System information

Bor client version: v2.7.0 (also affects latest main)

OS & Version: Linux (Kubernetes)

Environment: Polygon Mainnet

Type of node: Sentry

Overview of the problem

All Prometheus histogram metrics using ResettingSample have broken _count and _sum values. They decrease on every scrape instead of monotonically increasing, which violates the Prometheus counter type spec and breaks rate(),
increase(), and average latency calculations.

https://prometheus.io/docs/concepts/metric_types/#counter

A counter is a cumulative metric that represents a single monotonically increasing counter whose value can only increase or be reset to zero on restart.

Root cause: ResettingSample.Snapshot() in metrics/resetting_sample.go calls Clear() which resets count and sum to 0. The Prometheus collector in metrics/prometheus/collector.go emits these reset values as counter type,
but they only contain the delta since the last scrape — not cumulative totals.

Affected files and metrics:

Expected: _count and _sum should be monotonically increasing as required by the Prometheus counter type spec.

Actual: Both reset to interval-only values on every scrape, causing:

• rate() returns incorrect results or no data
• increase() is unreliable
• Average latency calculation (rate(_sum) / rate(_count)) is broken

Reproduction Steps

1. Enable telemetry with metrics = true and prometheus-addr = "0.0.0.0:7071"
2. Send RPC requests (e.g. eth_blockNumber)
3. Scrape /debug/metrics/prometheus twice with 1 minute interval
4. Observe rpc_duration_eth_blockNumber_success_count and _sum values decrease on second scrape

Logs / Traces / Output / Error Messages

Scrape at T1 (100 requests since startup)

rpc_duration_eth_blockNumber_success_count 100
rpc_duration_eth_blockNumber_success_sum 5000000

Scrape at T2 (50 requests since T1)

rpc_duration_eth_blockNumber_success_count 50 ← decreased
rpc_duration_eth_blockNumber_success_sum 2500000 ← decreased

Code path:

1. rpc/metrics.go:46 — creates histogram with ResettingSample
2. metrics/resetting_sample.go:22 — Snapshot() calls Clear()
3. metrics/sample.go:165 — Clear() resets count = 0, sum = 0
4. metrics/prometheus/collector.go:98-107 — emits snapshot values as Prometheus counter