LXD 4.20 has been released

Introduction

The LXD team is very excited to announce the release of LXD 4.20!

This is one very busy release with a lot of new features.

VM users will be happy to see the initial implementation of live migration and core scheduling support. Container users are getting new configuration keys to set sysctls.

Then the bulk of the new features are all network related with peer network relationships, network zones for auto-generated DNS and SR-IOV accelerated OVN networks.

And lastly, on the clustering front, it’s now possible to better control what servers will be receiving new workloads.

Enjoy!

New features and highlights

Live migration of virtual machines

LXD now has initial support for live-migration of virtual machines.

This works by simply using lxc move to move between two separate LXD servers or lxc move --target to move within a cluster. Assuming the VM is running, live migration will be attempted.

Using this requires migration.stateful to be enabled on the VM which will prevent it from using GPUs, USB or host PCI devices as well as disable virtiofs (limiting filesystem passthrough to 9p).

The current implementation will effectively perform a stateful stop of the instance immediately followed with migration of the entire data (which will contain the runtime state) and restoration on the target.

This should just take 2-3s when using Ceph but can take significantly longer on other storage backends. We have work scheduled over the next 6 months to improve LXD’s ability to very cheaply and quickly refresh volumes on BTRFS and ZFS which will then allow similar performance for those storage backends.

Network peering for OVN

When using OVN for LXD networking and multiple networks are defined, routing from one network to another currently exits OVN, hits the uplink network and then re-enters OVN. While this is sometimes desired to fully control the network flows, it can be a huge bottleneck.

To address this, LXD now supports network peers. A peer is added on each side of a pair of networks (can be across projects). Once the peer relation is established, OVN will be configured to directly route from one network to the other with traffic never leaving OVN.

Peer relations also allow for easier ACL rules, making it possible to use @some-network/some-peer in the source or destination field to affect traffic coming in or out of a specific peer network.

A new lxc network peer command is used to manage those peers relations.
At the API level, this is all under /1.0/networks/NAME/peers

Specification: [LXD] OVN network to network routing
Documentation: Network Peers | LXD

Network zones (DNS)

Those managing a large set of instances across many projects will often appreciate all instances having valid forward and reverse DNS records available on their entire network.

Up until now, the only option for this was to use the built-in dnsmasq DNS server and its auto-generated DNS zone but this only really works for one network in one project and doesn’t easily integrate within a larger infrastructure. The alternative being to run a completely manually provisioned DNS server on the side.

With this release, LXD introduces the concept of network zones. Those are effectively DNS zones that are tied to LXD networks and can be used for forward DNS records, reverse IPv4 or reverse IPv6 DNS records.

To set it up, one would first create some zones (lxc network zone create), then assign them for the right type of records on the right networks by setting one of:

  • dns.zone.forward
  • dns.zone.reverse.ipv4
  • dns.zone.reverse.ipv6

Lastly the zone must get configured with at least one peer DNS server. This is done using the following configuration keys on the zone itself:

  • peers.NAME.address
  • peers.NAME.key

Either one, or both, of those must be set for a client DNS server to be able to pull the zone from LXD.

Once that’s all setup, your external DNS server can perform a zone transfer (AXFR) from LXD for that zone and then serve it. LXD itself only allows for zone transfers and cannot be directly queried.

Specification: [LXD] Built-in DNS server
Documentation: Network Zones | LXD

SR-IOV acceleration for OVN networking

To further improve the performance of OVN based networking on LXD, we have now added support for SR-IOV acceleration.

The way this works is by using physical network cards which can operate in switchdev mode. Such cards will then provide both a guest facing VF and a host facing representor port. When a system with a suitable card has it in switchdev mode, has the PF added to the OVS bridge and has OVS configured for SR-IOV offload, LXD can automatically allocate VFs for use by containers or VMs.

When all the prerequisites are met, all that needs to be done is set acceleration=sriov on the LXD nic device and LXD will do the rest.

This can lead to an extreme improvement in performance especially on 40G, 100G or 200G networks as it will effectively offload the bulk of the traffic processing directly onto the physical NIC. In such an environment only the first packet in most connections will ever go through the host system and OVS/OVN, all traffic after that point is directly handled in hardware.

Documentation: Instances | LXD

Linux sysctl configuration on containers

A new linux.sysctl.* set of config keys has been introduced. This allows directly setting a specific sysctl to a specific value on container startup.

This can be more flexible than having a sysctl.d entry inside the container and can also enable accessing sysctls which are properly namespaced but require elevated privileges by having LXD apply those from the host.

Documentation: Instances | LXD

Core scheduling for virtual machines

Following our work on core scheduling for containers in LXD 4.19, we have now extended it to virtual machines. When running LXD on a kernel that supports core scheduling, LXD will automatically ensure that all vCPU threads for a particular VM are part of the same core scheduling group.

Core scheduling is designed to allow the use of SMT on CPUs that would otherwise be at risk of attack by the guest due to the Spectre vulnerabilities. Core scheduling ensures that a guest will either use a core/thread pair or that the thread will remain unused while the guest uses the associated core.

Cluster member configuration

Cluster members now have support for configuration keys.

Other than the free form user.KEY=VALUE, the other key being introduced is scheduler.instance which can be used to prevent a particular cluster member from getting automaticly placed instances, instead only getting instances that are directly placed on it through the --target option of the CLI.

Documentation: Clustering | LXD

Improvement to network leases

As network leases are heavily relied upon for the new network zones feature, some improvements have been done.

Some of the improvements are:

  • Support for network leases on OVN networks
  • Uplink networks now showing up in the list of leases
  • IPv6 EUI64 addresses included in the list (when stateful DHCPv6 isn’t enabled)

On the CLI, those records can be seen with lxc network list-leases NETWORK-NAME

Complete changelog

Here is a complete list of all changes in this release:

Full commit list
  • lxd/certificates: remove explicit calls to UpdateCertificateProjects
  • lxd/db/certificates: remove Cluster.UpdateCertificateProjects
  • lxd/db/generate/db/method: fill entity id association tables on create/update
  • lxd/db/certificates.mapper: update generated code
  • lxd/checkfeature: check whether the kernel supports core scheduling
  • lxd/daemon: Fix crash on lxd start when another lxd already running
  • lxd/daemon: Don’t fail shutdown if fail to close cluster DB
  • lxd/daemon: Don’t use Infof and Errorf
  • lxd/metrics: Change ProcsTotal to gauge
  • lxd/instance/drivers: Log metrics failures
  • lxd-agent: Log metrics failures
  • lxd/instance/operationalock: Change lock from using instance ID to use project and instace name
  • lxd/instance/operationalock: Use %q for error quoting
  • lxd/instance/operationlock: Get lock after checking for non-nil operation
  • lxd/instance/drivers/driver/common: operationlock usage
  • lxd/instance/drivers/driver/lxc: operationlock usage
  • lxd/instance/drivers/driver/qemu: operationlock usage
  • lxd/instance/instance/utils: operationlock usage
  • lxd-agent: Drop aggregated cpu stats in metrics
  • test: Kill LXD process if doesn’t start in time
  • lxd/main/shutdown: Fix shutdown regression when running in snap
  • lxc: suggest 20.04 as the first container to launch instead of 18.04
  • lxc: switch from 18.04 to 20.04 for examples of Ubuntu instances
  • i18n: Update translation templates
  • lxc: update wording when a cert is successfully trusted by a remote
  • i18n: Update translation templates
  • lxd/backup/backup/config: Adds ToInstanceDBArgs function
  • lxd/instance/instance/utils: Adds LoadFromBackup function
  • lxd/project/project: Update comment of InstanceParts
  • lxd/instances: Reworks instancesOnDisk to return slice of instance.Instance
  • lxd/instances: Updates instancesShutdown to use instancesOnDisk
  • lxd/patches: Updates patchUpdateFromV11 and patchUpdateFromV15 to use instancesOnDisk
  • lxd/api/internal: Use backupConf.ToInstanceDBArgs in internalImportFromBackup
  • lxd/api/internal/recover: Updates internalRecoverImportInstance to use backupConf.ToInstanceDBArgs
  • lxd/instances: Don’t clear last power state of all instances in a cluster in instancesShutdown
  • lxd/db/instances: Removes ResetInstancesPowerState function
  • lxd/instances: Move shutdown timeout logic into per-instance go routine in instancesShutdown
  • lxd/instances: Reworks instancesShutdown to handle and log shutdown failures by forcefully stopping
  • lxd/instances: Updates instancesShutdown to accept a slice of instances
  • lxd/instances: Renames containerAutostartList to instanceAutostartList
  • lxd/instances: Renames instancesRestart to instancesStart
  • lxd/daemon: Updates init to use instancesStop and instancesStart with preloaded container list
  • lxd/daemon: Updates Ready to use updated instancesStart
  • lxd/daemon: Updates Stop to load instances once
  • lxd/daemon: Updates numRunningInstances to accept a list of instances to check
  • shared/osarch/architectures: Use ARCH_UNKNOWN rather than 0 in ArchitectureId
  • lxd/db/instances: Removes UpdateInstancePowerState function
  • lxd/instance/drivers/driver/common: Adds recordLastState function
  • lxd/instance/drivers: Use d.VolatileSet in onStop hook to record last power state
  • lxd/instance/drivers: d.recordLastState usage
  • lxd/instances/drivers: Call d.UpdateBackupFile just before starting instance process
  • lxd/daemon: Close global database after query failure in Stop
  • lxd/daemon: Use consistent terminology of global rather than remote database in Stop
  • lxd/api/internal: Update internalContainerHookLoadFromReference to try and load instance from backup if DB not available
  • lxd/instance/drivers/driver/qemu: Update getMonitorEventHandler to try and load instance from backup if DB not available
  • lxd/storage/drivers/driver/zfs/volumes: Log dev path in UnmountVolume
  • lxd/migration: Update protobuf config
  • lxd/migration: Update generated protobuf
  • gomod: Update dependencies
  • doc/network: Avoid referring to releases by name
  • lxd/apparmor: Allow remount using noatime
  • lxd/apparmor: remove mount options alternations
  • lxd/apparmor: remove another mount options alternations
  • lxd/apparmor: remove spaces between mount options for consistency
  • lxd/apparmor: remove duplicated mount rules (ro,remount,bind)
  • lxd/api/cluster: Fail on no leader in internalClusterPostHandover
  • lxd/instance: Fix image download race condition in instanceCreateFromImage
  • lxd/networks: Report uplink networks in leases
  • lxd/api/cluster: Fail on no leader in handoverMemberRole
  • lxd/cluster/gateway: Log partial and initial heartbeat as info
  • test: Adds better logging and removes handover sleeps in test_clustering_handover
  • lxd/cluster/heartbeat: No need to log heartbeat restart
  • test: Don’t use pid files in test_clustering_shutdown_nodes
  • test: Actually ensure cluster DB isn’t reachable after its lost quorum in test_clustering_shutdown_nodes
  • test: Use timeouts in kill_lxd
  • lxd/storage/drivers/driver/zfs/volumes: Use normal mount rather than zfs mount
  • tests: Unify how the instance’s PID is looked up
  • tests: add missing --force-local to lxc stop
  • tests: use CSV format and column filtering where applicable
  • tests: use CSV format and column filtering to find the name of the newly created instance
  • tests: use grep -F when the match pattern contains regex/wildcard
  • tests: replace grep | cut by awk
  • lxd/instance/operationlock: Adds TimeoutSeconds constant
  • lxd/instance/drivers/driver/common: Error quoting in onStopOperationSetup
  • lxd/instance/drivers/driver/qemu: Updates onStop to be more like lxc driver
  • test/suites: Always provide project arg in volume test
  • lxd/sys/os: reorder kernel features
  • os: add separate entries for pure core scheduling kernel feature and container support
  • lxd: support core scheduling for virtual machines
  • test/suites: Fix cephfs backup test
  • lxd/instance/drivers/driver/qemu: Improve comments in Shutdown
  • lxd/instance/operationlock: Add ErrNonReusuableSucceeded error and Action type and action constants
  • lxd/instance/operationlock: Reworks Create to use Action type
  • lxd/instance/operationlock: Reworks CreateWaitGet
  • lxd/instance/drivers/driver/common: operationlock.Action usage
  • lxd/instance/drivers/driver/lxc: operationlock.CreateWaitGet and operationlock.Create usage
  • lxd/instance/drivers/driver/qemu: operationlock.CreateWaitGet and operationlock.Create usage
  • lxd/instance/drivers/driver/qemu: Add comment in Stop about operation lock
  • lxd/instance/drivers/driver/qemu: Keep operation alive in Shutdown
  • lxd/instance/drivers/driver/lxc: Keep operation alive in Shutdown
  • lxd/network: Move Leases to network package
  • lxd: support core scheduling for container even without LXC library support
  • lxd/instance/lxc: Properly report mapped memory
  • lxd/daemon: Updates NodeRefreshTask to accept an isLeader and unavailableMembers argument
  • lxd/api/cluster: Improves logging in internalClusterPostHandover
  • lxd/cluster/gateway: Adds shutdownCtx to NewGateway and return 503 in heartbeat if shutting down
  • lxd/storage/drivers/driver/zfs/volumes: Set mountpoint=none for filesystem volumes
  • lxd/storage/drivers/driver/zfs/patches: Update patches to set mountpoint=none
  • lxd/cluster/gateway: Adds HeartbeatHook type
  • lxd/cluster/gateway: Reject heartbeat if shutting down
  • lxd/cluster/gateway: Rework HandlerFuncs heartbeat handling
  • lxd/cluster/heartbeat: Update heartbeatRestart to return bool if heartbeat restarted
  • lxd/cluster/heartbeat: Pass non-updated heartbeat members as unavailable to heartbeat refresh task in heartbeat
  • lxd/cluster/membership: Add logging to notifyNodesUpdate
  • lxd/cluster/membership: Improve logging in Assign
  • lxd/cluster/membership: Adds unavailableMembers support to Rebalance and newRolesChanges
  • lxd/api/cluster: Adds unavailableMembers support to rebalanceMemberRoles
  • lxd/api/cluster: Improve logging in rebalanceMemberRoles
  • lxd/cluster/gateway/test: cluster.NewGateway usage
  • lxd/cluster/gateway: Remove unnecessary logging
  • lxd/cluster/gateway: Don’t stop enrichhing raft nodes if one member name not found in currentRaftNodes
  • test: Reduce sleeps and offline threshold in clustering tests to speed them up
  • lxd/db/networks: Reworks GetNetworkInAnyState and its ilk to split the functionality out into separate functions
  • shared/api/error: Improve argument name in StatusErrorMatch
  • shared/api/error: Adds StatusErrorCheck helper function
  • lxd/networks: Use api.StatusErrorCheck to check for not found error from d.cluster.GetNetworkInAnyState
  • lxd/instance/drivers/driver/lxc: Fix restart locking
  • lxd/network/driver/ovn: Fix comment on getLoadBalancerName
  • lxd/cluster/heartbeat: Use api.StatusErrorf in error returned from HeartbeatNode
  • test: Improve test_clustering_remove_raft_node reliability
  • lxd/network/ovn: Add support for leases
  • api: Add image_source_project extension
  • shared/api: Add Project to ImagesPostSource
  • lxd/storage/drivers/volume: Adds IsCustomBlock function
  • lxd/storage/drivers/generic/vfs: Fixes regression in genericVFSBackupUnpack for VM config volume import
  • lxd/network/openvswitch/ovn: Adds OVNRouterRoute type
  • lxd/network/openvswitch/ovn: Updates LogicalRouterRouteAdd to accept multiple OVNRouterRoute args
  • lxd/network/openvswitch/ovn: Harmonise naming conventions in LogicalRouterRouteDelete with OVNRouterRoute type
  • lxd/network/network/utils: Removes unnecessary DB lookup via NICType function in isInUseByDevice
  • lxd/network/network/utils: Updates UsedBy to use usedByInstanceDevices
  • lxd/network/network/utils: Updates usedByInstanceDevices to use updated isInUseByDevice
  • lxd/network/network/utils: Updates isInUseByProfile to use updated isInUseByDevice
  • lxd/network/network/utils: Renames isInUseByProfile to usedByProfileDevices
  • lxd/network/openvswitch/ovn: Update LogicalRouterRouteDelete to accept net.IPNet rather than pointer
  • lxd/network/driver/ovn: Remove default routes and re-add as needed in setup
  • lxd/network/driver/ovn: Update InstanceDevicePortSetup to use static route port hints
  • lxd/network/driver/ovn: InstanceDevicePortDelete updated to handle non-pointer IPNets
  • client: Support source project in image copies
  • lxd/images: properly return project name in error
  • lxd: Support source project in image copies
  • lxc/image: Support source project in image copies
  • doc/rest-api: Refresh swagger YAML
  • lxd/db/config: rename UpdateConfig to UpdateClusterConfig
  • lxd/db/generate/lex/form: smarter pluralize function
  • lxd/db/generate/db/method: fix stmt type for generating URIs
  • lxd/db/generate/db/parse: check stmt and method for omitting fields
  • lxd/db/generate/db/mapping: pass table name to FieldColumnName
  • lxd/db/generate/db/stmt: pass variable name to register function
  • lxd/db/operations.mapper: remove ProjectID omission from operations
  • lxd/instance/drivers/qmp/monitor: Update run to accept an interace{} for args and JSON encode internally
  • lxd/instance/drivers/qmp/monitor: m.run usage
  • lxd/instance/drivers/qmp/commands: m.run usage
  • lxd/instance/drivers/qemu: Adds workaround for QEMU 6.x regression in handling memory object host-nodes setting
  • lxd/db/cluster/update: updateFromV50
  • lxd/db/cluster/schema: Update schema
  • lxd/db/node: add Config field to NodeInfo
  • lxd/db/node: populate config on Get
  • lxd/db/node: add UpdateNodeConfig method
  • lxd/db/node: add default config on Create
  • lxd/db/node: bootstrap nodes with BootstrapNode
  • lxd/db/node: skip database-standby in UpdateRoles
  • lxd/db/node: skip nodes with manual scheduler
  • lxd/device: Make sure vfio-pci is loaded
  • lxc/cluster: Add --yes to remove
  • tests: Update for change to cluster remove
  • shared/version/api: add clustering_config
  • shared/api/cluster: add Config to ClusterMemberPut
  • lxd/api/cluster: use updateClusterNode for PUT/PATCH
  • lxd/api/cluster: validate and update config on PUT/PATCH
  • lxc/cluster: add cluster member get/set/unset
  • doc/rest-api: Refresh swagger YAML
  • i18n: Update translation templates
  • doc/clustering: add clustering configuration docs
  • test/suites/clustering: add clustering_autotarget
  • lxd/db/config.mapper: placeholder config methods
  • lxd: Add fsmonitor package
  • lxd/state: Add DevMonitor to state
  • lxd: Initialize DevMonitor in daemon
  • lxd/device: Switch to DevMonitor
  • lxd/device: Check prefix path in source
  • lxd/device: Remove old inotify code
  • test/suites: Add fsmonitor to static analysis
  • lxd/db/networks: Fixes getStoragePool to support NULL description fields
  • lxd/api/cluster: fix comment on clusterGet clusterPut
  • lxd/device/nic/ovn: Improve error in Start
  • lxd/network/acl/acl/ovn: Adds OVNIntSwitchPortGroupAddressSetPrefix function
  • lxd/network/openvswitch/ovn: Adds address management functions
  • lxd/network/openvswitch/ovn: Adds router policy management function
  • lxd/network/driver/ovn: Move logical switch creation after internal network IP validation
  • lxd/network/driver/ovn: Add address set that represents internal switch subnets and NIC routes
  • lxd/network/driver/ovn: Adds instanceNICGetRoutes function
  • test: Set LXD_DEVMONITOR_DIR
  • doc/environment: Add LXD_DEVMONITOR_DIR
  • lxd/network/driver/ovn: Reworks Update to populate active NIC routes into internal switch’s address set
  • api: Adds network_peer extension
  • lxd/network/acl/driver/common: ValidName usage
  • lxd/network/acl/acl/validation: Adds ValidName function
  • lxd/network/acl/driver/common: Adds ReservedNetworkSubects slice
  • shared/api/network/peer: Adds network peering API types
  • lxd/db/cluster: Adds networks_peers and networks_peers_config tables
  • lxd/db/network/peers: Peer management functions
  • doc/rest-api: Refresh swagger YAML
  • lxd/lifecycle/network/peer: Peer lifecycle type
  • lxd/network/network/utils: Updates UsedBy to check if the network has any created peers
  • lxd/networks: network.UsedBy usage
  • lxd/network/driver/common: UsedBy usage
  • lxd/network: Adds peer management function definitions and common not implemented implementation
  • lxd/network/driver/common: Adds Peering feature indicator
  • lxd/network/driver/common: Adds common peer validation function
  • lxd/network/openvswitch/ovn: Adds peer management functions
  • lxd/network/driver/ovn: Adds peering management functions
  • lxd/network/peer: Network peer API endpoints
  • client/interfaces: Adds network peer management function definitions
  • client/lxd/network/peer: Adds network peer management client functions
  • lxc/network/peer: Adds network peer CLI commands
  • i18n: Update translation templates
  • doc: Adds network peers documentation
  • lxd/network/driver/bridge: Fix leases
  • lxd-p2c: Allow passing existing certificate
  • client/connection: Typo in ConnectLXD
  • lxd/instance/post: Improve error in instancePostClusteringMigrate
  • lxd/move: Adds stateful migration support for cross-pool instance migrations
  • lxd/move: Consistent comment line endings in moveInstancePool
  • lxd/instance/post: Refactor how backward compat handling of Live field in instancePost
  • lxd/instance/post: Error quoting
  • lxd/instance/post: Comment typo in instancePostPoolMigration
  • lxd/instance/post: Adds stateful migration support to instancePostPoolMigration
  • i18n: Update translation templates
  • process_utils: move wait_for_pid_status_nointr() to common code
  • lxd: consolidate functions into common header
  • lxd: make C functions static
  • lxd: add config.h for common options
  • doc: Drop part about building LXC
  • doc: Add bind9-dnsutils to testsuite requirements
  • lxd: use argument struct for manip_file_in_ns()
  • forkfile: small coding style fixes
  • forkfile: decrease indendation level
  • lxd/fsmonitor/drivers: Add missing FAN_MARK_FILESYSTEM
  • doc/enviroment: Clarify that LXD_DEVMONITOR_DIR is for testing
  • lxd/fsmonitor: Ensure path is a mountpoint
  • lxd/instance/instance: Renames IsMigratable to CanMigrate
  • lxd/api/cluster: inst.CanMigrate usage
  • lxd/move: Add stateful move support to moveClusterInstance
  • lxd/instance/post: Adds stateful migration support to instancePostClusteringMigrate
  • i18n: Update translation templates
  • test: Use mountpoint for test devices
  • lxd/db/node: don’t autofill node config
  • lxd/api/cluster: make scheduler.instance optional
  • lxd/db/query/transaction: Adds 10s timeout to Transaction
  • lxd/cluster/gateway: Add 30s idle timeout in dqliteProxy
  • lxd/cluster/gateway: Close remote connection in dqliteProxy
  • lxd/cluster/events: Disconnect event listeners for offline members in eventsUpdateListeners
  • lxd/events/events: Rework events.Listener to use websocket heartbeats
  • shared/instance: Add linux.sysctl.*
  • lxd: Support for linux.sysctl.* configuration keys
  • doc: Add linux.sysctl.*
  • api: Add linux_sysctl extension
  • lxd/device/nic/bridged: Prevent specifying ipv{n}.address when NIC is connected to unmanaged parent bridge
  • test: Add test for bridged NIC not able to specify static IPs when using unmanaged parent bridge
  • lxd/events/events: Moves blocking reader into heartbeat function
  • lxd: Removes blocking reader from event client
  • lxd-agent: Removes blocking reader from event client
  • lxd/fsmonitor/drivers: Log warning instead of failing
  • lxd/cluster/gateway: Removes dqliteProxy idle timeout
  • lxd/cluster/gateway: Update logging in dqliteProxy
  • lxd/cluster/gateway: Set 30s TCP_USER_TIMEOUT and keep alive timers in dqliteProxy
  • lxd/util/net: Adds SetTCPUserTimeout function
  • lxd/db/db: retry cluster transactions once if context deadline exceeded
  • seccomp: Pass the caller TGID to pidfd_open instead of TID
  • lxd/instance/drivers: Check swap for metrics
  • seccomp: verify retrieved fds when
  • shared/validate: Add IsListOf
  • doc: Move metrics under API
  • api: network_dns extension
  • doc/projects: Add restricted.networks.zones
  • lxd/projects: Add restricted.networks.zones
  • doc/networks: Add dns.zone
  • lxd/network: Add dns.zone
  • doc/server: Add core.dns_address
  • lxd/node: Add core.dns_address
  • shared: Add network zone API
  • client: Rename lxd_network_forwards for consistency
  • client: Add network zone functions
  • lxd/db: Add networks_zones schema
  • lxd/db: Add network zones helpers
  • lxd/network/zone: Initial package
  • lxd/lifecycle: Add network zones
  • lxd: Add network zone API
  • doc/rest-api: Refresh swagger YAML
  • lxc/network: Add zone sub-command
  • i18n: Update translation templates
  • lxd/dns: Add DNS server
  • lxd/daemon: Integrate DNS server
  • tests: Add DNS zone tests
  • doc: Add network zones documentation
  • lxd/instance/qemu: Fix host-nodes on multi-node
  • lxd/instance/drivers: Use existing DeviceTotalMemory
  • lxd/instance/drivers: Check memory limit value for metrics
  • lxd/migrate: Fix deadlock in sendControl
  • lxd/migrate: Time out when waiting for connections
  • lxd/storage/backend/lxd: Clarify errors in CreateInstanceFromCopy and CreateInstanceFromMigration
  • lxd/instances/post: Wrap error in createFromMigration
  • lxd/instances/post: Update comment in createFromMigration
  • lxd/instance/drivers/driver/qemu: Implements Migrate to support stateful start after migration
  • lxd/migrate/instance: Update new migration source and sink functions to detect container instance type for CRIU
  • lxd/migrate/instance: Move predump setup logic inside CRIU section
  • lxd/migrate/instance: Keep vol source arg setup logic together
  • lxd/migrate/instance: Make CRIU logic conditional on container instance type
  • lxd/migrate/instance: Statefully stop running VM for live migration in migrationSourceWs
  • lxd/migrate/instance: Code style tweak in migrationSink
  • lxd/migrate/instance: Update migrationSink Do to support VM stateful migration
  • lxd/migrate: Update critieria for live migration to detect container instance type
  • lxc/copy: Add comment clarifying post-migration start up logic in copyInstance
  • lxd/instance/post: Simplify arguments to instance migration functions
  • lxd: migrateInstance usage
  • lxd/instance/post: Remove container references in migrateInstance
  • lxd/instance/post: Improve comment in instancePostClusteringMigrateWithCeph
  • lxd/storage/drivers/driver/ceph/volumes: Improve comment in MigrateVolume
  • lxd/instance/post: Avoid loading storage pool twice in instancePostClusteringMigrateWithCeph
  • lxd/instance/post: Remove out of date comment from instancePostClusteringMigrateWithCeph
  • lxd/instance/post: Remove container reference in error in instancePostClusteringMigrateWithCeph
  • lxd/instance/post: Remove trailing newline in instancePostClusteringMigrateWithCeph
  • lxd/instance/post: Use http.StatusOK in instancePostClusteringMigrateWithCeph
  • lxd/instance/post: Switch to api.NewURL() in instancePostClusteringMigrateWithCeph
  • lxd/db/instances: Adds volumeType argument to UpdateInstanceNode
  • lxd/instance/post: Replace node in errors with member
  • lxd/instance/post: Rework instancePostClusteringMigrateWithCeph to support VM stateful migration
  • lxd/instance/post: Update instancePostCreateInstanceMountPoint to accept an instance
  • lxd/instance/post: Don’t allow migration to same location of instance in migrateInstance
  • lxd/instances/post: Replace node with member in error in clusterCopyContainerInternal
  • lxd/storage/drivers/driver/ceph/volumes: Fix MigrateVolume to support VM cluster migration
  • lxd/instance/post: Update instancePostClusteringMigrate with support for starting instance during stateless migration
  • test: Update teardown_clustering_netns to succeed if process has already gone
  • lxd/device/nic/bridged: Allow static IP on unmanaged bridge when MAAS subnet specified
  • lxd/device/nic/ovn: Remove setting accept_ra sysctl unnecessarily as disable_ipv6 is set
  • Replace ‘which’ with ‘command -v’
  • Use which to resolve lxc’s path without resolving the shell function of the same name
  • lxc: better advertise support for VMs
  • i18n: Update translation templates
  • lxd/device/device/utils/network: Renames networkSnapshotPhysicalNic to networkSnapshotPhysicalNIC
  • lxd/device: networkSnapshotPhysicalNIC usage
  • lxd/device/device/utils/network: Renames networkRestorePhysicalNic to networkRestorePhysicalNIC
  • lxd/device: networkRestorePhysicalNIC usage
  • lxd/device/nic: Split up SR-IOV functions
  • lxd/device: Records SRIOV VF parent device in last_state.vf.parent
  • shared/instance: Add .last_state.vf.parent support to ConfigKeyChecker
  • lxd/device/device/utils/network: Adds useSpoofCheck arg to networkSRIOVSetupVF and networkSRIOVRestoreVF
  • lxd/device/device/utils/network: Add support for last_state.vf.parent in networkSRIOVRestoreVF
  • lxd/device/nic/sriov: useSpoofCheck argument usage
  • lxd/network/network/utils/sriov: Rename sriovFindFreeVirtualFunctionMutex to SRIOVVirtualFunctionMutex
  • lxd/network/network/utils/sriov: Remove use of lock from SRIOVFindFreeVirtualFunction
  • lxd/device/nic/sriov: Use network.SRIOVVirtualFunctionMutex.Lock
  • api: Add ovn_nic_acceleration extension
  • doc/instances: Add network acceleration config
  • lxd/network/openvswitch: Add HardwareOffloadingEnabled
  • lxd/network: Adds SRIOVFindFreeVFAndRepresentor
  • lxd/device/nic_ovn: Add SR-IOV support
  • lxd/device/nic/ovn: Remove incorrect call to networkSetupHostVethLimits
  • lxd/cluster: Fix forceful stop on restore
  • bash-completion: Add restore snapshots
  • lxd/operations: Updates waitForOperations to accept consoleShutdownTimeout arg
  • lxd/daemon: Load shutdown timeout from DB in Stop()
  • lxd/daemon: Only call waitForOperations during Stop if cluster DB is available
  • lxd/api/internal: Wait for daemon to finish starting up before shutting down in internalShutdown
  • forkuevent: minor fixes
  • forkuevent: fix “–” placement
  • lxd/instances/qemu: Enable topoext on x86_64 with SMT
  • lxc/import: read from stdin
  • doc/instances: Explain CPU topology in VMs
  • cleanup: remove subtest
  • Remove the last ‘which’ occurence.
  • lxc/file: Don’t use HostPath on temp files
  • lxd/cluster: Log leader address on failure
  • lxd/api/cluster: Sleep for 100ms to allow http.Flush to render
  • lxd/fsmonitor: Don’t crash on Walk errors
  • lxd/instance/lxc: Only mount for file operations if stopped
  • lxd/instance_exec: Improve and expand default PATH setting
  • lxd/network/driver/ovn: Comment typo
  • lxd/network/driver/ovn: Add external inbound peer address spoof protection to logicalRouterPolicySetup
  • lxd/network/driver/ovn: Update PeerDelete to remove security policy rules from local and target routers
  • lxd/network/acl/driver/common: Small optimisation in validateRule
  • lxd/network/acl/driver/common: Allow named peers starting with @ in validateRuleSubjects
  • lxd/db/network/peers: Adds GetNetworkPeersTargetNetworkIDs function
  • lxd/network/acl/acl/ovn: Adds support for peer subjects using @<network_name>/<peer_name> in OVNEnsureACLs
  • doc/network-acls: Adds concept of peer network subject selectors
  • lxd/instance/drivers/driver/lxc: Adds release function to reduce repetition of liblxc clearing logic
  • lxd/instance/drivers/driver/lxc: Release liblxc cache when stopping or shutting down
  • lxd/network/network/interface: Adds PeerUsedBy function
  • shared/api/network/peer: Adds UsedBy field to NetworkPeer
  • lxd/network/peer: Populate UsedBy field
  • lxd/network/driver/common: Adds peer used by functions
  • lxd/network/driver/ovn: Prevent peer from being deleted if in use
  • doc/rest-api: Refresh swagger YAML
  • gomod: Update dependencies
  • i18n: Update translations from weblate
  • lxd: Move to new protobuf when possible
  • gomod: Pin x/net to version supporting 1.13
  • lxd/storage/btrfs: Support 5.14.2
  • lxd/db: Use upstream context package
  • gomod: Update dependencies
  • gomod: Downgrade dqlite to 1.10.0
  • lxd/storage/ceph: Properly support pre-existing empty pools

Try it for yourself

This new LXD release is already available for you to try on our demo service.

Downloads

The release tarballs can be found on our download page.

Binary builds are also available for:

  • Linux: snap install lxd
  • MacOS: brew install lxc
  • Windows: choco install lxc
5 Likes

This is currently available in the candidate snap channel for those using the snap.

A live stream going over each of the new features is scheduled for 2pm US eastern on Monday:

Very nice a lot of good features , the highest of all releases !
Gonna upgrade it.

What a huge release! The cluster member targeting is quite simple: either automatic placement or manual. That’s a step in the right direction.

The default Kubernetes approach is to schedule based on a lot of factors, mostly around the available resources (CPU, memory). I hardly ever do anything but let K8s just handle it.

With LXD I think it would be good to develop an explicit metadata-based approach first. Something like K8s node affinity, but with an LXD flavor. Imagine dividing up your cluster like this:

  • edge proxy nodes
  • database nodes
  • App 1
  • App 2
  • general compute (the rest of them)

Capacity planning on K8s is actually not as easy as it claims to be. The promise of having one giant blob of Compute and Disk that the K8s scheduler efficiently slices up has not panned out for me. It’s actually pretty hard to know what CPU and Memory reservations you want. It’s hard to balance all the concerns of redundancy, efficient capacity use, and performance. (PS I do not run bare metal, maybe k8s is more helpful there.)

And I still do not recommend stateful workloads on K8s.

But if LXD could give a little bit more help in categorizing nodes and placing instances based on those categories, that’s enough for me. I guess this is just a long way of saying that I’d prefer setting key-value pairs on nodes according to their role in my infrastructure. A sophisticated placement algorithm based on CPU, Memory, and a dozen other things is less interesting to me. If I really am over/under provisioned, I can scale the underlying VM.

We have work on cluster member groups planned for the next few months which will let you do --target @something and will have LXD pick the least busy of the servers in that group.

This will be combined with a change to that new cluster config key to have have some servers only take in instances when targeted through a group.

1 Like

Submitted to openSUSE. :smiley:

1 Like

Thanks!