Using Time to Live (TTL) with primary (partition) key columns

[Guy]

Originally from the User Slack

@Daria_Fedorova: Hi! I am reading on TTL in scylla and one thing seems strange to me
If TTL is only defined for non-primary columns , it is not possible to define TTL for a table that has only primary columns ?

@Felipe_Cardeneti_Mendes: you can have TTL, but you can’t indeed use the ttl() function on a primary column, yes.

@Daria_Fedorova: do you by any chance know the reason for this ?

@Felipe_Cardeneti_Mendes: I know that historically Cassandra also lacked support for it - since a decade now https://issues.apache.org/jira/browse/CASSANDRA-9312

It’s a bit of an edge case tho. You can’t really update a primary key, so the TTL will always decrease relative to its writetime() . It is true you can’t use the writetime() function on a key column, but if that’s a necessary use case, you can easily add a non-key column which simply ingests a now() timeuuid. Then you can use either the ttl() function on top of it, or simply deduce from it if you have a fixed TTL

[Mihir]

For a table that has only primary columns. we insert a row with ttl. now how do we check what is the ttl associated with the row.

[Mihir]

I am talking about data inserted by this dml
INSERT INTO test.table (id, wsid) VALUES (‘test-rand-id’, ‘test-ws-id’) USING TTL 10;
and table created using this ddl
CREATE TABLE IF NOT EXISTS test.table ( id text, wsid text,PRIMARY KEY ((id, wsid))) WITH bloom_filter_fp_chance = 0.005;