agent: Add per-VM metric for desired CU(s) #1108
Open
+123
−9
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
No changes to the coverage.
HTML Report |
sharnoff
force-pushed
the
sharnoff/scaling-event-reporting-2
branch
from
693b601 to
a3cf0fa
Compare
sharnoff
force-pushed
the
sharnoff/agent-desired-cu-metrics
branch
from
c9f1d75 to
fe3b019
Compare
sharnoff
force-pushed
the
sharnoff/scaling-event-reporting-2
branch
from
a3cf0fa to
16c0917
Compare
sharnoff
force-pushed
the
sharnoff/agent-desired-cu-metrics
branch
from
fe3b019 to
244473b
Compare
sharnoff
force-pushed
the
sharnoff/scaling-event-reporting-2
branch
from
16c0917 to
d2b4d45
Compare
sharnoff
force-pushed
the
sharnoff/agent-desired-cu-metrics
branch
from
244473b to
8008093
Compare
This commit adds a new per-VM metric: autoscaling_vm_desired_cu. It's based on the same "desired CU" information exposed by the scaling event reporting, but updated continuously instead of being rate limited to avoid spamming our reporting. The metric has the same base labels as the other per-VM metrics, with the addition of the "reason" label, which is one of: * "total" - the goal CU, after taking the maximum of the individual parts and rounding up to the next unit. * "cpu" - goal CU size in order to fit the current CPU usage * "mem" - goal CU size in order to fit the current memory usage, which includes some assesssment * "lfc" - goal CU size in order to fit the estimated working set size All of these values are also multiplied by the same Compute Unit factor as with the normal scaling event reporting, so that Neon's fractional compute units are exposed as such in the metrics, even as we use integer compute units in the autoscaler-agent. Also note that all values except "total" are NOT rounded, and instead show the fractional amounts to allow better comparison. KNOWN LIMITATION: If ReportDesiredScaling is disabled at runtime for a particular VM, the metrics will not be cleared, and instead will just cease to be updated. I figured this is a reasonable trade-off for simplicity.
sharnoff
force-pushed
the
sharnoff/scaling-event-reporting-2
branch
from
d2b4d45 to
8c60b7f
Compare
sharnoff
force-pushed
the
sharnoff/agent-desired-cu-metrics
branch
from
8008093 to
8639da0
Compare
Omrigan
reviewed
Dec 20, 2024
| // that when we set the desired CU from internal information, we can check whether the VM still | ||
| // exists. | ||
| // Otherwise it's not possible to prevent data races that would result in leaking metric labels. | ||
| activeMu sync.Mutex |
There was a problem hiding this comment.
Why did this become needed only now, how desiredCU is different from other metrics we already had?
There was a problem hiding this comment.
IIRC, it wasn't needed before because the metrics were all derived from the state of the VM, whereas desiredCU is derived from our internal state.
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.
This commit adds a new per-VM metric:
autoscaling_vm_desired_cu.It's based on the same "desired CU" information exposed by the scaling event reporting, but updated continuously instead of being rate limited to avoid spamming our reporting.
The metric has the same base labels as the other per-VM metrics, with the addition of the "reason" label, which is one of:
total- the goal CU, after taking the maximum of the individual parts and rounding up to the next unit.cpu- goal CU size in order to fit the current CPU usagemem- goal CU size in order to fit the current memory usage (including some information derived from LFC, to make sure there's room for cache too)lfc- goal CU size in order to fit the estimated working set sizeAll of these values are also multiplied by the same Compute Unit factor as with the normal scaling event reporting, so that Neon's fractional compute units are exposed as such in the metrics, even as we use integer compute units in the autoscaler-agent.
Also note that all values except "total" are NOT rounded, and instead show the fractional amounts to allow better comparison.
KNOWN LIMITATION: If
ReportDesiredScalingis disabled at runtime for a particular VM, the metrics will not be cleared, and instead will just cease to be updated. I figured this is a reasonable trade-off for simplicity.Notes for review: Tested this locally with the following patch to vm-deploy.yaml:
AFAICT it works as intended, but metrics are sometimes tricky. I plan to test it on staging before merging.
Also note: This PR builds on #1107 and must not be merged before it.