[Bug] event_store on_disk_data_size can overflow to EiB

[tenfyzhong]

Description

The Grafana panel "Data Size On Disk" occasionally shows EiB-scale spikes, which are far larger than the real data size.

Context

Panel query:
sum(ticdc_event_store_on_disk_data_size{...}) by (instance)

Metric source:
collectAndReportStoreMetrics calls diskSpaceUsage(stats) and exports
ticdc_event_store_on_disk_data_size.

In diskSpaceUsage, pebble.Metrics.Compact.InProgressBytes (int64) is cast
directly to uint64 and added:
usageBytes += uint64(m.Compact.InProgressBytes)

In Pebble, InProgressBytes is signed and may go negative when compactions finish.
If it becomes negative, the uint64 cast wraps to ~2^64, producing an EiB-scale value.

Relevant code:

• logservice/eventstore/event_store.go:1090-1103, 1138-1144
• pebble/metrics.go: Compact.InProgressBytes is int64

Expected vs Actual

Expected: on_disk_data_size reflects real on-disk bytes and never shows EiB spikes.
Actual: occasional EiB-level spikes on the panel.

Proposed Solution

Clamp Compact.InProgressBytes to 0 when it is negative (or otherwise ensure a non-negative
value) before casting to uint64 in diskSpaceUsage.

Screenshot