scheduler: Ingore instance's consumed resources on resize #203
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mariusleu
reviewed
Jun 3, 2021
mariusleu
previously approved these changes
Jun 3, 2021
grandchild
suggested changes
Jun 3, 2021
joker-at-work
force-pushed
the
ignore-consumer-on-resize
branch
from
b9d477a to
4dacb86
Compare
grandchild
previously approved these changes
Jun 3, 2021
|
Found out the depending patch with |
Instead of passing the "_nova_check_type" scheduler-hint on in the dictionary containing the "filter_properties", we set the "scheduler_hints" attribute on the request-spec object instead. This is necessary, because we will look at this object later with nova.scheduler.utils.request_is_resize and is in line with how a rebuild is marked in Nova. Passing "_nova_check_type" through the "scheduler_hint" parameter doesn't work, because only certain values from this dictionary are later on used by the conductor and we thus wouldn't recognize a resize at all. Change-Id: If4aa0c29ef04e7b4b9a9baef40206b4a553ef415
This function is similar to set_and_clear_allocations in that it's meant to update the instance's allocations at the same time as the migration's allocations, making it an atomic update. The difference is, that this function can update the migration's allocations with the supplied data, which is supposed to be used in same-host resizes, where we don't have to reserve all the resources twice. Change-Id: I85ed3611392e6ed9521465dadb96605220863cda
joker-at-work
force-pushed
the
ignore-consumer-on-resize
branch
from
4dacb86 to
521b5e8
Compare
|
This is a major extension of the previous code, as the testing showed that the previous code only went maybe half the way. The commit-message was extended, too and should explain everything (hopefully). If not, please ask. |
|
Just quickly: Typo in commit message: |
mariusleu
reviewed
Jun 14, 2021
When getting an allocation-candidate from placement, we want to ignore the currently-used resources of the VM we're resizing. Ignoring the resources makes it possible to resize on the same host, even if the VM would only fit there once. Since the resource allocations have already moved to the Migration object, we use its UUID in the "ignore_consumer" parameter when calling /allocation_candidates in placement. But since the resources are claimed by the Migration now, the allocation-requests returned by placement can still not fit on the host. Therefore, we not only claim those resources, but also update the Migration's resources at the same time. We don't remove all resource-allocations from the Migration, because we need to keep resources reserved for a possible revert by the customer. Therefore, we update the Migration's resources to keep the difference between its resources and the instance's resources allocated. One exception is DISK_GB, which needs to be kept allocated, as the disk will be copied and thus allocated twice. On reverting, we have lost some information that we didn't put into the Migration's allocations, because the instance had more. Therefore, we use the method of "nova-manage placement heal_allocations" to re-create the alloations for the original flavor. This might be problematic, if the instance had allocations against multiple providers. Since we're ignoring the whole Migration consumer in placement, double-spending on DISK_GB is not computed in and might make the resize fail if the host is low in DISK_GB. Additionally, the resize might fail if the host is overprovisioned, even if we patch placement to allow swapping resources of consumers. We might be able to accomodate for that in a future patch, though. Change-Id: Ie068a4a22edc37aa1e1173be9b2e7823fd3c5890
joker-at-work
force-pushed
the
ignore-consumer-on-resize
branch
from
521b5e8 to
844369c
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When getting an allocation-candidate from placement, we don't want to
double-spend the resources on the same host, because the VM will only
run once. Therefore, we pass "ignore_consumer" for the current
instance_uuid to placement's allocation_candidates/ endpoint. This has
the advantage, that we can resize on the same host, even if there are
only enough resources available for fitting in the VM once and don't
have to explain to customers that a resize is not possible, because
there are not enough resources available.
Change-Id: Ie068a4a22edc37aa1e1173be9b2e7823fd3c5890