ClickHouse no longer fills the disk with its own logs

Proxus 2.0.2 fixes the ClickHouse settings that ship with the installation. The application itself is unchanged, and all fixes from 2.0.1 are included.

Fixes

• ClickHouse no longer fills the disk with its own logs. The shipped settings turned off most of the database server's internal log tables but left the two largest, query_log and text_log, on. Proxus reads none of them. On one customer installation these tables reached 277 GB while telemetry used 21 GB, and query_log grew by about 9 GB a day. Now query_log, text_log, query_metric_log, latency_log and asynchronous_insert_log are off as well, in Docker Compose, Kubernetes (local and HA) and Siemens Industrial Edge installations. The Kubernetes HA installation turned off no internal logs at all; it now uses the same list as Docker.

Upgrade notes

• The setting is in the compose file, not the image. Installations updated with the installer get the new file. If you manage the compose file yourself, add these five lines below <backup_log remove="1"/> in the ClickHouse section and restart only ClickHouse (docker compose up -d clickhouse). Writes wait a few seconds and no data is lost: <query_log remove="1"/>, <text_log remove="1"/>, <query_metric_log remove="1"/>, <latency_log remove="1"/>, <asynchronous_insert_log remove="1"/>.
• Remove the logs already collected (optional). List them with SELECT name, formatReadableSize(total_bytes) FROM system.tables WHERE database='system' AND engine='MergeTree' AND match(name, '_log(_[0-9]+)?$'), then run DROP TABLE system.<name> SYNC for each. Telemetry tables are not touched; ClickHouse recreates the small ones itself if needed.
• From 2.0.0. Set the images to 2.0.2; the 2.0.1 fixes are included. See Upgrade.