Changelog
0.17.0 (2024-04-17)
Fix
Added mising network param for external network or nova pool name. [cloudguruab]
Patch to storage migrations feature and docs. [cloudguruab]
0.16.0 (2023-09-13)
New
Documentation for new storage migration modes. [cloudguruab]
Strg migration. [cloudguruab]
Fix
Defaulting to release zed for devstack. [cloudguruab]
Revert install guestfs-tools on CentOS 9 (#620) (#627) [Adrian Brown]
Set default username to centos (#625) [Nicholas]
change default username for CentOS 9 cloud image
Install guestfs-tools on CentOS 9 (#620) (#627) [Adrian Brown, Nicholas Natale]
Change install content task on conversion hosts to install guestfs-tools for CentOS 9 systems (guestfs-tools provides virt-sparsify)
Other
Merge branch ‘main’ of https://github.com/cloudguruab/os-migrate. [cloudguruab]
Merge branch ‘main’ of https://github.com/cloudguruab/os-migrate. [cloudguruab]
Merge branch ‘main’ of https://github.com/cloudguruab/os-migrate. [cloudguruab]
0.15.1 (2023-06-06)
New
Stable 0.15.0. [cloudguruab]
Mac preservation enhancement for server_port instance (#603) [Adrian Brown]
new: mac preservation enhancement for server_port instance
dev: fix to cachecontrols error
Fix
Update to CentOS-8 qcow image version for e2e devel env setup. [cloudguruab]
0.15.0 (2023-04-20)
New
Port creation via neutron api. [cloudguruab]
Neutron port creation for workloads. [cloudguruab]
Patch to user project role assignment bug. [cloudguruab]
Process for cleaning stuck lockfiles in use_lock (#590) [Adrian Brown]
Validate changelog using lint-commit-messages target. [Jiří Stránský]
Fix
Support cacert in clouds.yaml template. [Pierre Riteau]
0.14.0 (2022-09-22)
New
Conversion host updates. [cloudguruab]
Update to auth on missed ansible parts. [cloudguruab]
Use clouds.yml for auth config on project resources (#563) [Adrian Brown]
Use cloud config for auth on workload resources (#569) [Adrian Brown]
Use clouds config for auth on users resources (#568) [Adrian Brown]
Use cloud config for auth on subnet resource (#566) [Adrian Brown]
Use cloud config for auth on security group resources (#565) [Adrian Brown]
Use clouds.yml config for auth on routers resource (#564) [Adrian Brown]
Use clouds.yml config for auth on flavor resources (#562) [Adrian Brown]
Use clouds.yml config for auth on image resources (#561) [Adrian Brown]
Use clouds.yaml in keypair migration. [cloudguruab]
Use clouds.yaml auth approach in network import/export. [cloudguruab]
Use clouds.yaml auth approach in network import/export (#559) [Adrian Brown]
Write auth config to clouds.yaml (#552) [Adrian Brown]
Sort exported server ports as they appear in Nova API. [Jiri Stransky]
We export server port info from Neutron API, because it gives more information about the port. However, querying ports from Neutron seems to give a somewhat random order of the ports.
Now we still export the same Neutron info about ports, but we sort them based on the order in which the VM IP addresses appear in OpenStack SDK’s ‘compute.get_server’ return value.
This approach cannot guarantee that the original order of NICs is preserved during migration (Nova does not guarantee NIC ordering [1]), but it makes it more likely that the NIC order will be preserved during migration.
Skip serialization if dns_domain is false. [cloudguruab]
Import user project role assignment. [cloudguruab]
User project role assignments export. [Adrian Brown]
Introduced validate_resource_files role (#532) [Adrian Brown]
Changes
Documentation for keypair migration pitfalls and solutions. [cloudguruab]
Fix
Update to resource type. [cloudguruab]
Export of subnets with multiple allocation pools. [Jiri Stransky]
Subnets with multiple allocation pools previously crashed on export due to an issue with parameter sorting. This is now fixed.
0.13.0 (2022-03-30)
New
Customize availbility zone for conversion hosts. [adrian]
It is now possible to customize conversion host availability zones via these new variables:
os_migrate_src_conversion_availability_zone os_migrate_dst_conversion_availability_zone
Ensure correct mode on conversion host ssh key. [Jiri Stransky]
Sometimes when using pre-existing key and mounting it into a container, the mode seems to be set wrong (reason not discovered). We now make sure that the conversion host key has mode ‘0600’ before using it, otherwise ssh would refuse to work with it anyway.
Allow using conversion hosts without floating IPs. [Jiri Stransky]
Conversion hosts now can be created without floating IPs. Whether FIPs should be managed or not is controlled via these new variables:
os_migrate_src_conversion_manage_fip os_migrate_dst_conversion_manage_fip
Having conversion hosts without FIPs is useful when attaching the hosts directly onto public networks. This use case is described in more detail in the docs (Variables Guide).
Customizing src/dst names of conversion host dependency resources. [Jiri Stransky]
Conversion host dependency resource names can now be customized differently per source/destination cloud. The relevant new variables are:
os_migrate_src_conversion_net_name os_migrate_dst_conversion_net_name os_migrate_src_conversion_subnet_name os_migrate_dst_conversion_subnet_name os_migrate_src_conversion_router_name os_migrate_dst_conversion_router_name os_migrate_src_conversion_secgroup_name os_migrate_dst_conversion_secgroup_name os_migrate_src_conversion_keypair_name os_migrate_dst_conversion_keypair_name
Allow using existing network with conversion hosts. [Jiri Stransky]
It is now possible to disable creation and deletion of private network+subnet+router for the conversion hosts via new variables:
os_migrate_src_conversion_manage_network os_migrate_dst_conversion_manage_network
The hosts will attach to pre-existing network defined by os_migrate_conversion_net_name.
Docs page with troubleshooting tips. [Jiri Stransky]
Changes
Improve error message on data version mismatch. [Jiri Stransky]
After upgrading without clearing old exported data, OS Migrate refuses to parse the old data files. The resulting error message was not clear enough about the likely root cause and solution. Path to the problematic file and extra hints are now added to the exception message, as this is a relatively common pitfall.
Fix
Replace akward openstacksdk version check. [Dmitriy Rabotyagov]
Using shell for getting pip version leads to several highly possible side effects. As example, you verify opensatcksdk persistance against ENV however following tasks will use ansible_python_interpreter that likely differ if virtualenv is used for ansible.
Another side effect is that openstacksdk could be installed in venv which is not respected by shell when you provide interpreter for host explicitly.
With this patch we replace shell with pip_package_info module that collects all installed packages against ansible_python_interpeter.
Enforce the executable argument when executing shell. [Carlos Camacho]
Currently the migrator host is not limited to be only CentOS, Fedora, or RHEL. In the case a user runs OS migrate from a Linux distribution which is not using Bash as the default terminal, options like pipefail might not be availabe (i.e. Ubuntu and ZSH) for those cases we need to make explicit that the shell we will use is Bash.
This commit enforce to use and set explicitly the executable when running ansible.builtin.shell.
Other
F/dev env setup improvements (#520) [Adrian Brown, Jiri Stransky]
New: Customize availbility zone for conversion hosts
It is now possible to customize conversion host availability zones via these new variables:
os_migrate_src_conversion_availability_zone os_migrate_dst_conversion_availability_zone
Dev: Don’t run lints with Ansible 2.9
This is to work around an issue with ansible-lint:
https://github.com/ansible-community/ansible-lint-action/issues/41
Dev: Refactor argument ordering to satisfy pylint
We shouldn’t change the ordering of arguments of create_or_update when subclassing Resource, we should just add a new parameter at the end.
dev: allow vagrant on headless servers
dev: update vagrant to fedora35
horizon was broken on fedora35, dont need so it is disabled.
dev: documentation update for variables in e2e setup
Docs: Fix link on install-from-galaxy which address to install-from- source. [Juan Badia Payno]
0.12.0 (2021-11-19)
New
Allow creating conversion hosts as boot-from-volume servers. [Jiri Stransky]
The operator can now control if conversion hosts are backed by a local Nova disk (still the default), or if they are created as boot-from-volume. The parameters which control the behavior are:
os_migrate_src_conversion_host_boot_from_volume os_migrate_dst_conversion_host_boot_from_volume
Include an openstacksdk check in the preludes roles. [Carlos Camacho]
This commit adds a package version check in the prelude_src and prelude_dst roles so we verify the user has a known working openstacksdk version before running the migrations.
Fix
Compatibility with PyYAML 6.x, safe YAML. [Jiri Stransky]
When PyYAML 6.x is used as a dependency, the load_resources_file function crashes with:
TypeError: load() missing 1 required positional argument: ‘Loader’
The best fix here should be switching to the YAML’s safe_load and safe_dump functions. These should be preferred because we do not need to serialize arbitrary Python objects, we serialize basic data types, dicts and lists.
0.11.1 (2021-10-19)
Fix
Compatibility with PyYAML 6.x, safe YAML. [Jiri Stransky]
When PyYAML 6.x is used as a dependency, the load_resources_file function crashes with:
TypeError: load() missing 1 required positional argument: ‘Loader’
The best fix here should be switching to the YAML’s safe_load and safe_dump functions. These should be preferred because we do not need to serialize arbitrary Python objects, we serialize basic data types, dicts and lists.
(cherry picked from commit 40982ecc8ad7b32f1b906d67dd998f884bae9de0)
0.11.0 (2021-10-08)
New
Allow using pre-existing floating IPs for migrated servers. [Jiri Stransky]
Two new floating_ip_mode migration parameter values are introduced: ‘new’ and ‘existing’. The modes now behave this way:
‘skip’ - Do not create any floating IPs.
‘new’ - Create a new floating IP (auto-assigned address).
‘existing’ - Assume the floating IP address specified in the workload serialization is already assigned to the destination project, but not attached. Attach this floating IP. If this is not possible for some reason, fail.
‘auto’ - Attempt the ‘existing’ method of floating IP assignment, but instead of failing, fall back to the ‘new’ method.
Allow deleting or detaching the conversion hosts FIPs. [Carlos Camacho]
This commit enables users to detach or demove floating IPs when deleting the conversion hosts.
Changes
Install conversion host content in parallel again. [Jiri Stransky]
Content installation went from parallelized to non-parallelized when we introduced pre/post-content hooks, so that we could feed different variable values per host.
This commit keeps the per-host variables by setting them via the set_fact module prior to executing the conversion_host_content role. The role itself then does not need per-host variables and runs in parallel on both conversion hosts. This can speed up the conversion host deployment playbook noticeably.
Fix
Remove unnecessary 10s delay in conversion host discovery. [Jiri Stransky]
Probably sleep (pause between attempts) was meant instead of delay (pause before the first attempt). Use a 5 second sleep.
Only perform keypair lookup in dst if server has keypair. [Philip Roche]
When validating workload migration we should only validate if the keypair specified in the workload/vm is present in the destination environment if the workload/vm has a keypair associated with it.
Validate keypairs exist in destination cloud before migrating workload. [Philip Roche]
dst_prerequisites_errorsclass method is used to validate prerequisites before migrating workloads but the parent class dst_prerequisites_errors does not validate keypairs exist in destination cloud. To do this we do a lookup to see if keypair name exists and append to errors list if not
Other
Oth: Improve workload data flow diagram. [Jiri Stransky]
The data flow diagram in “OS Migrate Walkthrough” and “How It Works: Workload Migration” now has more clarity on how (meta)data travels during workload migration.
Docs: fix walkthrough links. [Carlos Camacho]
This commit fixes some wrong links in the docs
0.10.1 (2021-09-02)
Fix
Make explicit the conversion hosts inventory update. [Carlos Camacho]
0.10.0 (2021-09-01)
New
Add a floating IP after creating the server instance. [Carlos Camacho]
This commit disables the auto floating IP allocation when the conversion host is created and explicitly assigns the floating IP once the instance is running. Also allows to specify custom floating IPs.
Conditionally stop VMs before migration. [Philip Roche]
Set os_migrate_workload_stop_before_migration to true to stop VMs before migration occurs.
Default is false but set to true for e2e tests
Allow disabling the RHSM tasks. [Carlos Camacho]
This commit allows to disable the execution of the RHSM tasks.
Allow configuring the conversion hosts names. [Carlos Camacho]
This commit allows to configure the conversion hosts names.
Allow running scripts in the conversion hosts. [Carlos Camacho]
This commit allows to run custom bash scripts in the conversion hosts before and after the content playbook.
Include additional RHSM repositories. [Carlos Camacho]
This commit allows to enable a list of new RHSM repositories in the conversion hosts.
Ability to specify boot volume params when source VM has no boot volume. [Jiri Stransky]
When the source VM does not have a boot volume, its boot disk should be copied (boot_disk_copy is true), then a boot volume will be created only in destination. Since there was no boot volume serialized in the workload YAML, we need an alternative way to allow editing volume creation parameters. New boot_volume_params migration parameter is introduced for this purpose.
Customizable attached volume properties in workload migration. [Jiri Stransky]
Attached volumes are now exported in the ‘params’ section of workload serialization, as nested resources. This allows editing these desired properties of attached volumes that should be created in the destination cloud: name, description, availability zone, volume type.
Allow ssh access to conversion host using a default password. [Carlos Camacho]
This commit allows to configure a default password in both centos and cloud-user users, this can help connecting directly to the hosts with a default password disabled by default.
Added os_migrate_conversion_host_ssh_user_enable_password_access disabled by default.
Fix
For resources that have name, reject migration of resources with empty name. [Philip Roche]
For resources which have a name property validate that it is non empty.
This fixes an issue where resources with empty names are not included in the check for uniqueness/idempotence. By checking for empty names we avoid this issue.
In rhsm auto_attach once defaulted will always collide with activationkey. [Carlos Camacho]
Once a variable is defaulted to an existing value it can not be omited, auto_attach is defaulted to true so if activationkey it can not be put back to omit.
omit is a special variable for omitting an individual argument to a module with only that value, and once defined it can not be omited again.
Glance API attribute ‘stores’ is read-only now (#443) [Philip Roche]
Fix: Glance API attribute ‘stores’ is read-only now
Due to bug https://bugs.launchpad.net/glance/+bug/1889676 glance now sets stores as a read only property. As such we should remove it from any create image calls.
See https://github.com/openstack/glance/commit/e1f0e94b90b162b1cdca5bdd7af3c09a90ca80ac#diff-3a53e94e6a14f08e4e3305f4e4388819cef4f35a9277afeab2f07bc2664489d6 for the Glance API change and also see conversation on terraform-provider-openstack repo @ https://github.com/terraform-provider-openstack/terraform-provider-openstack/issues/1091#issuecomment-806003499 for some related context.
This bug was found when testing e2e tests using an infrared deployed Openstack 16.1.
Other
Oth: User documentation on the inner workings of workload migration. [Jiri Stransky]
Docs: Add password-based authentication variable docs. [Carlos Camacho]
This commit adds the information for the ssh access using password in the conversion hosts.
Oth: General usage notes and TLS variables docs. [Jiri Stransky]
The documentation is extended with a separate page for general usage notes, and info about API TLS-related variables is added to the variables guide.
Oth: Add upgrade documentation. [Jiri Stransky]
Update comment in image.py. [Yaniv Kaul]
Fixes issue #437
0.9.0 (2021-07-08)
New
Import of users’ Nova keypairs as admin. [Jiri Stransky]
Import of Nova keypairs can now be done for other users as well, provided that the authenticated user running the playbook is authorized to read and create keys for other users (an admin user is typically authorized to do that). The playbook to perform this import is called import_users_keypairs.yml.
Both os_migrate_users_filter and os_migrate_keypairs_filter variables are respected when iterating over users and their keypairs.
Export of users’ Nova keypairs as admin. [Jiri Stransky]
Export of Nova keypairs can now be done for other users as well, provided that the authenticated user running the playbook is authorized to list users and their keys and read those keys (an admin user is typically authorized to do that). The playbook to perform this export is called export_users_keypairs.yml.
Both os_migrate_users_filter and os_migrate_keypairs_filter variables are respected when iterating over users and their keypairs.
Delete volumes when deleting the conversion hosts. [Carlos Camacho]
This commit adds the hability to remove automatically the volume created with the conversion host once the VM is destroyed.
Changes
Stop capping recommended OpenStack SDK on 0.52. [Jiri Stransky]
The security groups bug [1] introduced in OpenStack SDK 0.53 has been fixed, we start recommending and testing with the latest SDK again.
Increase the linelenght for the lint tests. [Carlos Camacho]
Lint tests are failing with the line lenght <=100. This commits increases the limit to 150.
Add retry logic to RHSM. [Carlos Camacho]
This commit adds the retry logic when using RHSM in the conversion hosts.
Add the conversion hosts to specific groups. [Carlos Camacho]
This commit allows having two additional groups for the conversion hosts. conversion_hosts_src and conversion_hosts_dst
Fix
Consider same-named keys for different users as valid for import. [Jiri Stransky]
The OS Migrate import validation used resource type + name to define identity. This works as expected when working in project scope. But as we added keypair migration by admin on behalf of other users, and keypairs are scoped to users, having multiple keypairs of the same name is valid as long as they are owned by different users. This commit refactors validation mechanisms to support custom notions of import identity per resource type, and keypairs use a tuple of resource type + resource name + user name + user domain.
Allow exporting same-named keypairs for different users. [Jiri Stransky]
Keypair IDs are typically the same as their names, they aren’t randomized UUIDs. We cannot rely on them for distinguishing uniqueneess on export and performing idempotency checks when exporting to a file which already contains some resources. The sameness check now uses ID + user_ref tuple, making it possible to export keys with the same ID values for mutliple users.
Conversion host network MTUs default to cloud defaults. [Jiri Stransky]
Previously we had conversion host network MTUs decided by variables os_migrate_src_release and os_migrate_dst_release and decided MTUs for conversion host network based on those. These were test-specific decisions leaking into production settings. We now default the MTU parameters to Ansible’s omit special value, which means the module responsible for creating the network will not specify any value in the network creation API request, and the MTU will be autoselected by the cloud.
If the defaults result in undesirable values, the MTUs can still be explicitly overriden via os_migrate_src_conversion_net_mtu and os_migrate_dst_conversion_net_mtu variables.
This fix also removes dependency of production code on os_migrate_src_release and os_migrate_dst_release variables, as currently OS Migrate does not alter behavior based on src/dst release.
Remove unnecessary delay in conversion host reachability check. [Jiri Stransky]
In conversion host reachability check, with the wait_for Ansible module we used delay of 10 seconds, which caused a delay before the first check, causing unnecessary slowdown (20 seconds in total) of the import_workloads playbook. What we probably meant to use was sleep, which sets the interval between checks. This is now fixed and the sleep property is set to 5 seconds.
Only use ‘type’ on keypairs when it is supported. [Jiri Stransky]
SDK should support ‘type’ on keypairs since 0.32 [1], however even with 0.36 this error still appears when testing against latest Devstack:
openstack.exceptions.BadRequestException: BadRequestException: 400: Client Error for url: http://192.168.122.93/compute/v2.1/os-keypairs, Invalid input for field/attribute keypair. Value: {‘type’: ‘ssh’, <REDACTED>, ‘name’: ‘osm_keypair’}. Additional properties are not allowed (‘type’ was unexpected).
This error disappears when testing with latest SDK. It’s not easy to track down what exact combinations of SDK/API are needed to allow for ‘type’ in keypair creation request. So we’ll first try with ‘type’, and if that request fails with error 400 (Bad Request), then we retry without ‘type’. If that request still fails, the keypair import module fails too.
[1] https://github.com/openstack/openstacksdk/commit/15baef656ac56421a71e691982a70b218110f18d
Flavor extra_specs import with OpenStack SDK older than 0.51. [Jiri Stransky]
The previous implementation of extra_specs import on flavors only worked with SDK >=0.51. A fallback implementation is now added for running with older SDKs. This fallback implementation can be dropped in the future after we drop support for SDK <0.51.
Handle when flavor extra_specs are reported as None from API. [Jiri Stransky]
Most recent APIs+SDKs always give empty hash for extra_specs. However, some older versions may also return extra_specs as None. This could break OS Migrate. The issue is now fixed by making sure we set internal variables to {} whenever some extra_specs parameter (either loaded from YAML or queried from SDK) is None.
Do not check for the subscribed_pool_ids. [Carlos Camacho]
Our internal RHSM server does not reply with subscribed_pool_ids, as it is always an empty list. This commit checks that the machines are correctly subscribed.
Increase timeouts when deployment takes more than default. [Carlos Camacho]
This commit allows to increase the hardwired default timeout of 600 seconds when deploying the conversion hosts.
In this case, when the overlay network provider takes longer to assign a floating IP, there might be needed to increase this default.
Export RouterInterface of HA and DVR routers. [Jiri Stransky]
Previously when exporting router interfaces, only ports with device_owner value of ‘network:router_interface’ were picked up. This would not export interfaces correctly for HA/DVR routers. We now also export ports with device_owner values of ‘network:ha_router_replicated_interface’ and ‘network:router_interface_distributed’ for HA/DVR routers.
Make sure the src/dst release is an integer. [Carlos Camacho]
This commit cast the os_migrate_src_release and os_migrate_dst_release variables to integers.
0.8.4 (2021-06-16)
Fix
Conversion host network MTUs default to cloud defaults. [Jiri Stransky]
Previously we had conversion host network MTUs decided by variables os_migrate_src_release and os_migrate_dst_release and decided MTUs for conversion host network based on those. These were test-specific decisions leaking into production settings. We now default the MTU parameters to Ansible’s omit special value, which means the module responsible for creating the network will not specify any value in the network creation API request, and the MTU will be autoselected by the cloud.
If the defaults result in undesirable values, the MTUs can still be explicitly overriden via os_migrate_src_conversion_net_mtu and os_migrate_dst_conversion_net_mtu variables.
This fix also removes dependency of production code on os_migrate_src_release and os_migrate_dst_release variables, as currently OS Migrate does not alter behavior based on src/dst release.
(cherry picked from commit c58f7d94e2ccfe52eaa8105dfa460cf0ccf7b15c)
0.8.3 (2021-06-10)
Fix
Remove unnecessary delay in conversion host reachability check. [Jiri Stransky]
In conversion host reachability check, with the wait_for Ansible module we used delay of 10 seconds, which caused a delay before the first check, causing unnecessary slowdown (20 seconds in total) of the import_workloads playbook. What we probably meant to use was sleep, which sets the interval between checks. This is now fixed and the sleep property is set to 5 seconds.
(cherry picked from commit de1647b8ac43461d521e9e7d002eebc3a2227135)
Only use ‘type’ on keypairs when it is supported. [Jiri Stransky]
SDK should support ‘type’ on keypairs since 0.32 [1], however even with 0.36 this error still appears when testing against latest Devstack:
openstack.exceptions.BadRequestException: BadRequestException: 400: Client Error for url: http://192.168.122.93/compute/v2.1/os-keypairs, Invalid input for field/attribute keypair. Value: {‘type’: ‘ssh’, <REDACTED>, ‘name’: ‘osm_keypair’}. Additional properties are not allowed (‘type’ was unexpected).
This error disappears when testing with latest SDK. It’s not easy to track down what exact combinations of SDK/API are needed to allow for ‘type’ in keypair creation request. So we’ll first try with ‘type’, and if that request fails with error 400 (Bad Request), then we retry without ‘type’. If that request still fails, the keypair import module fails too.
[1] https://github.com/openstack/openstacksdk/commit/15baef656ac56421a71e691982a70b218110f18d
(cherry picked from commit 3edc278aaca222d48b587caa9e8c9f1d3162d98d)
Flavor extra_specs import with OpenStack SDK older than 0.51. [Jiri Stransky]
The previous implementation of extra_specs import on flavors only worked with SDK >=0.51. A fallback implementation is now added for running with older SDKs. This fallback implementation can be dropped in the future after we drop support for SDK <0.51.
(cherry picked from commit 224b779da2e552094964bde5127a715c58e48e22)
Handle when flavor extra_specs are reported as None from API. [Jiri Stransky]
Most recent APIs+SDKs always give empty hash for extra_specs. However, some older versions may also return extra_specs as None. This could break OS Migrate. The issue is now fixed by making sure we set internal variables to {} whenever some extra_specs parameter (either loaded from YAML or queried from SDK) is None.
(cherry picked from commit 29c7d17856984487f628676478fbdecdd8526ecd)
0.8.2 (2021-05-25)
Fix
Make sure the src/dst release is an integer. [Carlos Camacho]
This commit cast the os_migrate_src_release and os_migrate_dst_release variables to integers.
Fixes: #408 (cherry picked from commit 6902b3d9be04823d5ef654a909a5b1baafec7d8e)
Export RouterInterface of HA and DVR routers. [Jiri Stransky]
Previously when exporting router interfaces, only ports with device_owner value of ‘network:router_interface’ were picked up. This would not export interfaces correctly for HA/DVR routers. We now also export ports with device_owner values of ‘network:ha_router_replicated_interface’ and ‘network:router_interface_distributed’ for HA/DVR routers.
(cherry picked from commit d486fe9132760bc054e66bddbdc4f6702550da81)
0.8.1 (2021-05-12)
Fix
Project parent references allow domains too. [Jiri Stransky]
Keystone has 2 types of domains - domain objects, and project objects with is_domain=True. OS Migrate correctly worked only with the latter until now. Domain objects are now working too, and it is tested in functional tests.
Preserve domain on user export/import. [Jiri Stransky]
We now preserve domain when migrating users. This means the domain of a given name needs to exist in the destination cloud.
Preserve domain and parent for projects. [Jiri Stransky]
The references to project parents and domains are now exported and used during import. This makes it possible to export/import projects from non-default domains, but also requires making sure that the domain exists in destination.
Preserve keypair type on migration. [Jiri Stransky]
Type can be ‘ssh’ or ‘x509’, let’s preserve the value.
The ‘private_key’ param is removed since it is always null for exports/imports (it is only used when Nova auto-generates a new key).
The keypair migration test is moved from as-admin test suite into as-tenant test suite.
0.8.0 (2021-04-23)
New
Configurable timeout for non-immediate actions. [Jiri Stransky]
Until now, non-immediate actions and blocking calls to OpenStack SDK within workload migration have used a hard-coded timeout of 30 minutes. New variable os_migrate_timeout has been introduced for setting the timeout (in seconds), the value is 1800 (= 30 minutes).
Add volume attachements info to workload exports. [Jiri Stransky]
In the workloads YAML file we now write information about volume attachments into the _info section. The actual volumes to migrate are looked up again during the import phase, so this new serialized information does not have any effect on the migration, but it can be helpful when debugging.
Ensure Nova server creation is successful and emit detailed info. [Jiri Stransky]
The import_workload_create_instance module now outputs a full workload serialization of the server created in the destination cloud. When running Ansible in verbose mode, it can be used as debugging information, and allows for direct comparison of the destination workload to the contents of workloads.yml file.
This requires that the import_workload_create_instance module waits for the Nova server to enter ACTIVE state, to have full information available. Typically this only adds several seconds to the workload migration time, which is negligible in the full picture. By waiting, we can also make sure the server indeed does go ACTIVE and not into ERROR state, which may have previously gone unnoticed with the “fire and forget” server creation approach.
Validate destination prerequisites before migrating a workload. [Jiri Stransky]
Before workload storage migration is started, it is now validated that the server’s references to resources like flavor, image, security group, network, or subnet can be properly resolved in the destination cloud. This aims to avoid situations where the block storage is first successfully copied and then the server creation fails due to a trivial problem.
Validate resource data structure. [Jiri Stransky]
When resources are being loaded, we now validate that the resource has all the expected params and _migration_params and _info.id. Errors are reported for resources that cannot be parsed. The resources in error messages are identified by type+name+id tuple, if those attributes of the resource aren’t missing.
Filters in import playbooks. [Jiri Stransky]
The os_migrate_<resources>_filter variables that are already used to filter resources during export, are now respected during import too. That means user can choose whether they want to filter during export or during import, or even combine the approaches to export a subset of the existing resources, and further limit them into smaller chunks during import.
Fix
Flavor export/import. [Jiri Stransky]
Flavors were prevented from importing due to several issues around parameters passed into the flavor creation API request, this is now fixed. Extra specs are now being set after the flavor is created.
Ensure new security groups don’t have auto-created rules. [Jiri Stransky]
When creating a security group, OpenStack automatically inserts 2 basic rules to allow all egress traffic (one for IPv4, one for IPv6). Creating these blanket-allow rules automatically in destination is undesirable, as the user may have replaced the default rules with some more restrictive ones in the source security group.
When creating a new destination security group, we now immediately delete the auto-created rules. If the source security group kept the default rules, they got exported like any other rules, and will be imported into destination during rule import.
Skip migration failure cleanup when there is nothing to clean up. [Jiri Stransky]
Previously the source/destination cleanup would be attempted on a failed migration, even though the migration didn’t progress far enough to create the variables that are needed for cleanup. This would then produce a false alarm error during the cleanup. Now it’s fixed, the cleanups are attempted only if there is necessary info available (if the migration progressed far enough for the cleanup to be necessary).
Project sameness handling consistent with other resources. [Jiri Stransky]
We were missing ‘id’ in project resource and the sameness was determined by ‘name’. This could run into issues in multi-domain setups. The sameness check is now consistent with other resources and is based on ‘id’.
0.7.0 (2021-02-18)
New
Testing with lower bound requirements. [Jiri Stransky]
We started testing with two OpenStack SDK versions:
0.52.x which is close to latest but not yet affected by a security group rules bug [1].
0.36.x which is in Train release (OSP 16).
Adds Logic to Unsubscribe RHEL Conversion Hosts. [Ryan Brady]
This patch adds logic to unsubscribe systems from RHSM when using RHEL as the OS for a conversion host. The unsubscribe is called for both the source and destination conversion hosts just before they are deleted.
Creation of Floating IPs on migrated workloads. [Jiri Stransky]
A new migration parameter ‘floating_ip_mode’ is added into the workload serialization. The current possible values are ‘auto’ and ‘skip’.
‘auto’ means that floating IP creation is attempted for the VM, if the source VM had floating IP(s). The floating IP info is already being exported with the workload.
Currently we only support auto-assignment of the floating IP address. (Exact assignment is typically not allowed by policy for tenants, but this should be investigated too.)
Each floating IP serialization has a ‘floating_network_ref’ reference, which specifies the network where the floating IP should be created. The ‘fixed_ip_address’ parameter controls which port/address will the floating IP attach to.
‘skip’ means that even if the workload serialization does include serialized floating IP subresources, their creation in destination cloud will not be attempted.
The ‘floating_ip_mode’ parameter defaults to ‘auto’.
Parallel installation of conversion hosts. [Jiri Stransky]
The updating and installing of packages on conversion hosts can take minutes. These actions were done in series until now (src then dst), but now they’re done in parallel, which should speed up execution of the conversion host deployment playbook.
Floating IP export for workloads (servers) [Jiri Stransky]
When exporting servers, export information about their floating IPs.
Guide for contributing changes to documentation. [Jiri Stransky]
For developers, but also for users who find the docs lacking in some way. They can either report the issue on Github, or submit a patch with the changes.
Changes
Make boot volume prefix configurable, default to ‘os-migrate-’ [Jiri Stransky]
The boot volume prefix has previously been hard-coded to ‘rhosp-migration-’ and ‘rhosp-migration-root-’, we now make the prefix configurable via os_migrate_workload_boot_volume_prefix variable and it is defaulted to ‘os-migrate-‘.
Preserve tags during migration of routers, networks, subnets. [Jiri Stransky]
The tags on networking resources cannot be set during the initial creation request. Previously we weren’t setting tags at all during migration of networking resources. We are now migrating tags correctly on routers, networks and subnets using a post-create/update hook.
Convert to fully qualified collection names. [Carlos Camacho]
Starting on Ansible 2.9 all tasks can use the fully defined collection name the builtin runtime is specified in: https://github.com/ansible/ansible/blob/devel/lib/ansible/config/ansible_builtin_runtime.yml
Pass through all RHSM parameters for conversion hosts. [Jiri Stransky]
All parameters supported by the redhat_subscription Ansible module are now exposed as variables for the conversion host content installation.
Shorter names for RHSM variables. [Jiri Stransky]
The RHSM variables are renamed from previous pattern of e.g. ‘os_migrate_conversion_host_content_rhsm_user’ to e.g. ‘os_migrate_conversion_rhsm_user’ which is shorter but should still be descriptive enough and not risk a naming collision.
The ‘user’ variable is also renamed to ‘username’ to match the parameter naming on the redhat_subscription module.
Don’t feed availability_zone_hints into requests when empty. [Jiri Stransky]
The availability_zone_hints parameter for routers and networks can often be an empty list. Previously we’ve fed it as-is into the API requests. However, if the extension for network/router availability zones isn’t enabled, this can cause the request to fail.
We now only use the availability_zone_hints parameter when its not empty. This should have no effect on correctness, but it can prevent false-alarm failures.
Resilient RHEL unregistration mechanism. [Jiri Stransky]
Previously the deletion of RHEL conversion hosts would require the hosts to exist and always execute the unregisteration from RHSM. But if at least one of the hosts didn’t exist, or the host got somehow damaged and wasn’t reachable via ssh, the playbook would fail and would not delete the conversion hosts. The deletion playbook is now resilient in these scenarios and progresses towards ensuring the conversion hosts are deleted.
Workload Source Check. [Ryan Brady]
During the workload import, the code was checking the workload status via the export file. If a server had been turned on or moved into another state other than “SHUTOFF” since the export occured, the workload import would fail. This patch addresses that by checking the source cloud via the openstacksdk to ensure the workload status is “SHUTOFF” before attempting to migrate it.
Fix
Clean up in the destination cloud after a failed migration. [Jiri Stransky]
After a workload migration fails, we now clean up any migrated volumes that are in the destination cloud but weren’t in fact used in creation of a new server.
Clean up in the source cloud after a failed migration. [Jiri Stransky]
Previously when a migration would fail, the volumes in the source cloud would remain in incorrect state: boot volume copies still existing, and all VM’s volumes still attached to conversion host.
This is now fixed - source cloud cleanup is called the same way as after a successful migration. NBD exports are closed, the duplicated boot volume is detached from the conversion host and deleted, and any additional volumes are detached from the conversion host and re-attached to the source server.
This change also results in module renames:
import_workload_source_check to import_workload_src_check
import_workload_cleanup to import_workload_src_cleanup
This is to keep consistent naming of decent length, as src/dst prefixes are already used elsewhere, and likely we will also introduce import_workload_dst_cleanup module in the future.
Sparsification of volumes. [Jiri Stransky]
To have the volume sparsification work, two things have changed:
We install and start libvirt in the conversion hosts.
We specify that the NBD backends of volume qcow2 images are of raw format.
Exact name match on idempotency check during workload migration. [Jiri Stransky]
Previously we just fed the desired server_name into the compute.servers() query when checking whether a server has already been migrated. However, Nova API performs a regular expression search, so this condition matched on any substring occurence. This is now fixed by wrapping the search term between ^ and $.
Note: We use Glance API in a similar way due to it also not having a find method with project filters. Glance API seems to do exact matching on the name query.
Compatibility with OpenStack SDK 0.36.x. [Jiri Stransky]
It seems like the only place we’re hitting problems is using ‘owner’ property of an image resource. Both 0.36 and 0.52 SDK support the ‘owner_id’ property there, so let’s just use that.
Interestingly enough, this is not the case for the query parameters in conn.image.images(). There image_id is not working anywhere, it has to be image in both 0.36 and 0.52, so we’ll stick to that.
Loosen OpenStack operation timeouts in workload migration. [Jiri Stransky]
The current timeout for OpenStack operations during workload migration is 10 minutes. This can be too short when working with large storage volumes. We should make the timeout configurable, but as first alleviation we increase the timeout to 30 minutes. This should still be enough to fail in reasonable time when things get stuck, but it shouldn’t produce as many false alarms.
Always stop on errors in conversion host deployment. [Jiri Stransky]
Previously when the source or destination conversion host encountered some errror, Ansible would mark that host as failed and not run any other tasks there, but it would continue to run tasks on other hosts, most notably the migrator. This meant that e.g. subscription errors wouldn’t stop the playbook immediately, and if conversion host deployment was included into a more complex migration playbook, the migration itself would fail much later with a non-obvious error. This is now fixed: whenever an error during conversion host deployment happens, the playbook stops right away.
Fix image lookup for images owned by tenant. [Jiri Stransky]
The connection.image.find_image method doesn’t support filter arguments (like project_id), unlike other find methods. This caused failures when looking up images to be used for servers, when the image was owned by tenant (non-public).
A custom image lookup method is now implemented in our reference.py module, utilizing images() rather than find_image().
End-to-end tests are also amended to test one workload spawned from public image, and one from tenant-owned image.
Do not look up server image when boot_disk_copy: true. [Jiri Stransky]
When migrating a server which is boot-from-image in source cloud, and boot-from-volume in destination cloud (due to setting boot_disk_copy: true), the server creation would still try to look up image_ref in the destination cloud, and fail if the image doesn’t exist.
This is now fixed. With boot_disk_copy: true we don’t need any images in destination, so we no longer perform an image lookup.
Fix building OS Migrate from source. [Jiri Stransky]
The conditional for sourcing virtualenv was incorrect in the script, and virtualenv wasn’t being sourced automatically. This is now fixed for users who want to build OS Migrate from source via the toolbox container.
Other
Oth: Documentation for using RHEL as conversion host. [Jiri Stransky]
Oth: Documentation for ‘floating_ip_mode’ workload migration parameter. [Jiri Stransky]
Pin openstacksdk version due to os_security_group_rule module breakage. [Jiri Stransky]
The pin works around this breakage:
‘BadRequestException: 400: Client Error for url: http://10.0.110.233:9696/v2.0/security-group-rules, Unrecognized attribute(s) ‘’remote_address_group_id’’’
It seems like remote_address_group_id is fed into the request even though we’re not specifying it, which makes the os_security_group_rule module incompatible with Train release and lower.
Docs: include information in all roles. [Carlos Camacho]
Update packages when deploying conversion hosts. [Jiri Stransky]
I ran into a bug when trying to migrate workloads, the volume export step crashed with:
qemu-img: Unable to initialize gcrypt
It turns out there was desync in package versions, we were installing latest qemu-img but libgcrypt was old. We need to update to latest when deploying the conversion hosts.
0.6.2 (2021-01-28)
Fix
Always stop on errors in conversion host deployment. [Jiri Stransky]
Previously when the source or destination conversion host encountered some errror, Ansible would mark that host as failed and not run any other tasks there, but it would continue to run tasks on other hosts, most notably the migrator. This meant that e.g. subscription errors wouldn’t stop the playbook immediately, and if conversion host deployment was included into a more complex migration playbook, the migration itself would fail much later with a non-obvious error. This is now fixed: whenever an error during conversion host deployment happens, the playbook stops right away.
(cherry picked from commit e867b359371d3bfc250b45e4c9e176b3d020d37c)
Loosen OpenStack operation timeouts in workload migration. [Jiri Stransky]
The current timeout for OpenStack operations during workload migration is 10 minutes. This can be too short when working with large storage volumes. We should make the timeout configurable, but as first alleviation we increase the timeout to 30 minutes. This should still be enough to fail in reasonable time when things get stuck, but it shouldn’t produce as many false alarms.
(cherry picked from commit 790a44059e15632c4ac008ba3639700238c4b98f)
Fix image lookup for images owned by tenant. [Jiri Stransky]
The connection.image.find_image method doesn’t support filter arguments (like project_id), unlike other find methods. This caused failures when looking up images to be used for servers, when the image was owned by tenant (non-public).
A custom image lookup method is now implemented in our reference.py module, utilizing images() rather than find_image().
End-to-end tests are also amended to test one workload spawned from public image, and one from tenant-owned image.
Resolves: https://github.com/os-migrate/os-migrate/issues/346 (cherry picked from commit 24d383dd1373bc4e09210f56b85052c57338583c)
Do not look up server image when boot_disk_copy: true. [Jiri Stransky]
When migrating a server which is boot-from-image in source cloud, and boot-from-volume in destination cloud (due to setting boot_disk_copy: true), the server creation would still try to look up image_ref in the destination cloud, and fail if the image doesn’t exist.
This is now fixed. With boot_disk_copy: true we don’t need any images in destination, so we no longer perform an image lookup.
Resolves: https://github.com/os-migrate/os-migrate/issues/347 (cherry picked from commit 5a57a52778a8668b5ca1b2787b5cc89d221d4449)
Other
Prepare release 0.6.2. [Jiri Stransky]
Pin openstacksdk version due to os_security_group_rule module breakage. [Jiri Stransky]
The pin works around this breakage:
‘BadRequestException: 400: Client Error for url: http://10.0.110.233:9696/v2.0/security-group-rules, Unrecognized attribute(s) ‘’remote_address_group_id’’’
It seems like remote_address_group_id is fed into the request even though we’re not specifying it, which makes the os_security_group_rule module incompatible with Train release and lower.
Resolves: https://github.com/os-migrate/os-migrate/issues/350 (cherry picked from commit 9e1109c57326c728486d85f4c6774f89bfef7710)
0.6.1 (2020-12-10)
Update packages when deploying conversion hosts. [Jiri Stransky]
I ran into a bug when trying to migrate workloads, the volume export step crashed with:
qemu-img: Unable to initialize gcrypt
It turns out there was desync in package versions, we were installing latest qemu-img but libgcrypt was old. We need to update to latest when deploying the conversion hosts.
Resolves: https://github.com/os-migrate/os-migrate/issues/322 (cherry picked from commit 3e68a4aad182472d8d343769c83640512551a673)
0.6.0 (2020-11-25)
New
Documentation for resource filters. [Jiri Stransky]
Guide for using migration parameters. [Jiri Stransky]
Include changelog in docs. [Carlos Camacho]
This patch includes the changelog in the documentation automatically.
Changes
User documentation improvements. [Jiri Stransky]
Improve clarity of user docs in various spots and add more workload migration info to the walkthrough doc.
Take info about server ports from Neutron. [Jiri Stransky]
Previously we took it from Nova but that does not contain all the needed info for advanced use cases. Now we export info from Neutron which should allow us to export more port properties.
The creation of ports during server import is still kept simple for now, via Nova. This has the benefit that if the server is deleted later, its ports are deleted automatically too. We may add an opt-in advanced port creation method via Neutron to support additional properties (e.g. setting a MAC address). However, IIUC, the ports created via Neutron don’t get automatically deleted on server deletion and have to be cleaned up afterwards.
More metadata to migrate with servers. [Jiri Stransky]
Fix
Detach stuck volumes in unit tests. [Jiri Stransky]
Revert to conditional empty variable check. [Carlos Camacho]
Im following now the same approach we have in another variables check in the same makefile. Even if locally the check works in the job we have errors like: if [[ EUSE_TOOLBOX -eq “0” ]]; then Notice that EUSE instead or REUSE
Assign a default value for reusing the toolbox image. [Carlos Camacho]
Currently, the container image build job is failing as we are calling the toolbox build without the variable TOOLBOX_BUILD. This patch defaults the value to 0 if not defined, which means, build the container image.
Other
Update contributing.rst. [Ryan Brady]
minor edit, changed titles to title case
Move resource checks from utility to Resource classes. [Ryan Brady]
This patch moves the responsibility of comparing resources from the serialization class into the Resource class hierarchy so subclasses can add custom comparisons as needed.
Store workload migration logs in a separate dir. [Jiri Stransky]
Previously workload migration logs were stored directly in os_migrate_data_dir together with resource YAMLs. This meant that when migrating multiple workloads, the directory would get cluttered. Let’s separate workload migration logs into its own directory.
Reuse container images if possible. [Carlos Camacho]
Preserve keypair when creating a server from an image. [Jiri Stransky]
This will ensure ssh-ability of servers which were migrated using the from-image method. Naturally the ssh keypair needs to be migrated before migrating the server.
Migration of rebuildable servers. [Jiri Stransky]
This enables the boot_disk_copy migration parameter for workloads. When it’s false, server is booted from an image. When it’s true, server boot disk is copied and the destination server is created as boot-from-volume.
Default value for boot_disk_copy depends on each server. Servers booted from a volume default to true (as that is the only sensible option), servers booted from an image default to false.
image_ref is added to the server serialization to track the Glance image that should be used to create the server.
E2e tests for workload migration are now testing these cases:
Instance booted from an image, with boot_disk_copy: false.
Instance booted from an image, with boot_disk_copy: true.
Add migration param defaults to the resource class. [Jiri Stransky]
This will allow us to assume that some keys exist in the migration params. We won’t have to check for their existence before using them in code.
Specify the image format. [Carlos Camacho]
Create servers using proper Nova bindings, support server description. [Jiri Stransky]
This commit moves us off old bindings for server creation onto the current ones. Points of interest:
Only block_device_mapping is now used for all volumes, we can get rid of special handling for boot_volume_id.
We can and do support the description parameter on servers.
The move forced changes in how we specify security groups and NICs but we still support the important features we supported before, e.g. fixed IP preservation.
In general this enable us to support more Nova API features / instance parameters.
Resilience in e2e cleaning against failed migrations. [Jiri Stransky]
If workload migration fails, volumes might still be attached to the conversion host(s) which prevents them from being deleted. Make sure they are detached before deleting.
Stop busy-looping during volume transfer (fix for 100% CPU usage) [Jiri Stransky]
The EAGAIN error code is used when polling, meaning there’s no data to read right now but it should be tried again later. Instead of trying right away, sleep for a second. This will prevent Ansible from eating 100% CPU on the migrator machine.
Add Tenant(Project) Roles. [Ryan Brady]
This patch adds the roles and supporting code to import and export OpenStack projects.
Use an additional non-boot volume in e2e workload tests. [Jiri Stransky]
This should improve our test coverage, only instance with no attached volumes was tested until now.
Fix non-boot volume attachement. [Jiri Stransky]
After migrating, the additional non-boot volume didn’t attach onto the instance correctly, instead it was stuck in “reserved”. This was due to malformed block_device_mapping_v2 data structure. See docs [1].
Include lint checks. [Carlos Camacho]
Export/import users with functional tests. [Carlos Camacho]
Adjust docs job and build nightly image. [Carlos Camacho]
0.5.0 (2020-10-02)
Add missing role readme files. [Jiri Stransky]
Galaxy releasing fails without these.
Release 0.5.0. [Jiri Stransky]
Separates Functional Tests into Admin/Tenant. [Ryan Brady]
This patch changes the func tests to run separately for both admin and tenant roles. It follows the pattern done for the e2e testing a short time ago.
Allow running e2e tests in smaller chunks for development. [Jiri Stransky]
Currently running e2e tests does all this: deploy conversion hosts, seed pre-workload resources, then workload resources, then migrate pre-workload, then migrate workload, then clean everything up.
For working on workload migration, it would be great if we could leave conversion hosts and pre-workload resources in place between test runs, and only cycle seed-migrate-clean on the actual workload. This is now implemented, but the default test run remains unchanged.
First i can just run everything without cleaning up:
./toolbox/run bash -c “E2E_TEST_ARGS=’–skip-tags test_clean_after -e test_clean_conversion_hosts_after=false’ make test-e2e-tenant”
And then i can run workload tests specifically, which will auto-clean before and after, but only the workload parts:
./toolbox/run bash -c “E2E_TEST_ARGS=’–tags test_workload’ make test-e2e-tenant”
Put e2e scenario variables last. [Jiri Stransky]
This allows them to also override the auth file, if desired.
Add ‘boot_disk_copy’ migration parameter to workloads. [Jiri Stransky]
Default it to true for now, as currently that’s the only supported migration method.
Move server creation logic into the resource class. [Jiri Stransky]
Add _migration dict into resource serialization. [Jiri Stransky]
Currently we have two dicts within the resource serialization:
params - parameters relevant for resource creation in destination cloud, and for idempotence,
_info - additional information about the resource from source cloud, which will not be passed on into the destination cloud when creating the resource.
With work on migration of rebuildable instances, we will add a third one:
_migration - parameters for the migration process itself, not capturing properties of the resource, but rather how we want the resource to be migrated.
The example use case here is the workload migration. There is one server in the source cloud but we have at least two options how to move it to destination:
full copy (current method) - root disk is copied as a volume, and the instance in the destination cloud is booted from that volume,
rebuild (to be implemented) - instance is booted from the same glance image in the destination cloud, putting the destination instance into a similar state as after using openstack server rebuild on the original. This is better suited for instances which are stateless or only save state on external volumes, not on the root disk.
To distinguish how we want a given server migrated, we will have a parameter in the _migration section.
Update wrong links. [Carlos Camacho]
Foundation for running e2e tests as admin. [Jiri Stransky]
Use separate tmp dir for e2e tests. [Jiri Stransky]
Split file paths to prepare separate e2e tenant/admin tests. [Jiri Stransky]
Prefix and shorten export variable names. [Jiri Stransky]
We had a bunch of unprefixed variables like:
export_networks_name_filter
We should add the os_migrate prefix and we can perhaps shorten the variable name to still remain descriptive enough, like this:
os_migrate_networks_filter
The rename in this sense is applied across the project to all affected variables.
This rename also opens up the possibility to use this filter on imports in the future, allowing to work on full YAML files while selecting just a few resources to import. It cannot be done right now because the stringfilter module only supports looking at direct attribute and not nested ones, so we cannot look at [‘params’][‘name’] presently. An enhancement for stringfilter could be implemented.
Allow setting conversion subnet DNS separately for src/dst. [Jiri Stransky]
DNS nameservers is a parameter for conversion subnet where it makes sense to allow different values between src/dst clouds.
Fix readme and include a link to the project in the docs root. [Carlos Camacho]
Filter exported images by the currently authenticated project. [Jiri Stransky]
This kind of filtering is standard for our other resources, but for Glance images it was missing due to Glance API lack of functionality which translates into lack of functionality in the os_image_info Ansible module. The filtering is done explicitly in the playbook now.
Docs refactor. [Carlos Camacho]
(WIP) Adds Nova Idempotence Tests. [Ryan Brady]
This patch adds idempotence tests for the Nova Flavor and Keypair resources.
Include the fixed IP when restoring the server. [Carlos Camacho]
The nics parameter is not documented in the upstream docs, also there is no way in inject the MAC address, or assign the floating IP when executing create_server
This should be fixed by creating a port and assign it to the node before is created.
Walkthrough doc updates and formatting fixes. [Jiri Stransky]
This makes walkthrough doc up to date and generally into a better shape. Some notable changes:
Prerequisites section regarding authentication and parameters which are used for both pre-workload and workload migration is consolidated and moved at the beginning of the doc.
Nesting of headlines is fixed.
Removed mentions RH OSP.
Updated conversion host deployment to use the playbook instead of manual wiring instructions.
Updated the workload migration process to the new one which runs on plain CentOS and doesn’t require VDDK.
Command shortcuts.
Adds Roles For Nova Keypair Import and Export. [Ryan Brady]
This patch adds the roles and associated code required to export and import Nova keypairs.
Add an update test for images. [Jiri Stransky]
Add idempotence tests for image export/import. [Jiri Stransky]
Implement Glance image import. [Jiri Stransky]
Fix pylint issues. [Jiri Stransky]
It seems like pylint got updated and our code started failing the lint, this is now fixed.
Adds Import and Export Roles for Nova Flavors. [Ryan Brady]
This patch adds import and export roles, supporting files and tests for migrating Nova Flavors between clouds.
Adds Nova services to the CI stack. [Ryan Brady]
This patch adds the nova services to the stack to support functional tests.
Export of Glance images. [Jiri Stransky]
Since Glance images contain binary data, the export is done via two modules:
export_image_meta – exports image metadata (YAML file)
export_image_blob – exports image binary data
The export of binary data is optional. This split between metadata and data exporting should leave us an open door towards migrating images in a direct way between clouds (e.g. via conversion hosts) rather than via the migrator host.
Update user docs to use Keystone v2. [Carlos Camacho]
Support migrations between keystone v2 and v3. [Carlos Camacho]
Update references from ‘master’ branch to ‘main’ [Jiri Stransky]
Allow configure the ssh user. [Carlos Camacho]
These are the pending changes to allow using the centos image. [Carlos Camacho]
We use become from vars and include internal DNS to be able to install packages.
We also increase the timeout to create the conversion hosts to 10 minutes, and change the default image to be CentOS.
Adds Role to Install Conversion Host Content. [Ryan Brady]
This patch adds a role to install content for the conversion host conditionally. The role currently supports CentOS and RHEL but is written in a way that should be simple to add additional OS support if needed in the future.
Adds Link For Contributing To Developer Doc. [Ryan Brady]
Adds a link for the the contributing doc to the README.
Adds Contribution Document. [Ryan Brady]
This patch adds information to help a potential contributor add new roles or supporting code to the project.
Fixes Walkthrough Images. [Ryan Brady]
When the walkthrough doc was merged, the URIs for the images were incorrect and the images were not displayed for users. This patch fixes the URIs for the images to the correct path and removed an erroneous TODO comment left in the previous version.
Fully qualified refs in workload serialization. [Jiri Stransky]
Add Matthew into authors. [Jiri Stransky]
Use cgroupfs as cgroup manager in Github Actions. [Jiri Stransky]
Format variable names as code. [Jiri Stransky]
Restructure user documentation landing page. [Jiri Stransky]
Remove the walkthrough-like part of the root user readme, we have a specific Walkthrough doc now. Make the main readme a link directory.
Configuring the conversion host. [Carlos Camacho]
Fully qualified refs in network serialization. [Jiri Stransky]
Fully qualified refs in security groups and rules serialization. [Jiri Stransky]
Fully qualified refs in router and router interface serialization. [Jiri Stransky]
Fully qualified refs in subnet serialization. [Jiri Stransky]
Adds Walkthrough Document. [Jiri Stransky, Matthew Arnold, Ryan Brady]
This patch adds documentation for executing a parallel cloud migration using os-migrate.
Fix. [Carlos Camacho]
Improve variable naming consistency in conversion host playbook. [Jiri Stransky]
0.4.1 (2020-06-26)
Prepare release of 0.4.1. [Jiri Stransky]
Fix. [Carlos Camacho]
0.4.0 (2020-06-26)
Prepare release 0.4.0. [Jiri Stransky]
Mainly utilizes a new workload migration approach with new conversion host and conversion host deployment/deletion playbooks.
Fix subnet update test. [Jiri Stransky]
The lineinfile module only makes sure one line is present. So after it does its replacement in one spot, it stops replacing. Due to random order of subnets in the file, it sometimes replaces the description of the wrong subnet, resulting in intermittent failures. To make the tests work, we simply replace descriptions of both subnets in the func test, and then we check only one.
To keep tests consistent, use ‘replace’ instead of ‘lineinfile’ module in other tests too.
Variables cleanup. [Carlos Camacho]
This patch makes the variables used in the configurable so we can trigger different scenarios just changing this default file.
Also separates the network/flavor/image names for src/dst conversion hosts
And remove some required variables that can be safely defaulted.
Adds Functional Update Test for Security Groups. [Ryan Brady]
This patch adds a test to ensure the updates for security group resources is working properly.
Adds Update Functional Tests for Subnets. [Ryan Brady]
This patch adds the code to test the update functionality of the subnet resources.
Adds Functional Test for Network Update. [Ryan Brady]
This patch adds a functional test to ensure that updates are working for the network resource.
Adds Functional Update Test For Router Resources. [Ryan Brady]
This patch adds a test to ensure updates work for router resources.
Use conversion host deployment/deletion playbooks in CI. [Jiri Stransky]
Set up conversion hosts to talk to each other. [Jiri Stransky]
A separate ssh key is used to connect from dst conversion host to src conversion host, so that user can’t mistakenly upload their own private key into the conversion host.
Fix Resource Updates. [Jiri Stransky, Ryan Brady]
For the resources that support updating, the code previously called the update sending the resource name causing the update to fail from the corresponding API with a “BadRequest” error. This patch updates the code to use the instance of the resource itself when calling the associated update method.
Hack for conversion host reachability in CI. [Jiri Stransky]
Update workload exporting for new openstacksdk. [Jiri Stransky]
Make sure we respect virtualenv in playbooks. [Jiri Stransky]
When we run our playbooks, it seems that virtualenv is ignored even when it’s active for the ansible-playbook command invocation. We set it into the inventory explicitly, so that when venv is active for ansible-playbook invocation, it will also be active for the Ansible modules invocations.
Print whether we’re sourcing venv or not in Makefile. [Jiri Stransky]
Fix timeout logic on snapshot creation. [Jiri Stransky]
In CI we’re hitting this error:
File “/tmp/ansible_os_migrate.os_migrate.import_workload_export_volumes_payload_4g3ps56s/ansible_os_migrate.os_migrate.import_workload_export_volumes_payload.zip/ansible_collections/os_migrate/os_migrate/plugins/modules/import_workload_export_volumes.py”, line 364, in _detach_data_volumes_from_source AttributeError: ‘NoneType’ object has no attribute ‘id’
Looks like the call to create snapshot returns None. We’ll try moving the timeout into the call itself and using wait=True.
Add playbooks for deployment and deletion of conversion hosts. [Jiri Stransky]
Deploys and deletes the conversion hosts including supporting resources like networking, security group, and key pair. The deployment and deletion are within a single role to make use of common defaults file.
Relax YAML lint - allow 100 character lines. [Jiri Stransky]
The 80 character limit is a bit too strict for dealing with long Ansible variables. 100 characters should still be comfortable enough for modern screens.
Update uch image. [Carlos Camacho]
Gitignore local directory. [Jiri Stransky]
This directory allows putting some personal files in, e.g. credential or parameter files for manual testing. This directory is available in ./toolbox/venv-shell containers.
Use nbdkit if available with fallback to qemu-nbd. [Matthew Arnold]
Clean up temporary volumes and SSH processes. [Matthew Arnold]
Create destination instance and attach volumes. [Matthew Arnold]
Also add number/names of networks, without preserving MAC addresses.
Transfer exported volumes to destination. [Matthew Arnold]
Also factor out common code between source and destination conversion hosts.
Export volumes from source conversion host. [Matthew Arnold]
Fix exporting router interfaces when router name filter is undefined. [Jiri Stransky]
Just a missing default.
Docs for publishing a release. [Jiri Stransky]
Make sure that docs CI passes ‘functional’ test automatically too. [Jiri Stransky]
Without this a pull request seems to be able to get stuck waiting for the functional test to complete.
0.3.1 (2020-06-05)
Prepare release 0.3.1. [Jiri Stransky]
Fix exporting router interfaces when router name filter is undefined. [Jiri Stransky]
Just a missing default.
Closes https://github.com/os-migrate/os-migrate/issues/201
(cherry picked from commit fc81f87d147980aae45f205959d42628885f4b36)
0.3.0 (2020-06-02)
Add readme files to all roles, pointing to user docs. [Jiri Stransky]
It looks like this is now required, otherwise the Galaxy publish script fails with:
ERROR! Galaxy import process failed: No role readme found. (Code: None)
Prepare the release of 0.3.0. [Jiri Stransky]
Adds additional user docs. [Ryan Brady]
This patch adds more data to the repo README and updates the the user doc to remove clouds.yml info and replace it with the newer auth vars.
Reintroduce project filtering into workload migration idempotence check. [Jiri Stransky]
This was added earlier but errorneously, reporting that server exists even if it wasn’t true, so it had to be reverted. The project id filtering is now reintroduced into the workload migration idempotence check, and it doesn’t produce false positives.
Include the workload migration. [Carlos Camacho]
Add dst filters to workload migration. [Jiri Stransky]
This is only necessary when looking up the conversion host and when checking for idempotence. The actual workload migration uses the VM’s ID from _info, so it should be already good to go:
Include network mappings in wrapper input. [Matthew Arnold]
Import lookups filtering by project_id for pre-workload migration. [Jiri Stransky]
Rename export_prelude role to prelude_src. [Jiri Stransky]
I realized that e.g. workload import might need to fetch authenticated project IDs for both src and dst. Rather than having an export/import prelude, it may make more sense to have prelude roles for interacting with src/dst clouds.
Add project id filter to the remaining export roles. [Jiri Stransky]
By default, make sure we export only resources in the currently autheticated project.
Remove unused (and incorrect) os_security_group_rules_info. [Jiri Stransky]
This isn’t used and it’s not utilizing the filters parameter correctly.
Implement export filter for resources in current project (for networks) [Jiri Stransky]
This adds the reusable code for fetching current project ID and setting a general export filter with that project ID by default. Networks export is amended to make use of this filtering parameter.
Adds test for Router Interface idempotence. [Ryan Brady]
This patch adds a test to ensure router interface idempotence and updates the router interface resource class to handle the case when a router interface already exists when attempting to create an identical one.
This patch also fixes a small bug in the router interface where it was using the “subnet_name” key where it should have been using a “subnet_id” key. This was causing the unintended behavior of never finding a matching port when checking for ports that may have already been created, so it always tried to create the port leading to conflicts.
Capture wrapper stdout/stderr to data directory. [Matthew Arnold]
Split import_workload into two modules. [Matthew Arnold]
The ‘prelim’ module configures everything right up to the point of running the virt-v2v-wrapper container, and returns paths to the expected log and state files at that point. The import_workload module just runs the container, so it will still sit without showing progress for potentially a very long time - but this way the operator can optionally monitor the previously-provided log files as needed.
Ansible automation for the e2e tests. [Carlos Camacho]
Adds Functional Test for Security Group Rule Idempotence. [Ryan Brady]
This patch adds a functional test to ensure security group rule idempotence.
Make sure releasing is done from toolbox container. [Jiri Stransky]
E.g. if shyaml is missing, the releasing will fail with a weird error. In toolbox, we have all the releasing dependencies.
Prepare for 0.0.2 release to Galaxy. [Jiri Stransky]
First attempt at workload migration. (#170) [Matthew Arnold]
First attempt at workload migration.
This runs virt-v2v-wrapper on the destination conversion host.
Squashed commits from review process:
Rename uci_container_id to uci_container_image.
Remove flavor and security_groups objects from info.
Rename ssh_key_path to clarify what it is used for.
Pass auth parameters straight through to wrapper.
No longer forces password authentication.
Rename server_address module to conversion_host_info.
Allow specifying conversion hosts with name or ID.
Include remote wrapper directory in return value.
This helps debug migration failures.
Remove flavor_id from Server’s info_from_sdk.
Run UCI container with host networking.
Save ether_type for security group rules. [Jiri Stransky]
This is a fundamental property of the rule, specifies whether the rule applies to IPv4 or IPv6 traffic.
Adds Router Idempotence Fuctional Test. [Ryan Brady]
This patch adds a test to ensure idempotence for the router resource.
Move instance name from info to params. [Matthew Arnold]
Basic workload information export. [Matthew Arnold]
Export instance information from a source OpenStack cloud to workloads.yml. Actual workload migration will require saving more properties than this, but they will get filled in later.
Allow skipping func test cleanup only after testing. [Jiri Stransky]
Previously it was possible to skip cleanup via –skip-tags test_clean, but that would affect cleanup both before and after testing. It is now possible to skip the cleanup only after testing, with –skip-tags test_clean_after.
Fix importing router external gateway. [Jiri Stransky]
Previously routers were created without an external gateway interface, this is now fixed. The interface is created on the network with the same name as in the source cloud.
Make sure router import/export doesn’t fail if external gateway is not present. [Jiri Stransky]
Add a Router import playbook. [Jiri Stransky]
Force upgrade virtualenv. [Carlos Camacho]
Galaxy sanity checks. [Carlos Camacho]
Check build for docs folder. [Carlos Camacho]
Do not ls, use always current version as latest. [Carlos Camacho]
Fixes Subnet Update. [Ryan Brady]
Intermittently, the subnet import functional tests fail for ip_version or network_id attributes. This patch adds a readonly_params collection to the Resource class and strips out any readonly properties during an attempt to update. For the Subnet resource this patch removes the blacklisted properties in neutron source for the subnet resource (project_id, network_id) from the update args.
A follow up issue and series of patches is forthcoming for adding functional testing for the update on all resources.
closes #142
Document class vars in Resource. [Jiri Stransky]
Fixes Subnet Update. [Ryan Brady]
Intermittently, the subnet import functional tests fail for ip_version or network_id attributes. This patch adds a readonly_params collection to the Resource class and strips out any readonly properties during an attempt to update. For the Subnet resource this patch removes the blacklisted properties in neutron source for the subnet resource (project_id, network_id) from the update args.
A follow up issue and series of patches is forthcoming for adding functional testing for the update on all resources.
closes #142
Update dev docs. [Jiri Stransky]
The dev docs for running functional tests went out of date as we changed auth mechanism and as we changed how to run a subset of functional tests. They are now updated to current workflows.
Also slight edit has been made to the design doc now that we’ve made some implementation decisions where previously there was uncertainty.
Start polishing the Galaxy sanity check to publish the collection. [Carlos Camacho]
There are several checks we need to pass in order to be able to publish the collection in both Galaxy and the Automation Hub.
This will be the first change to make it possible.
With this, we can pass the required to publish into galaxy.
OOP security group rules. [Carlos Camacho]
Move README. [Carlos Camacho]
Remove old auth mechanism. [Jiri Stransky]
The removal will make sure that we cannot depend on the old clouds.yaml based auth mechanism and usage of new auth is enforced.
Switch func tests seed/clean/verify to new auth. [Jiri Stransky]
We already have new auth used in all modules, but the func test code still used the old auth. Switch func tests too.
Switch remaining modules to new auth. [Jiri Stransky]
This swithces the remaining Ansible modules (for routers, router interfaces, subnets, security groups, security group rules) into using the new auth mechanism.
Move Network to use transferable auth instead of named clouds. [Jiri Stransky]
This move Network to use an auth mechanism that we can use also with Universal Conversion Hosts (something that is easily pushable as data rather than having to reference a clouds.yaml file).
Script to extract auth info from clouds.yml. [Jiri Stransky]
We’ll need to stop using named clouds because we’ll need to feed credentials into universal conversion host, we can’t simply copy user’s clouds.yml there. So let’s start moving towards the other auth mechanism used - auth + auth_type + region_name. IIUC should still support advanced mechanisms like Application Credentials, via different values in auth_type and corresponding auth dictionary.
Fix subnet import idempotence. [Jiri Stransky]
The last param that must be sorted for import idempotence to work is ‘host_routes’. It’s a list of dicts so it needs a list of keys to look at in each dict. The list-of-dicts sorting is implemented in Resource class to be reusable.
OOP sec group. [Carlos Camacho]
Split security groups and rules from networks. [Carlos Camacho]
Remove Transible code to cleanup the repo structure. [Jiri Stransky]
We’re not using it, and where OS-Migrate is heading now, it doesn’t seem we will literally need to take parts from it. It can only confuse folks who try to understand OS-Migrate, making them think that the code under openstack_ansible directory is being used for the migration. If we need to revive any of that code, we can find it in git history.
Add Sagi as a co-author. [Jiri Stransky]
The Transible code from Sagi has been an inspiration for parts of what we do in OS-Migrate.
Fix Subnet Update. [Ryan Brady]
During the idempotency testing, the subnet update calls to the API failed for ip_version and network_id. These parameters are accepted for the create, but not the update. Sorting the allocation_pools and dns_nameservers migtigated the condistion that caused this error.
Lint target needs to make sure the collection is installed. [Jiri Stransky]
Otherwise it fails with:
Couldn’t parse task at /root/os_migrate/tests/func/idempotence/security_group.yml:28 (couldn’t resolve module/action ‘os_migrate.os_migrate.os_security_groups_info’. This often indicates a misspelling, missing collection, or incorrect module path.
Fixes CI. [Ryan Brady]
The CI started failing today when attempting to run the checkout source step. I started researching online and found there is a race condition issue with the actions/checkout@v1[1]. This patch changes the checkout action to use @v2 which supposedly addresses this issue.
Security Group Idempotence Test. [Ryan Brady]
This patch adds idempotence tests for security group import and export.
partially-addresses: #21
Router interfaces import. [Jiri Stransky]
Import interfaces which Routers have in tenant Subnets. This is a more advanced resource than the basic “CRUD” approach. First we need to create a port with the desired IP addresses in the desired subnets, and then we attach those ports to the router. This ensures that the router interface IP addresses in tenant networks are migrated without change.
Router interfaces export. [Jiri Stransky]
Export information about interfaces which Routers have in Subnets.
Subnet Idempotence Testing. [Ryan Brady]
This patch adds idempotence test for subnet import and export.
partially-addresses #21
Adds Subnets Import Role. [Ryan Brady]
This patch adds the ability to import subnets from a os_migrate yaml structure.
Use mysql service instead. [Carlos Camacho]
Run using the mysql service. [Carlos Camacho]
Add basic import of Routers. [Jiri Stransky]
Import routers. Some params were refused on the router creation request so i had to move them to ‘info’.
Refactor router export via Resource subclass. [Jiri Stransky]
The router functionality is just export right now, let’s first refactor it as Resource subclass before we add import.
Subnets Refactor. [Ryan Brady]
This patch refactors the export subnets role to use the new Resource class.
Default implementation for Resource.create_or_update. [Jiri Stransky]
By extracting the find/create/update calls for the underlying SDK object, we can provide a default implementation for the create_or_update method on Resource. Subclasses can either provide the find/create/update mini-methods and reuse the create_or_update logic, or override the create_or_update method fully.
Do not lint YAML files in ./tests/func/tmpdata. [Jiri Stransky]
These resource files are generated in tests via YAML dumper and they don’t pass the lint check on indentation :).
Also reformat the linters.sh file to make the commands a bit easier to read.
Remove begining of line. [Carlos Camacho]
Run ansible-lint last, feed all files into a single process. [Carlos Camacho, Jiri Stransky]
Ansible-lint is the slowest linter, running it last should speed up catching Python and YAML lint issues via test-fast target.
Also feeding all files into a single ansible-lint process seems to be much faster than running ansible-lint per file.
Fix YAML nits. [Carlos Camacho]
Python lint checks. [Carlos Camacho]
Introduce Resource class. [Jiri Stransky, Ryan Brady]
We created a Resource class as the base class for all our exported resources. It improves the DRYness of our overall code and is suitable for both export and import. It uses method composition internally so that it can be easily “disconnected from network” for purposes of unit testing.
The common interfaces are now backwards compatible, e.g. the add_or_replace method for exporting resources now accepts both direct resource data or an instance of Resource.
Fis bashate. [Carlos Camacho]
Run Ansible lint. [Carlos Camacho]
Fix podman’s TTY-related options in vagrant-run. [Jiri Stransky]
The TTY-related options passed to create vs. start are not perfect substitutes of each other.
When calling start, we want to always pass -a even when not in interactive shell, simply to avoid launching the container in background (podman start ~= podman run -d; podman start -a ~= podman run).
There is no -t equivalent for podman start. It looks like we need to decide on allocating pseudo-TTY in podman create.
I’m not 100% sure on the -i behavior but from docs it looks like we either want to pass it to both or neither.
Allow running functional tests more individually. [Jiri Stransky]
Now one can run this to run only Network tests for example:
./toolbox/run bash -c “FUNC_TEST_ARGS=’–tags test_network’ make test-func”
or Network + Subnet tests:
./toolbox/run bash -c “FUNC_TEST_ARGS=’–tags test_network,test_subnet’ make test-func”
Special new use case is:
./toolbox/run bash -c “FUNC_TEST_ARGS=’–tags test_clean’ make test-func”
which will only clean up the test data from the named clouds.
Implement Router export. [Jiri Stransky]
Note that this doesn’t export router’s interfaces on private subnets, they will have to be exported/imported via a separate resource.
Write resource correctly when iterating over resources. [Carlos Camacho]
We are writing only the last item in the array to the output file.
Closes:#100
Import security groups and rules. [Carlos Camacho]
Including unit and functional tests.
Registry.fedoraproject.org consistently fails on pulling images. [Carlos Camacho]
We had several CI issues related to errors when pulling the container image from registry.fedoraproject.org
Maybe we wont hit this anymore if we use Docker Hub. Also, the image shuold be exactly the same.
Include functional tests from exporting security groups and security group rules. [Carlos Camacho]
We also change the way we query for the security group and rules before, we were querying the old API which returns a Munch object instead of the API object we were waiting for.
Correct $HOME ownership in vagrant-run and vagrant-shell. [Jiri Stransky]
I was considering several options how to solve the issue of incorrect $HOME ownership in containers:
Give the user sudo rights to do the chown in entrypoint. Bad mainly because we don’t know which user we’ll run as when creating the container image, and for sudo to work in a container we also need to mount in /etc/shadow (even when using NOPASSWD sudo).
Launch the container as root, perform chown in its entrypoint, and then su to another user (we’d pass it in via env variable) to run the real command. Better than previous but reimplementing the –user argument of podman run via an env variable and custom entrypoint still feels somewhat suboptimal, especially if Vagrant is not the only use case for the os_migrate_toolbox image.
First create the container without starting it, tweak the container’s filesystem as needed via podman mount, and then start the container.
The last options feels the cleanest and it’s the one implemented in this commit.
Add validation of unique resource naming. [Jiri Stransky]
This adds validation of resource files. Unique resource naming in ensured per resource type. Multiple resource files can be fed into the validation, so if someone had a resource file setup where resources of the same type are spread across multiple files, the names would still get validated properly.
Validation can currently be executed in 2 ways:
After loading resources, before importing them (can be disabled by a boolean flag).
Independently on all resources in a directory via validate_data_dir playbook.
Adds Subnet Export. [Ryan Brady]
This patch adds the ability to export subnets from an openstack cloud into a yaml file.
Use id-based iteration for Security Groups and Security Group Rules. [Carlos Camacho]
Including matching id by name for the security group name on rules param section. [Carlos Camacho]
This commit adds the parameter security_group_name to the params section.
Closes #79
Update Vagrant-Devstack setup to Fedora 30. [Jiri Stransky]
Fedora 29 is no longer supported.
Closes https://github.com/os-migrate/os-migrate/issues/74 Closes https://github.com/os-migrate/os-migrate/issues/65
Adding the export security groups and security group rules. [Carlos Camacho]
This patch exports the current deployed security groups into the yaml os-migrate format.
It also export each security group rule per security group.
Functional CI fixes. [Jiri Stransky]
This moves the container build earlier in the functional CI. There seem to be flakes recently in the image build, and moving it earlier in the CI steps should help us fail faster.
Also it makes sure that devstack is ran from /home/stack when it is ran as stack user. We hit weird ownership issues when running as stack user but devstack is under /home/runner.
Assert correct os_migrate_version when loading files. [Jiri Stransky]
Currently we do not guarantee any backward/forward compatibility in serialized files. Using different version of OS-Migrate between export and import isn’t supported. Should users get to a situation where differing versions are needed between export and import, they will have to investigate the risks and e.g. sed the version in files to the expected one. We’re defaulting to being defensive here.
I was thinking whether this could go into validations as one of the validation errors, but in fact if the versions are different, we cannot even guarantee that the validations will work correctly, so let’s be careful and just raise a clear exception right away.
Allow duplicate names on export - iterate over resources by ID. [Jiri Stransky]
We want to export the resources as they are, even if they’d fail validation e.g. due to duplicate naming. Validation will be performed on the exported files and/or prior to importing.
This requires an amendment to stringfilter to be able to filter a list of dicts and not just a simple list of strings. The attribute optional argument is added. This is consistent with behavior of Ansible’s sort filter. By default it operates on flat lists, but with attribute argument it can be made to operate on lists of dicts.
Decide resource sameness on export by ID rather than name. [Jiri Stransky]
This change is fueled by work on validations. We have essentially 2 options there:
Export all resources including ones with empty name and duplicate names. Then run validatiton on the exported files. This requires that resources are iterated over for export by ID and not by name.
This is nicer because the files can be perhaps edited by hand to amend the naming on destination cloud without forcing name changes on source cloud. Validation can be re-ran arbitrarily on the files to iterate forward.
Validate the state in the source cloud to verify whether export can be ran, and change the state in the source cloud until unique naming is achieved. Then run export.
This is less nice because amendments of the source cloud are necessary before an export can be performed.
To achieve the former (nicer) option, we need to start exporting by ID. First step, which is done in this commit, is to amend the resource sameness check which is used during export to figure out if a resource should be added into a file or if it’s already present.
Do not enable non-existent Copr repo. [Jiri Stransky]
Ian’s copr repo was fixing iSCSI in Fedora but it doesn’t exist anymore. Let’s just remove it because we don’t need Nova instances working right now, and we’ll probably be moving onto Fedora 30 or CentOS 8 subsequently.
Fix prerequisites for Devstack. [Jiri Stransky]
python2 is needed because /usr/bin/env python must work, and python3-pyyaml must be absent.
Better devstack logging. [Jiri Stransky]
Use the LOGFILE variable supported by devstack, instead of tee.
Initialize vagrant storage pool in vagrant-destroy if needed. [Jiri Stransky]
This way you can call ./vagrant-destroy right after starting vagrant-shell. Previously it would complain that the pool doesn’t exist. The libvirt session for vagrant is transient and the pool gets lost every time the container is stopped, but the storage content is not lost and the pool can simply be recreated on top of the same dir.
Allow local edits in Devstack. [Jiri Stransky]
Now when devstack fails, you can edit files locally and run vagrant provision to retry.
Do not build container with Vagrant in CI. [Jiri Stransky]
We don’t need Vagrant in CI and it pulls a lot of deps, skipping its installation in CI should speed up the jobs.
Fix checker of resource sameness, add tests for it. [Jiri Stransky]
When two resources have the same type but their names are undefined, we should not consider them the same. This is now the case after a copypasta typo is fixed, and tests are added to verify correct behavior.
Implement generic idempotence checker. [Jiri Stransky]
The resource_needs_update function should be generically usable on all our resources to check whether some resource is in target state or needs update (= idempotence checks). It is right away plugged into the check method for Network.
Simplify and speed up Network functional tests by using name filter. [Jiri Stransky]
This ensures we only export the network we’re interested in and we don’t have to hack the resulting file. There is also slight speedup due to making fewer API calls.
Implement advanced name filtering (exact match, regex) [Jiri Stransky]
This adds a new filter plugin ‘stringfilter’ which allows advanced conditioning which resources to export, by their name. Either by exact match or regular expression match, on the names of resources that we want to export.
Document reusable module_utils functions and network functions. [Jiri Stransky]
Documents what our functions (especially the reusable ones that are intended to be shared) should do.
Since Pydoc is a free-form format, i went for very simple way of describing the functions, trying to maximize the information provided per characters typed and limit rigid boilerplate. This should help us be diligent about documenting the functions by making documentation not much pain to write :).
Adapt conventions based on Security Groups discussion. [Jiri Stransky]
Work on Security Groups spawned a discussion about some conventions. This patch performs following adaptation:
Use ‘_info’ instead of ‘info’ for resource properties which are irrelevant for importing and idempotence checks. This should further reduce a chance of conflict with some real property and allow us to ignore any ‘_info’ also in nested dicts when importing / checking idempotence.
Better namespacing for Network type serialization. ‘openstack.network.Network’ rather than ‘openstack.network’.
Resource field names for type/params/_info are now constants to prevent typos.
Use tab in Makefile. [Jiri Stransky]
This is make’s weirdness, it wants tabs. Here the spaces worked by accident since they weren’t the beginning of a recipe, rather inside a bash multi-liner. But it can throw off indentation, so let’s stick to tabs everywhere.
Initial user documentation. [Jiri Stransky]
Make sure toolbox container RPMs are up-to-date when building. [Jiri Stransky]
We install a bunch of (latest) RPMs but until now we didn’t update the others. This could result in a mixture of RPMs that has never been tested together in Fedora. Let’s update packages before installing additional ones.
Remove Tmit PoC build requirements. [Jiri Stransky]
I thought i was pasting just vagrant-libvirt package list but in fact pasted full Tmit build package list. We don’t need the Rust bits in OS-Migrate.
GCC needs to be explicitly added because it was an unstated dependency pulled in by the previous package list.
Use tags in functional tests instead of separate playbooks. [Jiri Stransky]
To run subset of functional tests, the current pattern was to create a full playbook which would only include the desired parts.
An alternative pattern might be to use –skip-tags or –tags with the test_all.yml playbook. This might allow better mix-and-match execution for particular test environments and developer scenarios.
Due to env vars not being automatically passed into the toolbox container, currently there are two ways to test this approach e.g. by skipping the network tests:
./toolbox/shell FUNC_TEST_ARGS=’–skip-tags test_network’ make test-func
and
./toolbox/run bash -c “FUNC_TEST_ARGS=’–skip-tags test_network’ make test-func”
The 2nd case perhaps deserves some improvement (list of env vars we want to detect and pass into toolbox container automatically?) but that’s beyond the scope of this commit.
Split documentation between developer and user parts. [Jiri Stransky]
Exclude docs from CI. [Carlos Camacho]
Fix qos_policy ref in Network and remove TODOs. [Jiri Stransky]
Fix a case of copypasta in network_refs_from_ser function.
Remove the TODOs related to the Network refs. Originally i wanted to load also project_name and subnet_names list, but those are not necessary for importing under current OS-Migrate assumptions, and fetching this information in network_refs_from_sdk and network_refs_from_ser would slow down the import/export with unnecessary API requests. For now let’s assume this info does not need to be queried, we can re-evaluatethe decision if we get any hints that this info would be useful.
Add utility functions for fetching references. [Jiri Stransky]
References handling for import/export. [Jiri Stransky]
ID-based references to other resources need to be converted to name-based ones on export, and back to ID-based ones during import. This is now implemented for Network’s qos_policy_id/qos_policy_name reference.
The functions which talk to OpenStack API are still kept separate from functions which only do serialization/deserialization data mangling. This makes testing and hopefully also reasoning about what happens where a bit easier.
There’s perhaps still opportunity to extract some of the reference handling code into a new utility Python module, and add more references to Network (its Project name and contained Subnet names). This can be expanded on in another commit.
DRY parameter copying in serialization. [Jiri Stransky]
This DRY code doesn’t reduce line count but it does simplify it conceptually and is less prone to typos.
Check that what is passed into serialize_network is indeed Network. [Jiri Stransky]
Network import idempotence. [Jiri Stransky]
For idempotence of imports we largely have all the pieces, all that needs to be done is to serialize the existing network into our format and compare if the resulting ‘params’ are the same as the ones provided to the import_network module.
Crucial is looking at ‘params’ only and not ‘info’, as ‘params’ contain the important network parameters that are meant to be migrated, while ‘info’ contains non-migratable info, e.g. the network’s UUID.
Export idempotence common code, tests, idempotence for Networks. [Jiri Stransky]
This adds export idempotence common code - when replacing a resource in a resource YAML file, we first check if indeed anything changed. If it didn’t, we don’t overwrite that resource and don’t even write into the file.
This is now utilized for Network resources specifically. Some things coming from OpenStack API seem to have non-deterministic order, e.g. the list of networks itself, or list of subnet IDs in a Network. To make the idempotence work properly, i’ve added sorting where applicable.
Functional tests which re-execute the export and check that the file content is the same are added.
Ansible.cfg for func tests - timestamps and nicer output. [Jiri Stransky]
The profile_tasks callback plugin adds timestamps, allowing us to check what actions took how long. And YAML log output is generally easier/faster to read than the non-indented JSON.
Wait until the Devstack fix is merged. [Carlos Camacho]
We need to wait until https://review.opendev.org/#/c/704136/ is merged to remove the pin.
Design doc - challenges of problem domain and how we deal with them. [Jiri Stransky]
This is part of the design doc which i forgot to transfer from Tmit [1]. It is useful to have this rationale captured.
[1] https://gitlab.com/jistr/tmit/blob/4559f58718f133e8ce44e50f31fef8485ff66fc7/doc/design.md#L107-121
Make ‘build.sh’ use ‘releases’ dir and make ‘publish.sh’ use ‘build.sh’ [Carlos Camacho]
Build and publish the collection to Galaxy. [Carlos Camacho]
This patch builds on each merge a new version of the collection, then this build should be published to Galaxy.
Update version in const.py based on galaxy.yml when building the collection. [Jiri Stransky]
Initially i wanted to just load the collection manifest from const.py, but it’s not possible, more details at [1].
Our next best shot is probably to auto-update const.py constant OS_MIGRATE_VERSION each time we build the collection, which is what this commit implements.
[1] https://github.com/os-migrate/os-migrate/pull/28
Closes #12
Use network class from the .network. openstacksdk submodule. [Jiri Stransky]
I was initially surprised that the network object from openstacksdk is inconsistent between creation and fetching, and then i realized that openstacksdk probably has 2 different Python APIs. The Connection.get_network() function returns a different object than Connection.network.find_network().
I’m guessing the API calls directly on Connection are older and we should probably avoid using them. On brief inspection, the code under .network. seems better documented and a somewhat nicer API. It is also the way recommended in the relevant guide:
https://docs.openstack.org/openstacksdk/train/user/guides/network.html
Resource file reading and networks import. [Jiri Stransky]
The import_networks playbook and role, and import_network module are for importing networks specifically.
The read_resources module is meant for all importing in general. It loads the resource YAML file structure into an Ansible variable so that individual resources can be iterated on inside the roles.
Small deserialization utility functions are added too, for transforming serialized data into parameters for OpenStack SDK when creating/updating resources.
There is a block in functional tests that isn’t very nice but it can be fully removed once we implement name-based filtering on export:
Add minor fixes to unit/functional tests and readme. [Carlos Camacho]
This patch add:
Adds a name convention for the tests, currently [unit|functional]-all
where all should be replaced for the specific type of tests that need to run, i.e. unit-network or functional-network.
Moves the functional tests to run with latest python 3.7.
Removes unneeded python versions in unit tests.
Improves the README with links to current tests for merged commmits.
Add a very initial logo for the project.
Add serialization basics and network exporting. [Jiri Stransky]
This adds serialization and resource file read/write basics, and adds rudimentary network exporting.
We may further think whether using some classes for our structures makes sense, or if just plain dict/list structures are good enough (my initial approach in this commit). Given that Python is a dynamic language, we won’t get any compile time validation even if we do add classes/structures for the data, and any errors will pop up at runtime either way. This already manifested in dealing with the openstacksdk’s Network class – even though it’s a class, it didn’t feel like we’re getting many benefits from it over a plain dict, it is very free-form, see serialization method in network.py.
Fix filesystem ownership in CI. [Jiri Stransky]
We were hitting weird issues in filesystem ownership [1] which failed devstack install. Making sure the ownership is correct seems to fix the problems.
Add CI. [Carlos Camacho]
Improve documentation. [Carlos Camacho]
Add a note about Vagrant VM lifespan being tied to vagrant-shell. [Jiri Stransky]
Prevent nesting of toolbox commands. [Jiri Stransky]
Rough draft of workflow on Ansible level. [Jiri Stransky]
This is a very high-level idea of how we might structure the workflow to give a decent debuggability and hackability of our Ansible-based tooling.
High-level development goals. [Jiri Stransky]
This list was originally for Tmit PoC (https://gitlab.com/jistr/tmit), and i think a lot of it could apply for OS-Migrate, so i post it here, with minor amendments.
Add network serialization stub code. [Jiri Stransky]
The data from serialized network is split between “params” and “info”, the idea being that params are important values which are needed when importing the network, while info are not used when importing but might be useful for a human looking through the exported data.
Call vagrant-up from vagrant-snapshot-revert if needed. [Jiri Stransky]
Fix for Fedora 31 vagrant-run. [Jiri Stransky]
It seems that /etc/qemu isn’t required but /dev/net is.
Update toolbox to F31, mount /etc/qemu for Vagrant. [Jiri Stransky]
Vagrant container - get rid of env vars which are undefined on Fedora 31. [Jiri Stransky]
Add skeleton for unit and sanity tests. [Jiri Stransky]
This adds Makefile targets “test-sanity”, “test-unit” and a shorthand target for both, called “test-fast”. The ansible-test tool is used for sanity and unit testing, and an empty unit test is added to show that the unit tests run.
Autodetect whether containers run interactively. [Jiri Stransky]
This automatically enables e.g. colored terminal output.
Add a stub of network exporting. [Jiri Stransky]
This is a very rough “i do nothing yet” prototype, but it should give us something to talk about re structuring the code and tests. It is exercised from the functional tests already.
Func tests now seed and clean up a network. [Jiri Stransky]
Clarify functional tests’ clouds.yaml. [Jiri Stransky]
Remove Vagrant’s clouds.yaml, fix gitignore. [Jiri Stransky]
Make it possible to run functional tests just for network migration. [Jiri Stransky]
Rename test_full to test_all. [Jiri Stransky]
Document how to run functional tests. [Jiri Stransky]
Test-func target depends on reinstall. [Jiri Stransky]
Prior to using a collection, it needs to be installed in ~/.ansible in the toolbox container, so tests should make sure we’ve built and installed the latest state.
Functional tests skeleton. [Jiri Stransky]
Add python3-devel to toolbox and openstacksdk to virtualenv. [Jiri Stransky]
Make targets for building and installing the collection. [Jiri Stransky]
Go for os_migrate namespace right away. [Jiri Stransky]
It seems that the ‘jistr’ namespace might be referenced in may places, let’s just go for ‘os_migrate’ anticipated namespace right away and we’ll request it before publishing, or figure out alternatives then.
One more note about naming. [Jiri Stransky]
Fix capitalization in readme. [Jiri Stransky]
Document dev env setup steps. [Jiri Stransky]
Scripts for running vagrant halt and destroy. [Jiri Stransky]
Add openstack to ansible (transible) converter. [Sagi Shnaidman]
Add run/shell commands with venv active inside. [Jiri Stransky]
Interactive editing of toolbox image. [Jiri Stransky]
Move virtualenv into the container image. [Jiri Stransky]
Move venv setup into its subdir. [Jiri Stransky]
Add license file – Apache 2.0. [Jiri Stransky]
Add vagrantized devstack for running tests against (for now) [Jiri Stransky]
I guess ideally we’d deploy something more stable like RDO TripleO Standalone, but i had devstack for now which i’ve written for Tmit project, so let’s start with that for now.
Generated Ansible collection skeleton. [Jiri Stransky]
Add virtualenv for locking Python deps. [Jiri Stransky]
Pip install in the virtualenv only works if the toolbox is started with –security-opt label=disable.
Makefile and a simple toolbox container. [Jiri Stransky]
Naming convention - underscores rather than hyphens. [Jiri Stransky]
Init. [Jiri Stransky]