Traffic

The Traffic page answers a single question: how much data is flowing into this cluster? It combines last-24h ingestion KPIs, rows/bytes/insert-query volume over time, per-table ingestion, and — only when the cluster actually replicates, shards, or runs a PeerDB pipeline — the matching write-amplification and data-movement views.

What it does

Traffic reads two complementary data sources and is careful to keep them distinct, because they measure different things:

• system.query_log (Insert queries, type = 'QueryFinish', query_kind = 'Insert') — written_rows / written_bytes are the uncompressed payload as ingested. This is the "how much data did clients send us" number.
• system.part_log (NewPart events) — size_in_bytes is the on-disk (compressed) size of each new part. This is the "how much did we actually store" number. part_log is opt-in (see below), so every chart that reads it degrades gracefully when it is absent.

Comparing the two — uncompressed ingest vs on-disk bytes — is how you read the effective compression of incoming data over time.

Sections

KPI strip

A row of last-24h totals from system.query_log, each with a delta versus the previous 24 hours: rows ingested, bytes ingested (uncompressed), and insert queries, plus the cluster's overall compression ratio (uncompressed ÷ compressed across active parts, from system.parts).

Ingestion charts

A grid of time-series charts. Every chart has its own date-range selector for hour / day / month granularity.

Merges & Data Movement

Background write-amplification and data-movement throughput from system.part_log: merged bytes (MergeParts), part moves (MovePart), and a write-amplification view (how many bytes the server rewrites for every byte ingested). This section reads part_log and shows an empty state when it is not configured.

Top Tables by Ingestion (24h)

A table ranking tables by data ingested in the last 24 hours: rows added, on-disk bytes added, parts created (all rendered as background bars), and each table's overall compression ratio. Ingestion volume comes from system.part_log (NewPart), joined with the table's current compression from system.parts. Because it depends on part_log, the table is optional and shows an informative empty state when part_log is off.

Replication & Distribution (conditional)

This section only appears when the cluster actually replicates or shards. Traffic runs a smart cluster-shape probe against system.replicas / system.clusters first; on a single-node, non-replicated server the section is not rendered at all — no empty cards. When present, it shows:

• Replica fetches — bytes downloaded between replicas (part_log DownloadPart).
• Distributed queries — initial vs secondary query volume, so you can see fan-out from Distributed tables.

PeerDB Ingestion (conditional)

Also auto-detected: Traffic probes for a PeerDB replication pipeline (tables/rows written by PeerDB mirrors). The PeerDB Ingestion section renders only when PeerDB activity is detected, showing rows ingested through the PeerDB pipeline. If you run PeerDB, see the dedicated PeerDB feature for mirror, peer, and slot-level detail.

Enable system.part_log

The KPI strip, the two query-log charts, and the overall compression ratio work out of the box, because system.query_log and system.parts are on by default. The Bytes on Disk chart, Merges & Data Movement, Top Tables by Ingestion, and Replica fetches all read system.part_log, which is opt-in on some ClickHouse distributions. When it is absent, those views show an informative empty state instead of failing.

To enable it, add <part_log> to the ClickHouse server config and restart:

<clickhouse>
  <part_log>
    <database>system</database>
    <table>part_log</table>
    <flush_interval_milliseconds>7500</flush_interval_milliseconds>
  </part_log>
</clickhouse>

See Enable system tables for the full list and configuration.

Permissions & access

Traffic shares the metrics feature id.

CHM_FEATURE_METRICS_ACCESS=authenticated

# Disable entirely
CHM_FEATURE_METRICS_ENABLED=false
# or
CHM_DISABLED_FEATURES=metrics

# CHM_CONFIG_FILE (TOML)
[features.metrics]
access = "authenticated"

The global query timeout applies to all Traffic queries:

CLICKHOUSE_MAX_EXECUTION_TIME=60

Related