Agent Dashboard — CH Buffer Peak
What this chart shows
CH Buffer Peak (per heartbeat) — the maximum number of copy-log rows held in the agent's in-memory copy-log buffer at any point during the last heartbeat window (~30 s + 0–10 s jitter). On each heartbeat the agent reports the high-water mark across the window, then zeroes the counter. The buffer accumulates per-object copy records before flushing them to ClickHouse in batches.
How the buffer works
Each time the copy engine records a copied or skipped object, the record is added to an in-memory buffer rather than written to ClickHouse immediately. The buffer flushes when:
- It reaches the configured batch size threshold, or
- A periodic flush timer fires
This chart shows the peak depth seen during each heartbeat window — a leading indicator of flush frequency and memory pressure.
How to read it
| Value range | Interpretation |
|---|---|
| 0–500 | Normal — buffer flushing regularly |
| 500–2000 | Active copy job with high object throughput |
| > 2000 | High accumulation — flushes may be falling behind; ClickHouse write latency may be elevated |
| Sustained high + not dropping | Possible ClickHouse connectivity issue causing buffer to fill without flushing |
Why it matters for resume and Smart Skip
Copy log records written to ClickHouse are the baseline used by Smart Skip deduplication on resume. If a job crashes before a batch flushes, those objects are not in the copy_log and Smart Skip cannot skip them on the next run — they will be re-copied. Monitoring buffer depth helps understand the maximum number of objects that could be re-copied after a crash.
Polling interval
Sampled on every heartbeat (~30 s + 0–10 s jitter). The value is the peak depth observed during that window, not the depth at the instant of sampling — so a single brief spike between heartbeats will still appear on the chart.