Skip to content

Database delete

Delete is permanent

Deleting data from a database is permanent. It is recommended that a backup of the database is taken before performing a delete. Undoing a delete is only possible in certain circumstances, see Undoing a delete below for more details.

Beta Feature

Deleting data from a database is currently in beta. Deleting data already in the object store tier is not supported. When the date interval defined by startTS/endTS overlaps with dates in object store, the whole delete action is rejected.

Performing a delete

A database delete can be performed by REST directly on the Storage Manager.


In the example below, $SM is the hostname and port of the Storage Manager process on which to perform the delete operation.

curl -X POST "$SM/delete" \
    -H 'Content-Type: application/json' \
    -d '{"session":"","table":"","startTS":"","endTS":"","filter":[]}'

Example success response

HTTP/1.1 200
Content-Type: application/json
Connection: close
Content-Length: 99


Example failure response

HTTP/1.1 400
Content-Type: application/json
Connection: close
Content-Length: 48



name required type description
session no string An optional delete session name which can be used to undo the delete. See Undoing a delete for more details. If a session name is not provided, one will be generated and returned.
table yes string The name of the table to perform this delete operation on.
startTS no timestamp The start boundary for performing this delete operation. See endTS.
endTS no timestamp The end boundary for performing this delete operation. See below for how to specify a timestamp. Since we do not support deleting from the object store tier, the time window must restrict the delete to the dates in the local HDB tiers. A time window that overlaps with dates in object store will be rejected with an error message.
filter no list A list of triples that can be used for selecting specific data to be deleted. If a filter is provided, any data that matches the filter will be deleted. If no filter is provided, all data between the start and end timestamps will be deleted. See filtering below for more details.

Specifying timestamps

For getting the start and end times using bash, you may use date -u to print formatted dates:

# Define a time interval within the hour
startTS=$(date -u '+%Y.%m.%dD%H:00:00')
endTS=$(date -u '+%Y.%m.%dD%H:%M%:%S')

Times may be extended to contain a variable number of trailing digits to specify milliseconds and nanosecond precision.


The filter parameter is similar to the one used for querying. See Get Data filtering for details.

Undoing a delete

Performing a delete is an immediate action that should be considered irreversible. It is recommended that you create a backup of your database before performing a delete, but besides the backup, all RT logs since the time of issuing the delete must also be available for a successful restoration of all data.

When a delete is performed, it is sequenced in the source stream's control messages. If you try to undo a delete by restoring your database from a backup, the delete will be replayed and the matching data will be deleted again. To avoid this during a data recovery, the session of the delete can be ignored so that the delete is not performed and the data will remain. To ignore a delete session during a recovery operation, set the environment variable KXI_IGNORE_DELETE=<session>[,<session>,...].