Delete data

Row-level deletion is an InfluxDB 3 Enterprise feature that requires the storage engine upgrade (--use-pacha-tree). Use it to delete rows from a database by time range and an optional tag predicate.

How row deletion works

Row deletion is asynchronous. When you run influxdb3 delete rows, InfluxDB 3 Enterprise records a delete request and persists it to object storage. The compactor applies the request the next time it rewrites the affected run sets.

By default, deletes don’t apply for up to 24 hours after you submit the request. This delay is tunable with the --pt-row-delete-min-age server flag.

Because deletes are applied during compaction, rows remain queryable until the compactor processes the request. Don’t expect rows to disappear immediately after the command returns.

Delete rows

Use the influxdb3 delete rows command to delete rows from a database. For the complete command syntax and flags, see the influxdb3 delete rows CLI reference.

Specify a time range

A delete request must include an explicit time scope. Use one of the following:

  • --min-time: start of the range (inclusive)
  • --max-time: end of the range (exclusive)
  • --all-time: delete rows across all time

Filter with a tag predicate

You can narrow a delete to rows that match a tag predicate. Predicates support tag equality only:

  • Compare a tag to a string literal–for example, region = "us-west".
  • Combine conditions with AND only.
  • OR, NOT, and IN operators, and field columns, are not supported.

Required permissions

Deleting rows from a database requires the db:<DATABASE_NAME>:delete token permission. Deleting rows from the _internal database requires an admin token.

Cancel a pending delete

Use the influxdb3 cancel row-delete command to cancel a pending delete request before the compactor applies it.

InfluxDB 3 Enterprise allows at most 1000 pending delete requests across the cluster.

Monitor row deletes

Query the system.row_deletes system table to review the status of delete requests:

SELECT * FROM system.row_deletes

InfluxDB 3 Enterprise also emits nine influxdb3_compactor_row_delete_* metrics that report row-deletion activity in the compactor.

A dedicated system-tables reference is planned as a follow-up. Until then, this guide documents the system.row_deletes table.

Known issues

Ghost rows after a completed delete

Rows in the un-compacted ingest tail can survive after a delete request reports completed. If you observe rows that should have been deleted, re-issue the delete after the affected data compacts, then verify row counts.

system.row_deletes can return HTTP 500

Querying system.row_deletes after an --all-time delete that has no tag predicate can return an HTTP 500 error. As a workaround, use the GET /api/v3/row_delete_requests API endpoint to review delete requests instead.

Was this page helpful?

Thank you for your feedback!