Support Bundle

scripts/collect-logs.sh (in the Bifract repository) gathers what support needs to diagnose a problem into one archive. It detects Docker Compose or Kubernetes on its own and only reads; it never restarts, deletes or changes anything.

Collect a bundle

1. Run it on the host as a user in the docker group (Compose), or with a kubeconfig that can exec into pods (Kubernetes):

   ./scripts/collect-logs.sh
   

2. Send the resulting bifract-support-<hostname>-<timestamp>.zip to support. Without zip, it writes a .tar.gz.

To look through the contents yourself, write a directory instead: ./scripts/collect-logs.sh --dir bifract-support. Start with SUMMARY.txt.

Option	Description
--dir <path>	Write a directory instead of an archive
--docker / --k8s	Force the platform
--namespace <ns>	Kubernetes namespace. Default bifract
--tail <n>	Log lines per container. Default 3000
--quick	Skip the 8 second second sample that tells stuck merges from moving ones

What it collects

Folder	Contents
host/	OS, CPU, memory, disk, load, OOM-killer evidence, listening ports
docker/ or kubernetes/	Container or pod inventory and describes, events, manifests, PVCs, network policies, operator logs
bifract/	Version, image tags and digests, reverse proxy config
logs/	Tail of every container log, an errors/ excerpt per component, and the previous instance of crash-looping containers
clickhouse/nodes/<node>/	Version, topology, disks, settings, metrics, merges, mutations, queues, parts, table DDL, migrations, errors, failed and slow queries, server log
clickhouse/consistency/	Cross-node checks: schema drift, migration numbers, stuck merges
postgres/	Version, migrations, settings, table sizes, object counts, alert and archive state, activity and locks

SUMMARY.txt is a one-page triage view, README.txt describes the layout, and collection-errors.txt lists anything that could not be collected.

Size and privacy

• Output is capped at 4 MB per command, 6 MB per container log and 60 log targets. A single-node bundle is usually a few hundred KB.
• ClickHouse queries time out after 45 seconds and read system tables, never logs.
• The run stops if the target filesystem has less than 1 GB free.
• Values under secret-looking keys (passwords, tokens, API and access keys, bearer tokens, credentials in URLs) become ***REDACTED***. Kubernetes secret values are never read.

Note

The bundle still holds operational data: container log lines, query text (first 300 characters), table and column names, fractal IDs and object counts. Review it before sending if that matters.