Add configurable size_layers and size_derived_metrics for layer-specific size metric handling

[izzet]

This pull request introduces a more flexible system for handling "size" metrics and layers in the analyzer configuration and processing logic. The main improvements are the ability to specify which layers and metrics should be treated as "size" related, and to propagate these settings throughout the analyzer pipeline. This enhances configurability for different storage backends and analysis presets.

Configuration enhancements:

• Added size_derived_metrics and size_layers fields to the AnalyzerPresetConfig dataclass, with appropriate defaults for POSIX and DLIO presets, allowing each preset to specify which layers and metrics are considered "size"-related. [1] [2] [3]
• Introduced DERIVED_POSIX_SIZE_METRICS constant to centralize the list of metrics considered as size-related for POSIX.

Analyzer logic updates:

• Modified the set_layer_metrics method in analyzer.py to accept a size_derived_metrics argument, and use it to determine which metrics should be treated as size-related, improving flexibility and reducing hardcoding.
• Updated the _compute_main_view method to use the new configuration fields, ensuring that size columns and metrics are handled correctly based on the current layer and preset.

Bug fix / data extraction improvement:

• Fixed the logic for extracting image_size in io_function to only assign the size field when the operation name does not include "open", ensuring more accurate data attribution.

[Copilot]

Pull request overview

This PR makes “size” metric handling configurable per analyzer preset and propagates that configuration through the analyzer pipeline, while also improving image_size extraction in IO event parsing.

Changes:

• Added preset config fields size_layers and size_derived_metrics (with POSIX/DLIO defaults) and centralized POSIX size-derived metric names.
• Updated analyzer logic to keep/drop size columns by configured layers and to treat only configured derived metrics as “size” metrics.
• Adjusted IO parsing to populate size from image_size only for non-open operations.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
python/dftracer/analyzer/dftracer.py	Updates image_size → size attribution logic during IO event parsing.
python/dftracer/analyzer/config.py	Introduces configurable size layer/metric preset fields and a POSIX size metrics constant.
python/dftracer/analyzer/analyzer.py	Threads new size config into main view computation and derived metric column creation.

Comments suppressed due to low confidence (1)

python/dftracer/analyzer/config.py:1

• These fields are typed as Optional[...] but default to empty containers via default_factory, so they’ll never be None unless explicitly passed as None. This makes downstream code more complex (extra or {} / or []) and makes the API ambiguous. Prefer either (a) making them non-optional (Dict[...] / List[...]) with the current defaults, or (b) defaulting to None without default_factory if None is a meaningful “unset” state.

import dataclasses as dc

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[rayandrew]

looks good to me

[rayandrew]

@izzet let me know if i want to merge this first or u still need to test using Karim's new traces

[izzet]

I am merging this @rayandrew thanks