Self-hosted Langfuse: tracing 7% of my AI agents, and ClickHouse logging itself

Description and use

The author self-hosted Langfuse to monitor AI agents but initially enabled tracing only on one profile, resulting in 7 % coverage and excessive ClickHouse disk usage. After configuring per‑profile keys, environment tags, and disabling system logs, coverage rose to 100 % and disk space was cut by nearly half.

Problem

Only 7% of AI agent traffic was traced and ClickHouse consumed 6 GiB of disk for 2.2 MiB of trace data, obscuring performance insights and wasting resources.

Outcome

Tracing coverage increased from 7 % to 100 % across profiles, enabling accurate latency and cost analysis per role. ClickHouse disk usage dropped by ~50 % after disabling unnecessary system logs, improving I/O performance.

Workflow

• 1. Copy Langfuse API keys into each profile’s secret scope. 2. Enable the Langfuse tracing plugin for each profile. 3. Set the Langfuse environment to the profile name. 4. Restart the gateway and send a probe trace from each profile to verify tagging. 5. Create a ClickHouse config.d file that removes twelve system log tables and sets logger level to warning. 6. Mount the config file read‑only into the container and recreate it. 7. Verify container health, unchanged trace_log row count, and intact trace data.
• 8. Run a weekly script that compares session records with Langfuse traces to generate role‑based reports.

Evidence

Langfuse held 618 events across 79 traces in four weeks. That's about three traces a day.

Source

ClickHouse was using 6.03 GiB of disk. The actual Langfuse trace data in it was 2.2 MiB.

Source

The tracing plugin is enabled per profile, and it reads its Langfuse API keys through that profile's own secret scope.

Source

ClickHouse lets you override server config with a file dropped into /etc/clickhouse-server/config.d/. This one removes twelve system log tables...

Source