Consider the following setup:

CREATE TABLE tbl (
	id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
	text TEXT
) WITH (ttl_expire_after = '30 days');
ALTER TABLE tbl SPLIT AT VALUES (gen_random_uuid());
CREATE TABLE tbl2 (id INT PRIMARY KEY);
ALTER TABLE tbl2 SPLIT AT VALUES (1);

In this case, there may be a range that is valid between /tbl1/gen_random_uuid() and /tbl2/1. If this is the case, a TTL job may fail with a error decoding X bytes message. The impact here is that TTL does not succeed and rows do not get cleaned up (foreground traffic still occurs).

In practice, this would only happen if:

• the table is sufficiently large.
• there are multiple tables in the database
• usually only the last range on the table is affected - other rows are TTL'd

Jira issue: CRDB-15299