It Works
Line 150: Line 149:
the current manager status file and executes the respective commands.
the current manager status file and executes the respective commands.
pve-ha-crm
pve-ha-crm
The cluster resource manager (CRM), which makes the cluster wide
The cluster resource manager (CRM), which makes the cluster-wide
decisions. It sends commands to the LRM, processes the results,
decisions. It sends commands to the LRM, processes the results,
and moves resources to other nodes if something fails. The CRM also
and moves resources to other nodes if something fails. The CRM also
handles node fencing.
handles node fencing.
Locks in the LRM & CRMLocks are provided by our distributed configuration file system (pmxcfs).
Locks in the LRM & CRMLocks are provided by our distributed configuration file system (pmxcfs).
They are used to guarantee that each LRM is active once and working. As a
They are used to guarantee that each LRM is active once and working. As an
LRM only executes actions when it holds its lock, we can mark a failed node
LRM only executes actions when it holds its lock, we can mark a failed node
as fenced if we can acquire its lock. This lets us then recover any failed
as fenced if we can acquire its lock. This then lets us recover any failed
HA services securely without any interference from the now unknown failed node.
HA services securely without any interference from the now unknown failed node.
This all gets supervised by the CRM which holds currently the manager master
This all gets supervised by the CRM which currently holds the manager master
lock.
lock.
Service States
Service States
The CRM use a service state enumeration to record the current service
The CRM uses a service state enumeration to record the current service
state. We display this state on the GUI and you can query it using
state. This state is displayed on the GUI and can be queried using
the ha-manager command line tool:
the ha-manager command-line tool:
# ha-manager status
# ha-manager status
quorum OK
quorum OK
Line 190: Line 189:
LRM that the service is running.
LRM that the service is running.
fence
fence
Wait for node fencing (service node is not inside quorate cluster
Wait for node fencing as the service node is not inside the quorate cluster
partition). As soon as node gets fenced successfully the service will
partition (see Fencing).
be recovered to another node, if possible
As soon as node gets fenced successfully the service will be placed into the
(see Fencing).
recovery state.
recovery
Wait for recovery of the service. The HA manager tries to find a new node where
the service can run on. This search depends not only on the list of online and
quorate nodes, but also if the service is a group member and how such a group
is limited.
As soon as a new available node is found, the service will be moved there and
initially placed into stopped state. If it’s configured to run the new node
will do so.
freeze
freeze
Do not touch the service state. We use this state while we reboot a
Do not touch the service state. We use this state while we reboot a
Line 200: Line 207:
ignored
ignored
Act as if the service were not managed by HA at all.
Act as if the service were not managed by HA at all.
Useful, when full control over the service is desired temporarily,
Useful, when full control over the service is desired temporarily, without
without removing it from the HA configuration.
removing it from the HA configuration.
migrate
migrate
Migrate service (live) to other node.
Migrate service (live) to other node.
Line 213: Line 220:
Local Resource Manager
Local Resource Manager
The local resource manager (pve-ha-lrm) is started as a daemon on
The local resource manager (pve-ha-lrm) is started as a daemon on
boot and waits until the HA cluster is quorate and thus cluster wide
boot and waits until the HA cluster is quorate and thus cluster-wide
locks are working.
locks are working.
It can be in three states:
It can be in three states:
Line 226: Line 233:
file in /etc/pve/ha/manager_status and determines the commands it
file in /etc/pve/ha/manager_status and determines the commands it
has to execute for the services it owns.
has to execute for the services it owns.
For each command a worker gets started, this workers are running in
For each command a worker gets started, these workers are running in
parallel and are limited to at most 4 by default. This default setting
parallel and are limited to at most 4 by default. This default setting
may be changed through the datacenter configuration key max_worker.
may be changed through the datacenter configuration key max_worker.
Line 232: Line 239:
the CRM.
the CRM.
Maximum Concurrent Worker Adjustment TipsThe default value of at most 4 concurrent workers may be unsuited for
Maximum Concurrent Worker Adjustment TipsThe default value of at most 4 concurrent workers may be unsuited for
a specific setup. For example may 4 live migrations happen at the same
a specific setup. For example, 4 live migrations may occur at the same
time, which can lead to network congestions with slower networks and/or
time, which can lead to network congestions with slower networks and/or
big (memory wise) services. Ensure that also in the worst case no congestion
big (memory wise) services. Also, ensure that in the worst case, congestion is
happens and lower the max_worker value if needed. In the contrary, if you
at a minimum, even if this means lowering the max_worker value. On the
have a particularly powerful high end setup you may also want to increase it.
contrary, if you have a particularly powerful, high-end setup you may also want
Each command requested by the CRM is uniquely identifiable by an UID, when
to increase it.
the worker finished its result will be processed and written in the LRM
Each command requested by the CRM is uniquely identifiable by a UID. When
the worker finishes, its result will be processed and written in the LRM
status file /etc/pve/nodes/<nodename>/lrm_status. There the CRM may collect
status file /etc/pve/nodes/<nodename>/lrm_status. There the CRM may collect
it and let its state machine - respective the commands output - act on it.
it and let its state machine - respective to the commands output - act on it.
The actions on each service between CRM and LRM are normally always synced.
The actions on each service between CRM and LRM are normally always synced.
This means that the CRM requests a state uniquely marked by an UID, the LRM
This means that the CRM requests a state uniquely marked by a UID, the LRM
then executes this action one time and writes back the result, also
then executes this action one time and writes back the result, which is also
identifiable by the same UID. This is needed so that the LRM does not
identifiable by the same UID. This is needed so that the LRM does not
executes an outdated command.
execute an outdated command.
With the exception of the stop and the error command,
The only exceptions to this behaviour are the stop and error commands;
those two do not depend on the result produced and are executed
these two do not depend on the result produced and are executed
always in the case of the stopped state and once in the case of
always in the case of the stopped state and once in the case of
the error state.
the error state.
Line 268: Line 276:
lost agent lock
lost agent lock
The CRM lost its lock, this means a failure happened and quorum was lost.
The CRM lost its lock, this means a failure happened and quorum was lost.
It main task is to manage the services which are configured to be highly
Its main task is to manage the services which are configured to be highly
available and try to always enforce the requested state. For example, a
available and try to always enforce the requested state. For example, a
service with the requested state started will be started if its not
service with the requested state started will be started if its not
already running. If it crashes it will be automatically started again.
already running. If it crashes it will be automatically started again.
Thus the CRM dictates the actions which the LRM needs to execute.
Thus the CRM dictates the actions the LRM needs to execute.
When an node leaves the cluster quorum, its state changes to unknown.
When a node leaves the cluster quorum, its state changes to unknown.
If the current CRM then can secure the failed nodes lock, the services
If the current CRM can then secure the failed node’s lock, the services
will be stolen and restarted on another node.
will be stolen and restarted on another node.
When a cluster member determines that it is no longer in the cluster
When a cluster member determines that it is no longer in the cluster
quorum, the LRM waits for a new quorum to form. As long as there is no
quorum, the LRM waits for a new quorum to form. As long as there is no
quorum the node cannot reset the watchdog. This will trigger a reboot
quorum the node cannot reset the watchdog. This will trigger a reboot
after the watchdog then times out, this happens after 60 seconds.
after the watchdog times out (this happens after 60 seconds).
HA Simulator
By using the HA simulator you can test and learn all functionalities of the
Proxmox VE HA solutions.
By default, the simulator allows you to watch and test the behaviour of a
real-world 3 node cluster with 6 VMs. You can also add or remove additional VMs
or Container.
You do not have to setup or configure a real cluster, the HA simulator runs out
of the box.
Install with apt:
apt install pve-ha-simulator
You can even install the package on any Debian-based system without any
other Proxmox VE packages.  For that you will need to download the package and
copy it to the system you want to run it on for installation.  When you install
the package with apt from the local file system it will also resolve the
required dependencies for you.
To start the simulator on a remote machine you must have an X11 redirection to
your current system.
If you are on a Linux machine you can use:
ssh root@<IPofPVE> -Y
On Windows it works with mobaxterm.
After connecting to an existing Proxmox VE with the simulator installed or
installing it on your local Debian-based system manually, you can try it out as
follows.
First you need to create a working directory where the simulator saves its
current state and writes its default config:
mkdir working
Then, simply pass the created directory as a parameter to pve-ha-simulator:
pve-ha-simulator working/
You can then start, stop, migrate the simulated HA services, or even check out
what happens on a node failure.
Configuration
Configuration
The HA stack is well integrated into the Proxmox VE API. So, for example,
The HA stack is well integrated into the Proxmox VE API. So, for example,
HA can be configured via the ha-manager command line interface, or
HA can be configured via the ha-manager command-line interface, or
the Proxmox VE web interface - both interfaces provide an easy way to
the Proxmox VE web interface - both interfaces provide an easy way to
manage HA. Automation tools can use the API directly.
manage HA. Automation tools can use the API directly.
Line 291: Line 329:
The resource configuration file /etc/pve/ha/resources.cfg stores
The resource configuration file /etc/pve/ha/resources.cfg stores
the list of resources managed by ha-manager. A resource configuration
the list of resources managed by ha-manager. A resource configuration
inside that list look like this:
inside that list looks like this:
<type>: <name>
<type>: <name>
         <property> <value>
         <property> <value>
Line 326: Line 364:
ignored
ignored
The resource gets removed from the manager status and so the CRM and the LRM do
The resource gets removed from the manager status and so the CRM and the LRM do
not touch the resource anymore. All Proxmox VE API calls affecting this resource
not touch the resource anymore. All {pve} API calls affecting this resource
will be executed, directly bypassing the HA stack. CRM commands will be thrown
will be executed, directly bypassing the HA stack. CRM commands will be thrown
away while there source is in this state. The resource will not get relocated
away while there source is in this state. The resource will not get relocated
Line 339: Line 377:
ct: 102
ct: 102
     # Note: use default settings for everything
     # Note: use default settings for everything
Above config was generated using the ha-manager command line tool:
The above config was generated using the ha-manager command-line tool:
# ha-manager add vm:501 --state started --max_relocate 2
# ha-manager add vm:501 --state started --max_relocate 2
# ha-manager add ct:102
# ha-manager add ct:102
Line 366: Line 404:
behavior. For example, you may want to run a set of services on
behavior. For example, you may want to run a set of services on
node1 if possible. If node1 is not available, you want to run them
node1 if possible. If node1 is not available, you want to run them
equally split on node2 and node3. If those nodes also fail the
equally split on node2 and node3. If those nodes also fail, the
services should run on node4. To achieve this you could set the node
services should run on node4. To achieve this you could set the node
list to:
list to:
Line 375: Line 413:
restricted group with said nodes:
restricted group with said nodes:
# ha-manager groupadd mygroup2 -nodes "node1,node2" -restricted
# ha-manager groupadd mygroup2 -nodes "node1,node2" -restricted
Above commands created the following group configuration fils:
The above commands created the following group configuration file:
Configuration Example (/etc/pve/ha/groups.cfg)
Configuration Example (/etc/pve/ha/groups.cfg)
group: prefer_node1
group: prefer_node1
Line 386: Line 424:
The nofailback options is mostly useful to avoid unwanted resource
The nofailback options is mostly useful to avoid unwanted resource
movements during administration tasks. For example, if you need to
movements during administration tasks. For example, if you need to
migrate a service to a node which hasn’t the highest priority in the
migrate a service to a node which doesn’t have the highest priority in the
group, you need to tell the HA manager to not move this service
group, you need to tell the HA manager not to instantly move this service
instantly back by setting the nofailback option.
back by setting the nofailback option.
Another scenario is when a service was fenced and it got recovered to
Another scenario is when a service was fenced and it got recovered to
another node. The admin tries to repair the fenced node and brings it
another node. The admin tries to repair the fenced node and brings it
up online again to investigate the failure cause and check if it runs
up online again to investigate the cause of failure and check if it runs
stable again. Setting the nofailback flag prevents that the
stably again. Setting the nofailback flag prevents the recovered services from
recovered services move straight back to the fenced node.
moving straight back to the fenced node.
Fencing
Fencing
On node failures, fencing ensures that the erroneous node is
On node failures, fencing ensures that the erroneous node is
guaranteed to be offline. This is required to make sure that no
guaranteed to be offline. This is required to make sure that no
resource runs twice when it gets recovered on another node. This is a
resource runs twice when it gets recovered on another node. This is a
really important task, because without, it would not be possible to
really important task, because without this, it would not be possible to
recover a resource on another node.
recover a resource on another node.
If a node would not get fenced, it would be in an unknown state where
If a node did not get fenced, it would be in an unknown state where
it may have still access to shared resources. This is really
it may have still access to shared resources. This is really
dangerous! Imagine that every network but the storage one broke. Now,
dangerous! Imagine that every network but the storage one broke. Now,
while not reachable from the public network, the VM still runs and
while not reachable from the public network, the VM still runs and
writes to the shared storage.
writes to the shared storage.
If we then simply start up this VM on another node, we would get a
If we then simply start up this VM on another node, we would get a
dangerous race conditions because we write from both nodes. Such
dangerous race condition, because we write from both nodes. Such
condition can destroy all VM data and the whole VM could be rendered
conditions can destroy all VM data and the whole VM could be rendered
unusable. The recovery could also fail if the storage protects from
unusable. The recovery could also fail if the storage protects against
multiple mounts.
multiple mounts.
How Proxmox VE Fences
How Proxmox VE Fences
Line 423: Line 461:
isolate nodes by disabling complete network traffic on the switch
isolate nodes by disabling complete network traffic on the switch
self fencing using watchdog timers
self fencing using watchdog timers
Watchdog timers are widely used in critical and dependable systems
Watchdog timers have been widely used in critical and dependable systems
since the beginning of micro controllers. They are often independent
since the beginning of microcontrollers. They are often simple, independent
and simple integrated circuits which are used to detect and recover
integrated circuits which are used to detect and recover from computer malfunctions.
from computer malfunctions.
During normal operation, ha-manager regularly resets the watchdog
During normal operation, ha-manager regularly resets the watchdog
timer to prevent it from elapsing. If, due to a hardware fault or
timer to prevent it from elapsing. If, due to a hardware fault or
program error, the computer fails to reset the watchdog, the timer
program error, the computer fails to reset the watchdog, the timer
will elapse and triggers a reset of the whole server (reboot).
will elapse and trigger a reset of the whole server (reboot).
Recent server motherboards often include such hardware watchdogs, but
Recent server motherboards often include such hardware watchdogs, but
these need to be configured. If no watchdog is available or
these need to be configured. If no watchdog is available or
Line 443: Line 480:
# select watchdog module (default is softdog)
# select watchdog module (default is softdog)
WATCHDOG_MODULE=iTCO_wdt
WATCHDOG_MODULE=iTCO_wdt
This configuration is read by the watchdog-mux service, which load
This configuration is read by the watchdog-mux service, which loads
the specified module at startup.
the specified module at startup.
Recover Fenced Services
Recover Fenced Services
Line 457: Line 494:
of an overloaded node.
of an overloaded node.
On node failure, the CRM distributes services to the
On node failure, the CRM distributes services to the
remaining nodes. This increase the service count on those nodes, and
remaining nodes. This increases the service count on those nodes, and
can lead to high load, especially on small clusters. Please design
can lead to high load, especially on small clusters. Please design
your cluster so that it can handle such worst case scenarios.
your cluster so that it can handle such worst case scenarios.
Start Failure Policy
Start Failure Policy
The start failure policy comes in effect if a service failed to start on a
The start failure policy comes into effect if a service failed to start on a
node once ore more times. It can be used to configure how often a restart
node one or more times. It can be used to configure how often a restart
should be triggered on the same node and how often a service should be
should be triggered on the same node and how often a service should be
relocated so that it gets a try to be started on another node.
relocated, so that it has an attempt to be started on another node.
The aim of this policy is to circumvent temporary unavailability of shared
The aim of this policy is to circumvent temporary unavailability of shared
resources on a specific node. For example, if a shared storage isn’t available
resources on a specific node. For example, if a shared storage isn’t available
on a quorate node anymore, e.g. network problems, but still on other nodes,
on a quorate node anymore, for instance due to network problems, but is still
the relocate policy allows then that the service gets started nonetheless.
available on other nodes, the relocate policy allows the service to start
nonetheless.
There are two service start recover policy settings which can be configured
There are two service start recover policy settings which can be configured
specific for each resource.
specific for each resource.
max_restart
max_restart
Maximum number of tries to restart an failed service on the actual
Maximum number of attempts to restart a failed service on the actual
node.  The default is set to one.
node.  The default is set to one.
max_relocate
max_relocate
Maximum number of tries to relocate the service to a different node.
Maximum number of attempts to relocate the service to a different node.
A relocate only happens after the max_restart value is exceeded on the
A relocate only happens after the max_restart value is exceeded on the
actual node. The default is set to one.
actual node. The default is set to one.
Line 483: Line 521:
repeated.
repeated.
Error Recovery
Error Recovery
If after all tries the service state could not be recovered it gets
If, after all attempts, the service state could not be recovered, it gets
placed in an error state. In this state the service won’t get touched
placed in an error state. In this state, the service won’t get touched
by the HA stack anymore. The only way out is disabling a service:
by the HA stack anymore. The only way out is disabling a service:
# ha-manager set vm:100 --state disabled
# ha-manager set vm:100 --state disabled
Line 495: Line 533:
after you fixed all errors you may request that the service starts again
after you fixed all errors you may request that the service starts again
Package Updates
Package Updates
When updating the ha-manager you should do one node after the other, never
When updating the ha-manager, you should do one node after the other, never
all at once for various reasons. First, while we test our software
all at once for various reasons. First, while we test our software
thoughtfully, a bug affecting your specific setup cannot totally be ruled out.
thoroughly, a bug affecting your specific setup cannot totally be ruled out.
Upgrading one node after the other and checking the functionality of each node
Updating one node after the other and checking the functionality of each node
after finishing the update helps to recover from an eventual problems, while
after finishing the update helps to recover from eventual problems, while
updating all could render you in a broken cluster state and is generally not
updating all at once could result in a broken cluster and is generally not
good practice.
good practice.
Also, the Proxmox VE HA stack uses a request acknowledge protocol to perform
Also, the Proxmox VE HA stack uses a request acknowledge protocol to perform
actions between the cluster and the local resource manager. For restarting,
actions between the cluster and the local resource manager. For restarting,
the LRM makes a request to the CRM to freeze all its services. This prevents
the LRM makes a request to the CRM to freeze all its services. This prevents
that they get touched by the Cluster during the short time the LRM is restarting.
them from getting touched by the Cluster during the short time the LRM is restarting.
After that the LRM may safely close the watchdog during a restart.
After that, the LRM may safely close the watchdog during a restart.
Such a restart happens normally during a package update and, as already stated,
Such a restart happens normally during a package update and, as already stated,
an active master CRM is needed to acknowledge the requests from the LRM. If
an active master CRM is needed to acknowledge the requests from the LRM. If
Line 512: Line 550:
case, may result in a reset triggered by the watchdog.
case, may result in a reset triggered by the watchdog.
Node Maintenance
Node Maintenance
It is sometimes possible to shutdown or reboot a node to do
Sometimes it is necessary to perform maintenance on a node, such as replacing
maintenance tasks. Either to replace hardware, or simply to install a
hardware or simply installing a new kernel image. This also applies while the
new kernel image.
HA stack is in use.
The HA stack can support you mainly in two types of maintenance:
for general shutdowns or reboots, the behavior can be configured, see
  Shutdown Policy.
for maintenance that does not require a shutdown or reboot, or that should
  not be switched off automatically after only one reboot, you can enable the
  manual maintenance mode.
Maintenance Mode
You can use the manual maintenance mode to mark the node as unavailable for HA
operation, prompting all services managed by HA to migrate to other nodes.
The target nodes for these migrations are selected from the other currently
available nodes, and determined by the HA group configuration and the configured
cluster resource scheduler (CRS) mode.
During each migration, the original node will be recorded in the HA managers'
state, so that the service can be moved back again automatically once the
maintenance mode is disabled and the node is back online.
Currently you can enabled or disable the maintenance mode using the ha-manager
CLI tool.
Enabling maintenance mode for a node
# ha-manager crm-command node-maintenance enable NODENAME
This will queue a CRM command, when the manager processes this command it will
record the request for maintenance-mode in the manager status. This allows you
to submit the command on any node, not just on the one you want to place in, or
out of the maintenance mode.
Once the LRM on the respective node picks the command up it will mark itself as
unavailable, but still process all migration commands. This means that the LRM
self-fencing watchdog will stay active until all active services got moved, and
all running workers finished.
Note that the LRM status will read maintenance mode as soon as the LRM
picked the requested state up, not only when all services got moved away, this
user experience is planned to be improved in the future.
For now, you can check for any active HA service left on the node, or watching
out for a log line like: pve-ha-lrm[PID]: watchdog closed (disabled) to know
when the node finished its transition into the maintenance mode.
The manual maintenance mode is not automatically deleted on node reboot,
but only if it is either manually deactivated using the ha-manager CLI or if
the manager-status is manually cleared.
Disabling maintenance mode for a node
# ha-manager crm-command node-maintenance disable NODENAME
The process of disabling the manual maintenance mode is similar to enabling it.
Using the ha-manager CLI command shown above will queue a CRM command that,
once processed, marks the respective LRM node as available again.
If you deactivate the maintenance mode, all services that were on the node when
the maintenance mode was activated will be moved back.
Shutdown Policy
Below you will find a description of the different HA policies for a node
shutdown. Currently Conditional is the default due to backward compatibility.
Some users may find that Migrate behaves more as expected.
The shutdown policy can be configured in the Web UI (Datacenter → Options
→ HA Settings), or directly in datacenter.cfg:
ha: shutdown_policy=<value>
Migrate
Once the Local Resource manager (LRM) gets a shutdown request and this policy
is enabled, it will mark itself as unavailable for the current HA manager.
This triggers a migration of all HA Services currently located on this node.
The LRM will try to delay the shutdown process, until all running services get
moved away. But, this expects that the running services can be migrated to
another node. In other words, the service must not be locally bound, for example
by using hardware passthrough. As non-group member nodes are considered as
runnable target if no group member is available, this policy can still be used
when making use of HA groups with only some nodes selected. But, marking a group
as restricted tells the HA manager that the service cannot run outside of the
chosen set of nodes. If all of those nodes are unavailable, the shutdown will
hang until you manually intervene. Once the shut down node comes back online
again, the previously displaced services will be moved back, if they were not
already manually migrated in-between.
The watchdog is still active during the migration process on shutdown.
If the node loses quorum it will be fenced and the services will be recovered.
If you start a (previously stopped) service on a node which is currently being
maintained, the node needs to be fenced to ensure that the service can be moved
and started on another available node.
Failover
This mode ensures that all services get stopped, but that they will also be
recovered, if the current node is not online soon. It can be useful when doing
maintenance on a cluster scale, where live-migrating VMs may not be possible if
too many nodes are powered off at a time, but you still want to ensure HA
services get recovered and started again as soon as possible.
Freeze
This mode ensures that all services get stopped and frozen, so that they won’t
get recovered until the current node is online again.
Conditional
The Conditional shutdown policy automatically detects if a shutdown or a
reboot is requested, and changes behaviour accordingly.
Shutdown
Shutdown
A shutdown (poweroff) is usually done if the node is planned to stay
A shutdown (poweroff) is usually done if it is planned for the node to stay
down for some time. The LRM stops all managed services in that
down for some time. The LRM stops all managed services in this case. This means
case. This means that other nodes will take over those service
that other nodes will take over those services afterwards.
afterwards.
Recent hardware has large amounts of memory (RAM). So we stop all
Recent hardware has large amounts of RAM. So we stop all
resources, then restart them to avoid online migration of all that RAM. If you
resources, then restart them to avoid online migration of all that
want to use online migration, you need to invoke that manually before you
RAM. If you want to use online migration, you need to invoke that
shutdown the node.
manually before you shutdown the node.
Reboot
Reboot
Node reboots are initiated with the reboot command. This is usually
Node reboots are initiated with the reboot command. This is usually done
done after installing a new kernel. Please note that this is different
after installing a new kernel. Please note that this is different from
from “shutdown”, because the node immediately starts again.
“shutdown”, because the node immediately starts again.
The LRM tells the CRM that it wants to restart, and waits until the
The LRM tells the CRM that it wants to restart, and waits until the CRM puts
CRM puts all resources into the freeze state (same mechanism is used
all resources into the freeze state (same mechanism is used for
for Package Updates). This prevents
Package Updates). This prevents those resources
that those resources are moved to other nodes. Instead, the CRM start
from being moved to other nodes. Instead, the CRM starts the resources after the
the resources after the reboot on the same node.
reboot on the same node.
Manual Resource Movement
Manual Resource Movement
Last but not least, you can also move resources manually to other
Last but not least, you can also manually move resources to other nodes, before
nodes before you shutdown or restart a node. The advantage is that you
you shutdown or restart a node. The advantage is that you have full control,
have full control, and you can decide if you want to use online
and you can decide if you want to use online migration or not.
migration or not.
Please do not kill services like pve-ha-crm, pve-ha-lrm or
Please do not kill services like pve-ha-crm, pve-ha-lrm or
watchdog-mux. They manage and use the watchdog, so this can result
watchdog-mux. They manage and use the watchdog, so this can result in an
in a node reboot.
immediate node reboot or even reset.
Cluster Resource Scheduling
The cluster resource scheduler (CRS) mode controls how HA selects nodes for the
recovery of a service as well as for migrations that are triggered by a
shutdown policy. The default mode is basic, you can change it in the Web UI
(Datacenter → Options), or directly in datacenter.cfg:
crs: ha=static
The change will be in effect starting with the next manager round (after a few
seconds).
For each service that needs to be recovered or migrated, the scheduler
iteratively chooses the best node among the nodes with the highest priority in
the service’s group.
There are plans to add modes for (static and dynamic) load-balancing in
the future.
Basic Scheduler
The number of active HA services on each node is used to choose a recovery node.
Non-HA-managed services are currently not counted.
Static-Load Scheduler
The static mode is still a technology preview.
Static usage information from HA services on each node is used to choose a
recovery node. Usage of non-HA-managed services is currently not considered.
For this selection, each node in turn is considered as if the service was
already running on it, using CPU and memory usage from the associated guest
configuration. Then for each such alternative, CPU and memory usage of all nodes
are considered, with memory being weighted much more, because it’s a truly
limited resource. For both, CPU and memory, highest usage among nodes (weighted
more, as ideally no node should be overcommitted) and average usage of all nodes
(to still be able to distinguish in case there already is a more highly
committed node) are considered.
The more services the more possible combinations there are, so it’s
currently not recommended to use it if you have thousands of HA managed
services.
CRS Scheduling Points
The CRS algorithm is not applied for every service in every round, since this
would mean a large number of constant migrations. Depending on the workload,
this could put more strain on the cluster than could be avoided by constant
balancing.
That’s why the Proxmox VE HA manager favors keeping services on their current node.
The CRS is currently used at the following scheduling points:
Service recovery (always active). When a node with active HA services fails,
  all its services need to be recovered to other nodes. The CRS algorithm will
  be used here to balance that recovery over the remaining nodes.
HA group config changes (always active). If a node is removed from a group,
  or its priority is reduced, the HA stack will use the CRS algorithm to find a
  new target node for the HA services in that group, matching the adapted
  priority constraints.
HA service stopped → start transition (opt-in). Requesting that a stopped
  service should be started is an good opportunity to check for the best suited
  node as per the CRS algorithm, as moving stopped services is  cheaper to do
  than moving them started, especially if their disk volumes reside on shared
  storage. You can enable this by setting the ha-rebalance-on-start
  CRS option in the datacenter config. You can change that option also in the
  Web UI, under Datacenter → Options → Cluster Resource Scheduling.
</pvehide>
</pvehide>
<!--PVE_IMPORT_END_MARKER-->
<!--PVE_IMPORT_END_MARKER-->

Latest revision as of 12:09, 28 November 2024

Our modern society depends heavily on information provided by computers over the network. Mobile devices amplified that dependency, because people can access the network any time from anywhere. If you provide such services, it is very important that they are available most of the time.

We can mathematically define the availability as the ratio of (A), the total time a service is capable of being used during a given interval to (B), the length of the interval. It is normally expressed as a percentage of uptime in a given year.

Table 1. Availability - Downtime per Year
Availability % Downtime per year

99

3.65 days

99.9

8.76 hours

99.99

52.56 minutes

99.999

5.26 minutes

99.9999

31.5 seconds

99.99999

3.15 seconds

There are several ways to increase availability. The most elegant solution is to rewrite your software, so that you can run it on several hosts at the same time. The software itself needs to have a way to detect errors and do failover. If you only want to serve read-only web pages, then this is relatively simple. However, this is generally complex and sometimes impossible, because you cannot modify the software yourself. The following solutions works without modifying the software:

  • Use reliable “server” components

    Note Computer components with the same functionality can have varying reliability numbers, depending on the component quality. Most vendors sell components with higher reliability as “server” components - usually at higher price.
  • Eliminate single point of failure (redundant components)

    • use an uninterruptible power supply (UPS)

    • use redundant power supplies in your servers

    • use ECC-RAM

    • use redundant network hardware

    • use RAID for local storage

    • use distributed, redundant storage for VM data

  • Reduce downtime

    • rapidly accessible administrators (24/7)

    • availability of spare parts (other nodes in a Proxmox VE cluster)

    • automatic error detection (provided by ha-manager)

    • automatic failover (provided by ha-manager)

Virtualization environments like Proxmox VE make it much easier to reach high availability because they remove the “hardware” dependency. They also support the setup and use of redundant storage and network devices, so if one host fails, you can simply start those services on another host within your cluster.

Better still, Proxmox VE provides a software stack called ha-manager, which can do that automatically for you. It is able to automatically detect errors and do automatic failover.

Proxmox VE ha-manager works like an “automated” administrator. First, you configure what resources (VMs, containers, …) it should manage. Then, ha-manager observes the correct functionality, and handles service failover to another node in case of errors. ha-manager can also handle normal user requests which may start, stop, relocate and migrate a service.

But high availability comes at a price. High quality components are more expensive, and making them redundant doubles the costs at least. Additional spare parts increase costs further. So you should carefully calculate the benefits, and compare with those additional costs.

Tip Increasing availability from 99% to 99.9% is relatively simple. But increasing availability from 99.9999% to 99.99999% is very hard and costly. ha-manager has typical error detection and failover times of about 2 minutes, so you can get no more than 99.999% availability.

Requirements

You must meet the following requirements before you start with HA:

  • at least three cluster nodes (to get reliable quorum)

  • shared storage for VMs and containers

  • hardware redundancy (everywhere)

  • use reliable “server” components

  • hardware watchdog - if not available we fall back to the linux kernel software watchdog (softdog)

  • optional hardware fencing devices

Resources

We call the primary management unit handled by ha-manager a resource. A resource (also called “service”) is uniquely identified by a service ID (SID), which consists of the resource type and a type specific ID, for example vm:100. That example would be a resource of type vm (virtual machine) with the ID 100.

For now we have two important resources types - virtual machines and containers. One basic idea here is that we can bundle related software into such a VM or container, so there is no need to compose one big service from other services, as was done with rgmanager. In general, a HA managed resource should not depend on other resources.

Management Tasks

This section provides a short overview of common management tasks. The first step is to enable HA for a resource. This is done by adding the resource to the HA resource configuration. You can do this using the GUI, or simply use the command-line tool, for example:

# ha-manager add vm:100

The HA stack now tries to start the resources and keep them running. Please note that you can configure the “requested” resources state. For example you may want the HA stack to stop the resource:

# ha-manager set vm:100 --state stopped

and start it again later:

# ha-manager set vm:100 --state started

You can also use the normal VM and container management commands. They automatically forward the commands to the HA stack, so

# qm start 100

simply sets the requested state to started. The same applies to qm stop, which sets the requested state to stopped.

Note The HA stack works fully asynchronous and needs to communicate with other cluster members. Therefore, it takes some seconds until you see the result of such actions.

To view the current HA resource configuration use:

# ha-manager config
vm:100
        state stopped

And you can view the actual HA manager and resource state with:

# ha-manager status
quorum OK
master node1 (active, Wed Nov 23 11:07:23 2016)
lrm elsa (active, Wed Nov 23 11:07:19 2016)
service vm:100 (node1, started)

You can also initiate resource migration to other nodes:

# ha-manager migrate vm:100 node2

This uses online migration and tries to keep the VM running. Online migration needs to transfer all used memory over the network, so it is sometimes faster to stop the VM, then restart it on the new node. This can be done using the relocate command:

# ha-manager relocate vm:100 node2

Finally, you can remove the resource from the HA configuration using the following command:

# ha-manager remove vm:100
Note This does not start or stop the resource.

But all HA related tasks can be done in the GUI, so there is no need to use the command line at all.

How It Works

This section provides a detailed description of the Proxmox VE HA manager internals. It describes all involved daemons and how they work together. To provide HA, two daemons run on each node:

pve-ha-lrm

The local resource manager (LRM), which controls the services running on the local node. It reads the requested states for its services from the current manager status file and executes the respective commands.

pve-ha-crm

The cluster resource manager (CRM), which makes the cluster-wide decisions. It sends commands to the LRM, processes the results, and moves resources to other nodes if something fails. The CRM also handles node fencing.

Note
Locks in the LRM & CRM
Locks are provided by our distributed configuration file system (pmxcfs). They are used to guarantee that each LRM is active once and working. As an LRM only executes actions when it holds its lock, we can mark a failed node as fenced if we can acquire its lock. This then lets us recover any failed HA services securely without any interference from the now unknown failed node. This all gets supervised by the CRM which currently holds the manager master lock.

Service States

The CRM uses a service state enumeration to record the current service state. This state is displayed on the GUI and can be queried using the ha-manager command-line tool:

# ha-manager status
quorum OK
master elsa (active, Mon Nov 21 07:23:29 2016)
lrm elsa (active, Mon Nov 21 07:23:22 2016)
service ct:100 (elsa, stopped)
service ct:102 (elsa, started)
service vm:501 (elsa, started)

Here is the list of possible states:

stopped

Service is stopped (confirmed by LRM). If the LRM detects a stopped service is still running, it will stop it again.

request_stop

Service should be stopped. The CRM waits for confirmation from the LRM.

stopping

Pending stop request. But the CRM did not get the request so far.

started

Service is active an LRM should start it ASAP if not already running. If the Service fails and is detected to be not running the LRM restarts it (see Start Failure Policy).

starting

Pending start request. But the CRM has not got any confirmation from the LRM that the service is running.

fence

Wait for node fencing as the service node is not inside the quorate cluster partition (see Fencing). As soon as node gets fenced successfully the service will be placed into the recovery state.

recovery

Wait for recovery of the service. The HA manager tries to find a new node where the service can run on. This search depends not only on the list of online and quorate nodes, but also if the service is a group member and how such a group is limited. As soon as a new available node is found, the service will be moved there and initially placed into stopped state. If it’s configured to run the new node will do so.

freeze

Do not touch the service state. We use this state while we reboot a node, or when we restart the LRM daemon (see Package Updates).

ignored

Act as if the service were not managed by HA at all. Useful, when full control over the service is desired temporarily, without removing it from the HA configuration.

migrate

Migrate service (live) to other node.

error

Service is disabled because of LRM errors. Needs manual intervention (see Error Recovery).

queued

Service is newly added, and the CRM has not seen it so far.

disabled

Service is stopped and marked as disabled

Local Resource Manager

The local resource manager (pve-ha-lrm) is started as a daemon on boot and waits until the HA cluster is quorate and thus cluster-wide locks are working.

It can be in three states:

wait for agent lock

The LRM waits for our exclusive lock. This is also used as idle state if no service is configured.

active

The LRM holds its exclusive lock and has services configured.

lost agent lock

The LRM lost its lock, this means a failure happened and quorum was lost.

After the LRM gets in the active state it reads the manager status file in /etc/pve/ha/manager_status and determines the commands it has to execute for the services it owns. For each command a worker gets started, these workers are running in parallel and are limited to at most 4 by default. This default setting may be changed through the datacenter configuration key max_worker. When finished the worker process gets collected and its result saved for the CRM.

Note
Maximum Concurrent Worker Adjustment Tips
The default value of at most 4 concurrent workers may be unsuited for a specific setup. For example, 4 live migrations may occur at the same time, which can lead to network congestions with slower networks and/or big (memory wise) services. Also, ensure that in the worst case, congestion is at a minimum, even if this means lowering the max_worker value. On the contrary, if you have a particularly powerful, high-end setup you may also want to increase it.

Each command requested by the CRM is uniquely identifiable by a UID. When the worker finishes, its result will be processed and written in the LRM status file /etc/pve/nodes/<nodename>/lrm_status. There the CRM may collect it and let its state machine - respective to the commands output - act on it.

The actions on each service between CRM and LRM are normally always synced. This means that the CRM requests a state uniquely marked by a UID, the LRM then executes this action one time and writes back the result, which is also identifiable by the same UID. This is needed so that the LRM does not execute an outdated command. The only exceptions to this behaviour are the stop and error commands; these two do not depend on the result produced and are executed always in the case of the stopped state and once in the case of the error state.

Note
Read the Logs
The HA Stack logs every action it makes. This helps to understand what and also why something happens in the cluster. Here its important to see what both daemons, the LRM and the CRM, did. You may use journalctl -u pve-ha-lrm on the node(s) where the service is and the same command for the pve-ha-crm on the node which is the current master.

Cluster Resource Manager

The cluster resource manager (pve-ha-crm) starts on each node and waits there for the manager lock, which can only be held by one node at a time. The node which successfully acquires the manager lock gets promoted to the CRM master.

It can be in three states:

wait for agent lock

The CRM waits for our exclusive lock. This is also used as idle state if no service is configured

active

The CRM holds its exclusive lock and has services configured

lost agent lock

The CRM lost its lock, this means a failure happened and quorum was lost.

Its main task is to manage the services which are configured to be highly available and try to always enforce the requested state. For example, a service with the requested state started will be started if its not already running. If it crashes it will be automatically started again. Thus the CRM dictates the actions the LRM needs to execute.

When a node leaves the cluster quorum, its state changes to unknown. If the current CRM can then secure the failed node’s lock, the services will be stolen and restarted on another node.

When a cluster member determines that it is no longer in the cluster quorum, the LRM waits for a new quorum to form. As long as there is no quorum the node cannot reset the watchdog. This will trigger a reboot after the watchdog times out (this happens after 60 seconds).

HA Simulator

screenshot/gui-ha-manager-status.png

By using the HA simulator you can test and learn all functionalities of the Proxmox VE HA solutions.

By default, the simulator allows you to watch and test the behaviour of a real-world 3 node cluster with 6 VMs. You can also add or remove additional VMs or Container.

You do not have to setup or configure a real cluster, the HA simulator runs out of the box.

Install with apt:

apt install pve-ha-simulator

You can even install the package on any Debian-based system without any other Proxmox VE packages. For that you will need to download the package and copy it to the system you want to run it on for installation. When you install the package with apt from the local file system it will also resolve the required dependencies for you.

To start the simulator on a remote machine you must have an X11 redirection to your current system.

If you are on a Linux machine you can use:

ssh root@<IPofPVE> -Y

On Windows it works with mobaxterm.

After connecting to an existing Proxmox VE with the simulator installed or installing it on your local Debian-based system manually, you can try it out as follows.

First you need to create a working directory where the simulator saves its current state and writes its default config:

mkdir working

Then, simply pass the created directory as a parameter to pve-ha-simulator:

pve-ha-simulator working/

You can then start, stop, migrate the simulated HA services, or even check out what happens on a node failure.

Configuration

The HA stack is well integrated into the Proxmox VE API. So, for example, HA can be configured via the ha-manager command-line interface, or the Proxmox VE web interface - both interfaces provide an easy way to manage HA. Automation tools can use the API directly.

All HA configuration files are within /etc/pve/ha/, so they get automatically distributed to the cluster nodes, and all nodes share the same HA configuration.

Resources

screenshot/gui-ha-manager-status.png

The resource configuration file /etc/pve/ha/resources.cfg stores the list of resources managed by ha-manager. A resource configuration inside that list looks like this:

<type>: <name>
        <property> <value>
        ...

It starts with a resource type followed by a resource specific name, separated with colon. Together this forms the HA resource ID, which is used by all ha-manager commands to uniquely identify a resource (example: vm:100 or ct:101). The next lines contain additional properties:

comment: <string>

Description.

group: <string>

The HA group identifier.

max_relocate: <integer> (0 - N) (default = 1)

Maximal number of service relocate tries when a service failes to start.

max_restart: <integer> (0 - N) (default = 1)

Maximal number of tries to restart the service on a node after its start failed.

state: <disabled | enabled | ignored | started | stopped> (default = started)

Requested resource state. The CRM reads this state and acts accordingly. Please note that enabled is just an alias for started.

started

The CRM tries to start the resource. Service state is set to started after successful start. On node failures, or when start fails, it tries to recover the resource. If everything fails, service state it set to error.

stopped

The CRM tries to keep the resource in stopped state, but it still tries to relocate the resources on node failures.

disabled

The CRM tries to put the resource in stopped state, but does not try to relocate the resources on node failures. The main purpose of this state is error recovery, because it is the only way to move a resource out of the error state.

ignored

The resource gets removed from the manager status and so the CRM and the LRM do not touch the resource anymore. All {pve} API calls affecting this resource will be executed, directly bypassing the HA stack. CRM commands will be thrown away while there source is in this state. The resource will not get relocated on node failures.

Here is a real world example with one VM and one container. As you see, the syntax of those files is really simple, so it is even possible to read or edit those files using your favorite editor:

Configuration Example (/etc/pve/ha/resources.cfg)
vm: 501
    state started
    max_relocate 2

ct: 102
    # Note: use default settings for everything
screenshot/gui-ha-manager-add-resource.png

The above config was generated using the ha-manager command-line tool:

# ha-manager add vm:501 --state started --max_relocate 2
# ha-manager add ct:102

Groups

screenshot/gui-ha-manager-groups-view.png

The HA group configuration file /etc/pve/ha/groups.cfg is used to define groups of cluster nodes. A resource can be restricted to run only on the members of such group. A group configuration look like this:

group: <group>
       nodes <node_list>
       <property> <value>
       ...
comment: <string>

Description.

nodes: <node>[:<pri>]{,<node>[:<pri>]}*

List of cluster node members, where a priority can be given to each node. A resource bound to a group will run on the available nodes with the highest priority. If there are more nodes in the highest priority class, the services will get distributed to those nodes. The priorities have a relative meaning only.

nofailback: <boolean> (default = 0)

The CRM tries to run services on the node with the highest priority. If a node with higher priority comes online, the CRM migrates the service to that node. Enabling nofailback prevents that behavior.

restricted: <boolean> (default = 0)

Resources bound to restricted groups may only run on nodes defined by the group. The resource will be placed in the stopped state if no group node member is online. Resources on unrestricted groups may run on any cluster node if all group members are offline, but they will migrate back as soon as a group member comes online. One can implement a preferred node behavior using an unrestricted group with only one member.

screenshot/gui-ha-manager-add-group.png

A common requirement is that a resource should run on a specific node. Usually the resource is able to run on other nodes, so you can define an unrestricted group with a single member:

# ha-manager groupadd prefer_node1 --nodes node1

For bigger clusters, it makes sense to define a more detailed failover behavior. For example, you may want to run a set of services on node1 if possible. If node1 is not available, you want to run them equally split on node2 and node3. If those nodes also fail, the services should run on node4. To achieve this you could set the node list to:

# ha-manager groupadd mygroup1 -nodes "node1:2,node2:1,node3:1,node4"

Another use case is if a resource uses other resources only available on specific nodes, lets say node1 and node2. We need to make sure that HA manager does not use other nodes, so we need to create a restricted group with said nodes:

# ha-manager groupadd mygroup2 -nodes "node1,node2" -restricted

The above commands created the following group configuration file:

Configuration Example (/etc/pve/ha/groups.cfg)
group: prefer_node1
       nodes node1

group: mygroup1
       nodes node2:1,node4,node1:2,node3:1

group: mygroup2
       nodes node2,node1
       restricted 1

The nofailback options is mostly useful to avoid unwanted resource movements during administration tasks. For example, if you need to migrate a service to a node which doesn’t have the highest priority in the group, you need to tell the HA manager not to instantly move this service back by setting the nofailback option.

Another scenario is when a service was fenced and it got recovered to another node. The admin tries to repair the fenced node and brings it up online again to investigate the cause of failure and check if it runs stably again. Setting the nofailback flag prevents the recovered services from moving straight back to the fenced node.

Fencing

On node failures, fencing ensures that the erroneous node is guaranteed to be offline. This is required to make sure that no resource runs twice when it gets recovered on another node. This is a really important task, because without this, it would not be possible to recover a resource on another node.

If a node did not get fenced, it would be in an unknown state where it may have still access to shared resources. This is really dangerous! Imagine that every network but the storage one broke. Now, while not reachable from the public network, the VM still runs and writes to the shared storage.

If we then simply start up this VM on another node, we would get a dangerous race condition, because we write from both nodes. Such conditions can destroy all VM data and the whole VM could be rendered unusable. The recovery could also fail if the storage protects against multiple mounts.

How Proxmox VE Fences

There are different methods to fence a node, for example, fence devices which cut off the power from the node or disable their communication completely. Those are often quite expensive and bring additional critical components into a system, because if they fail you cannot recover any service.

We thus wanted to integrate a simpler fencing method, which does not require additional external hardware. This can be done using watchdog timers.

Possible Fencing Methods
  • external power switches

  • isolate nodes by disabling complete network traffic on the switch

  • self fencing using watchdog timers

Watchdog timers have been widely used in critical and dependable systems since the beginning of microcontrollers. They are often simple, independent integrated circuits which are used to detect and recover from computer malfunctions.

During normal operation, ha-manager regularly resets the watchdog timer to prevent it from elapsing. If, due to a hardware fault or program error, the computer fails to reset the watchdog, the timer will elapse and trigger a reset of the whole server (reboot).

Recent server motherboards often include such hardware watchdogs, but these need to be configured. If no watchdog is available or configured, we fall back to the Linux Kernel softdog. While still reliable, it is not independent of the servers hardware, and thus has a lower reliability than a hardware watchdog.

Configure Hardware Watchdog

By default, all hardware watchdog modules are blocked for security reasons. They are like a loaded gun if not correctly initialized. To enable a hardware watchdog, you need to specify the module to load in /etc/default/pve-ha-manager, for example:

# select watchdog module (default is softdog)
WATCHDOG_MODULE=iTCO_wdt

This configuration is read by the watchdog-mux service, which loads the specified module at startup.

Recover Fenced Services

After a node failed and its fencing was successful, the CRM tries to move services from the failed node to nodes which are still online.

The selection of nodes, on which those services gets recovered, is influenced by the resource group settings, the list of currently active nodes, and their respective active service count.

The CRM first builds a set out of the intersection between user selected nodes (from group setting) and available nodes. It then choose the subset of nodes with the highest priority, and finally select the node with the lowest active service count. This minimizes the possibility of an overloaded node.

Caution On node failure, the CRM distributes services to the remaining nodes. This increases the service count on those nodes, and can lead to high load, especially on small clusters. Please design your cluster so that it can handle such worst case scenarios.

Start Failure Policy

The start failure policy comes into effect if a service failed to start on a node one or more times. It can be used to configure how often a restart should be triggered on the same node and how often a service should be relocated, so that it has an attempt to be started on another node. The aim of this policy is to circumvent temporary unavailability of shared resources on a specific node. For example, if a shared storage isn’t available on a quorate node anymore, for instance due to network problems, but is still available on other nodes, the relocate policy allows the service to start nonetheless.

There are two service start recover policy settings which can be configured specific for each resource.

max_restart

Maximum number of attempts to restart a failed service on the actual node. The default is set to one.

max_relocate

Maximum number of attempts to relocate the service to a different node. A relocate only happens after the max_restart value is exceeded on the actual node. The default is set to one.

Note The relocate count state will only reset to zero when the service had at least one successful start. That means if a service is re-started without fixing the error only the restart policy gets repeated.

Error Recovery

If, after all attempts, the service state could not be recovered, it gets placed in an error state. In this state, the service won’t get touched by the HA stack anymore. The only way out is disabling a service:

# ha-manager set vm:100 --state disabled

This can also be done in the web interface.

To recover from the error state you should do the following:

  • bring the resource back into a safe and consistent state (e.g.: kill its process if the service could not be stopped)

  • disable the resource to remove the error flag

  • fix the error which led to this failures

  • after you fixed all errors you may request that the service starts again

Package Updates

When updating the ha-manager, you should do one node after the other, never all at once for various reasons. First, while we test our software thoroughly, a bug affecting your specific setup cannot totally be ruled out. Updating one node after the other and checking the functionality of each node after finishing the update helps to recover from eventual problems, while updating all at once could result in a broken cluster and is generally not good practice.

Also, the Proxmox VE HA stack uses a request acknowledge protocol to perform actions between the cluster and the local resource manager. For restarting, the LRM makes a request to the CRM to freeze all its services. This prevents them from getting touched by the Cluster during the short time the LRM is restarting. After that, the LRM may safely close the watchdog during a restart. Such a restart happens normally during a package update and, as already stated, an active master CRM is needed to acknowledge the requests from the LRM. If this is not the case the update process can take too long which, in the worst case, may result in a reset triggered by the watchdog.

Node Maintenance

Sometimes it is necessary to perform maintenance on a node, such as replacing hardware or simply installing a new kernel image. This also applies while the HA stack is in use.

The HA stack can support you mainly in two types of maintenance:

  • for general shutdowns or reboots, the behavior can be configured, see Shutdown Policy.

  • for maintenance that does not require a shutdown or reboot, or that should not be switched off automatically after only one reboot, you can enable the manual maintenance mode.

Maintenance Mode

You can use the manual maintenance mode to mark the node as unavailable for HA operation, prompting all services managed by HA to migrate to other nodes.

The target nodes for these migrations are selected from the other currently available nodes, and determined by the HA group configuration and the configured cluster resource scheduler (CRS) mode. During each migration, the original node will be recorded in the HA managers' state, so that the service can be moved back again automatically once the maintenance mode is disabled and the node is back online.

Currently you can enabled or disable the maintenance mode using the ha-manager CLI tool.

Enabling maintenance mode for a node
# ha-manager crm-command node-maintenance enable NODENAME

This will queue a CRM command, when the manager processes this command it will record the request for maintenance-mode in the manager status. This allows you to submit the command on any node, not just on the one you want to place in, or out of the maintenance mode.

Once the LRM on the respective node picks the command up it will mark itself as unavailable, but still process all migration commands. This means that the LRM self-fencing watchdog will stay active until all active services got moved, and all running workers finished.

Note that the LRM status will read maintenance mode as soon as the LRM picked the requested state up, not only when all services got moved away, this user experience is planned to be improved in the future. For now, you can check for any active HA service left on the node, or watching out for a log line like: pve-ha-lrm[PID]: watchdog closed (disabled) to know when the node finished its transition into the maintenance mode.

Note The manual maintenance mode is not automatically deleted on node reboot, but only if it is either manually deactivated using the ha-manager CLI or if the manager-status is manually cleared.
Disabling maintenance mode for a node
# ha-manager crm-command node-maintenance disable NODENAME

The process of disabling the manual maintenance mode is similar to enabling it. Using the ha-manager CLI command shown above will queue a CRM command that, once processed, marks the respective LRM node as available again.

If you deactivate the maintenance mode, all services that were on the node when the maintenance mode was activated will be moved back.

Shutdown Policy

Below you will find a description of the different HA policies for a node shutdown. Currently Conditional is the default due to backward compatibility. Some users may find that Migrate behaves more as expected.

The shutdown policy can be configured in the Web UI (DatacenterOptionsHA Settings), or directly in datacenter.cfg:

ha: shutdown_policy=<value>

Migrate

Once the Local Resource manager (LRM) gets a shutdown request and this policy is enabled, it will mark itself as unavailable for the current HA manager. This triggers a migration of all HA Services currently located on this node. The LRM will try to delay the shutdown process, until all running services get moved away. But, this expects that the running services can be migrated to another node. In other words, the service must not be locally bound, for example by using hardware passthrough. As non-group member nodes are considered as runnable target if no group member is available, this policy can still be used when making use of HA groups with only some nodes selected. But, marking a group as restricted tells the HA manager that the service cannot run outside of the chosen set of nodes. If all of those nodes are unavailable, the shutdown will hang until you manually intervene. Once the shut down node comes back online again, the previously displaced services will be moved back, if they were not already manually migrated in-between.

Note The watchdog is still active during the migration process on shutdown. If the node loses quorum it will be fenced and the services will be recovered.

If you start a (previously stopped) service on a node which is currently being maintained, the node needs to be fenced to ensure that the service can be moved and started on another available node.

Failover

This mode ensures that all services get stopped, but that they will also be recovered, if the current node is not online soon. It can be useful when doing maintenance on a cluster scale, where live-migrating VMs may not be possible if too many nodes are powered off at a time, but you still want to ensure HA services get recovered and started again as soon as possible.

Freeze

This mode ensures that all services get stopped and frozen, so that they won’t get recovered until the current node is online again.

Conditional

The Conditional shutdown policy automatically detects if a shutdown or a reboot is requested, and changes behaviour accordingly.

Shutdown

A shutdown (poweroff) is usually done if it is planned for the node to stay down for some time. The LRM stops all managed services in this case. This means that other nodes will take over those services afterwards.

Note Recent hardware has large amounts of memory (RAM). So we stop all resources, then restart them to avoid online migration of all that RAM. If you want to use online migration, you need to invoke that manually before you shutdown the node.
Reboot

Node reboots are initiated with the reboot command. This is usually done after installing a new kernel. Please note that this is different from “shutdown”, because the node immediately starts again.

The LRM tells the CRM that it wants to restart, and waits until the CRM puts all resources into the freeze state (same mechanism is used for Package Updates). This prevents those resources from being moved to other nodes. Instead, the CRM starts the resources after the reboot on the same node.

Manual Resource Movement

Last but not least, you can also manually move resources to other nodes, before you shutdown or restart a node. The advantage is that you have full control, and you can decide if you want to use online migration or not.

Note Please do not kill services like pve-ha-crm, pve-ha-lrm or watchdog-mux. They manage and use the watchdog, so this can result in an immediate node reboot or even reset.

Cluster Resource Scheduling

The cluster resource scheduler (CRS) mode controls how HA selects nodes for the recovery of a service as well as for migrations that are triggered by a shutdown policy. The default mode is basic, you can change it in the Web UI (DatacenterOptions), or directly in datacenter.cfg:

crs: ha=static
screenshot/gui-datacenter-options-crs.png

The change will be in effect starting with the next manager round (after a few seconds).

For each service that needs to be recovered or migrated, the scheduler iteratively chooses the best node among the nodes with the highest priority in the service’s group.

Note There are plans to add modes for (static and dynamic) load-balancing in the future.

Basic Scheduler

The number of active HA services on each node is used to choose a recovery node. Non-HA-managed services are currently not counted.

Static-Load Scheduler

Important The static mode is still a technology preview.

Static usage information from HA services on each node is used to choose a recovery node. Usage of non-HA-managed services is currently not considered.

For this selection, each node in turn is considered as if the service was already running on it, using CPU and memory usage from the associated guest configuration. Then for each such alternative, CPU and memory usage of all nodes are considered, with memory being weighted much more, because it’s a truly limited resource. For both, CPU and memory, highest usage among nodes (weighted more, as ideally no node should be overcommitted) and average usage of all nodes (to still be able to distinguish in case there already is a more highly committed node) are considered.

Important The more services the more possible combinations there are, so it’s currently not recommended to use it if you have thousands of HA managed services.

CRS Scheduling Points

The CRS algorithm is not applied for every service in every round, since this would mean a large number of constant migrations. Depending on the workload, this could put more strain on the cluster than could be avoided by constant balancing. That’s why the Proxmox VE HA manager favors keeping services on their current node.

The CRS is currently used at the following scheduling points:

  • Service recovery (always active). When a node with active HA services fails, all its services need to be recovered to other nodes. The CRS algorithm will be used here to balance that recovery over the remaining nodes.

  • HA group config changes (always active). If a node is removed from a group, or its priority is reduced, the HA stack will use the CRS algorithm to find a new target node for the HA services in that group, matching the adapted priority constraints.

  • HA service stopped → start transition (opt-in). Requesting that a stopped service should be started is an good opportunity to check for the best suited node as per the CRS algorithm, as moving stopped services is cheaper to do than moving them started, especially if their disk volumes reside on shared storage. You can enable this by setting the ha-rebalance-on-start CRS option in the datacenter config. You can change that option also in the Web UI, under DatacenterOptionsCluster Resource Scheduling.