refactor uncore event collection and handling

[harp-intel]

Expanding uncore events to per-device events creates a very long list of event groups. In some cases the list is too large for perf to handle resulting in an "Argument list too long" error from perf. We have worked around this in the past by abbreviating event names to shorten the argument list.

Through investigating the problem above, we found that the individual uncore device event counters are not used independently. They are added together, and the sum is used. The original intent of uncore event expansion is unknown.

In this PR, we eliminate the expansion of uncore events.

Copilot Description Follows

This pull request refactors the metrics event processing pipeline to simplify event handling and improve aggregation logic, especially for uncore events. It removes legacy abbreviation and group expansion logic, consolidates event bucketing and aggregation steps, and clarifies group assignment for metric variables. These changes streamline the codebase and make the event handling more explicit and maintainable.

Event bucketing and aggregation improvements:

• Replaced the coalesceEvents function with bucketEvents and added aggregateUncoreEvents to sum and deduplicate uncore events at socket granularity, removing the need for post-processing uncore group collapsing. (cmd/metrics/event_frame.go) [1] [2]
• Removed the legacy collapseUncoreGroupsInFrame and related helper functions, as uncore event aggregation is now handled earlier in the pipeline. (cmd/metrics/event_frame.go)

Group assignment and parsing changes:

• Added assignEventsToGroups to explicitly assign events to groups based on event group definitions after aggregation, separating parsing and grouping logic. (cmd/metrics/event_frame.go) [1] [2]
• Simplified parseEvents to only unmarshal JSON and parse values, removing abbreviation and group assignment logic from this step. (cmd/metrics/event_frame.go) [1] [2]

Removal of legacy abbreviation and expansion logic:

• Removed event name abbreviation and uncore group expansion functions, as these are no longer needed with the new aggregation approach. (cmd/metrics/event_defs.go, cmd/metrics/metric_defs.go) [1] [2] [3] [4]

Metric variable assignment clarification:

• Improved error handling and clarified logic for assigning metric variable values, ensuring correct group and variable name mapping, and adjusting C-state residency only when appropriate. (cmd/metrics/metric.go)

Dependency cleanup:

• Removed unused imports such as regexp and slices from cmd/metrics/event_defs.go.

[Copilot]

Pull Request Overview

This PR refactors the uncore event collection and handling pipeline to eliminate per-device event expansion, which was causing "Argument list too long" errors with perf. The refactoring simplifies event processing by aggregating uncore events at collection time rather than expanding and later collapsing them.

• Removes uncore event expansion and abbreviation logic that was previously used to manage long argument lists
• Introduces event bucketing and aggregation functions to handle uncore events more efficiently
• Separates event parsing from group assignment for clearer code organization

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 4 comments.

File	Description
cmd/metrics/event_defs.go	Removes event name abbreviation and uncore group expansion functions, along with unused imports
cmd/metrics/event_frame.go	Replaces coalesceEvents with bucketEvents and adds aggregateUncoreEvents function; removes legacy uncore group collapsing logic
cmd/metrics/metric.go	Improves error handling for metric variable assignment and clarifies C-state residency adjustment logic
cmd/metrics/metric_defs.go	Removes abbreviation of event names in metric expressions

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}