Skip to main content

Generating a support bundle

A support bundle collects logs, configuration and diagnostics from one cluster into a single archive you can hand to support. Collection is read-only, and one bundle is collected at a time.

Estimate before you collect​

A bundle can run to gigabytes. The estimate runs the same file selection the collector does, so the size it reports is the size you will get. The duration is a lower bound; collection can take longer. Estimate before committing to a wide time window.

Start from a scenario​

A scenario is a preset for a symptom. It selects the artifacts support usually asks for, so you do not have to choose them one by one.

For example:

  • Compute node crash gathers what it takes to triage a crash or a restart loop — output from the run before the crash, and the query that was in flight.
  • Query trace follows a single query ID across the coordinator logs, and needs you to supply the ID.

Others exist, and the set grows over time. The console carries the current list with a description of each; start there rather than from this page.

A scenario pre-selects a subset. It does not restrict what you can add, and everything it selects can be changed.

Reuse a selection with a bundle profile​

A bundle profile is a selection saved to a file: the artifacts, the pods, the time window and the search values. It carries no cluster name and no collection logic, so it is a way to hand someone an exact selection instead of describing one. The usual sender is support. Bundle profiles are available from v2.1.

Bundle profile sits at the top right of the Create support bundle page.

  • Save to file writes the current selection, with an optional note shown to whoever opens it. A half-finished selection will not save: a category ticked in one part and left empty in another is named so you can finish or clear it, because collection would otherwise drop it without saying so.
  • Load from file always opens a review step before anything changes. The review reports what the file asks for that this cluster does not have — a pod that is gone, a value not offered here — and names what needs a decision; a missing pod can be replaced with one of the same kind, or ignored. Applying replaces the whole selection, and collection still waits for Create bundle.

A profile holds a selection and nothing else, so loading one cannot collect anything the page could not already collect.

A file is refused outright, before any review, when it is not a bundle profile at all or was saved by a console too different from yours to read it. The message says which it is; where the answer is the version, ask whoever sent it to save it again on yours.

What gets collected​

A bundle can include, among other things:

  • Cluster definition and configuration files — how the cluster and its warehouses are set up
  • Log files and log search — everything in the window, or only the lines matching a search
  • Diagnostic queries — read-only queries describing cluster state, such as nodes and running jobs
  • Thread dumps and thread stacks — what coordinators and compute nodes were doing at the moment of collection
  • Query audit analysis — failed statements, and the heaviest queries in the window
  • Crash context and memory diagnostics — evidence for one specific class of failure
  • Metrics snapshot — the monitoring window, so support sees the same charts you do. It comes back empty if no Prometheus is wired to Anywhere, and nothing in the bundle says so — check the console's monitoring pages before collecting a bundle for anything performance-related

Getting the archive​

There are two ways to retrieve a finished bundle.

  • Download — the archive comes back through your existing session. Use it when you want to look at the bundle yourself, or send it to support manually.
  • Export link — a pre-signed URL with an expiry. Anyone holding the link can download the bundle, so treat it as a credential. Useful for sharing within your own network, or with support where outbound transfer is permitted by your policy.

Bundles are removed after 7 days.

What is masked, and what is not​

Credential values in manifests, in configuration files and in the coordinator configuration listing (commands/show-frontend-config.txt) are masked irreversibly.

Every other command output under commands/ is collected exactly as the cluster printed it. What a statement prints is the cluster's decision: DESC STORAGE VOLUME, SHOW CREATE CATALOG and SHOW ROUTINE LOAD mask their own secrets, and a custom command you saved yourself is never rewritten.

Audit logs contain user SQL verbatim and are deliberately not masked. Review a bundle before sending it outside your organisation.

Every bundle contains a safety-report.txt listing what ran and what the masking covered. It is generated from the artifacts that executed, so it describes that bundle rather than the product in general — see Reading a support bundle.