#7115 flytestdlib: add additional storage cache metrics and implement freecache prom collector Pull request opened by xjerod Adds Prometheus metrics to

flytestdlib/storage

to help identify Flytepropeller memory consumption from the in-memory data cache and large inputs/outputs — the two main controllable sources of propeller OOM pressure. •

rawstores.go

: Scopes the

cache

,

proto

, and

stow

metric groups into named sub-scopes so their metrics are easier to distinguish in dashboards. •

protobuf_store.go

: Adds

read_bytes_total

and

written_bytes_total

counters to track byte volume moving through the protobuf store, surfacing executions that pass large data between tasks. •

cached_rawstore.go

: Implements the freecache

prometheus.Collector

(resolving a long-standing TODO), exposing snapshot metrics scraped each Prometheus interval: •

entry_count

(gauge) — current occupancy of the in-memory cache •

evacuate_count_total

— evictions due to memory pressure; the primary signal that the cache is sized too small relative to workload •

overwrite_count_total

/

expired_count_total

— entry churn Also adds

cache_read_bytes_total

and

cache_write_bytes_total

counters to track cache throughput. flyteorg/flyte