CA-320203: Fixed the update wizard failure when installing two cumulative updates
The issue was that when installing 2 new versions in one go (e.g. two cumulative updates), the installation succeeds on master but it cannot continue on slaves, because it cannot find the first CU update anymore.
The fix is to refresh the update record before applying the update, with the record found in the cache. Also, if there is no update in the cache, try to re-introduce the update.
Signed-off-by: Mihaela Stoica <mihaela.stoica@citrix.com>
If a host needs to be rebooted after the update application, and the update is not a live patch, then do the host evacuation before applying the update.
Signed-off-by: Mihaela Stoica <mihaela.stoica@citrix.com>
* Simplification: use auto-properties.
* Removed unnecessary null check. Converted private methods to linq method chain.
* Removed unused file. Added null check.
* Corrected property names as it was difficult to distinguish between the alert,
the patch and the filepath.
* Removed duplicate lock.
* CA-296490: Rewrote the logic uploading single updates so as to facilitate uploading to multiple pools in parallel.
* Corrections as per code review.
* Account for the case of uploading from disk legacy updates that do not correspond to an update alert.
Removed unfriendly error messages (the failure is written out in the logs anyway).
* Wizard test corrections and some modernising:
- Only call cancel on the last page if the test allows us to.
- Do finish the UpdatesWizard test.
- Finish the RunWizardKeyboardTests using key strokes instead of button clicks.
- Ensure the Cancel/Previous buttons do not have focus when sending an Enter stroke.
- Use optional parameters in the WizardTest constructor and string interpolation.
* Renamed badly named property.
* Improved code cloning VMs by assigning new opaque_refs to cloned VBDs.
Compacted test object provider: use optional parameters instead of method overloads.
* Modifications to prevent multiple update of enabled state and ensure the wizard
buttons are updated last.
Signed-off-by: Konstantina Chremmou <konstantina.chremmou@citrix.com>
Also sort Messages.resx and call ReplaceProgressStep instead of AddProgressStep in the DownloadPatchPlanAction class (when the download is skipped)
Signed-off-by: Mihaela Stoica <mihaela.stoica@citrix.com>
* Deprecate set but unused Status property in favour of ProgressDescription.
* Deprecate set but unused TitlePlan property.
* Deprecated property Title in favour of ProgressDescription.
* Removed unused plan RebootHostPlanAction.
* Updated event declaration.
* Removed property Visible. The plan is to have the actions log their history as soon as some is available.
* Reworked the logging for the plan actions. The main change is the replacement of
the ProgressDescription with a stack where all the steps of an action are stored,
i.e. the action is responsible for its history, not the worker running it.
* Lock the progress history before accessing it.
* Better handling of error logging and user cancellation.
Signed-off-by: Konstantina Chremmou <konstantina.chremmou@citrix.com>
prevent them from being serialised alongside the API properties. This will also
be useful for moving the API bindings out of XenModel.
Signed-off-by: Konstantina Chremmou <konstantina.chremmou@citrix.com>
In rare cases, the GetMaster could return null, causing the update progress to fail with "Pool_patch or Pool_update not found". (when the Cache hasn't been populated by the execution gets here)
This commit drops GetMaster and uses a pre-saved masterUuid instead of waiting for the Cache what would be the other way to solve this issue.
Signed-off-by: Gabor Apati-Nagy <gabor.apati-nagy@citrix.com>
Using new XAPI functionality in Install Update Wizard required by the new packaging format (ISO updates)
Signed-off-by: Gabor Apati-Nagy <gabor.apati-nagy@citrix.com>
Comparing uuids instead of references that likely will be changed after the Cache has been repopulated.
Signed-off-by: Gabor Apati-Nagy <gabor.apati-nagy@citrix.com>
The PatchPrechecksOnMultipleHostsInAPoolPlanAction did not filter the patchmappings by masterhost, so with some probability, it could choose the wrong mapping (pool_patch object for the same update, but in different pool). This bug is being fixed in this commit, other changes only renamed (for more consistent naming) a class and removed an unused field.
Signed-off-by: Gabor Apati-Nagy <gabor.apati-nagy@citrix.com>
