The same SQL statement, sql-client and sql-gateway give different execution plans

We have a SQL job, like this
select ... from prod_kafka.f_alarm_tag_dev /*+ OPTIONS('scan.startup.mode' = 'latest-offset') */ as f
left join mysql_bnpmp.gem_bnpmp.f_a_alarm_filter
/*+ OPTIONS('lookup.cache.max-rows' = '5000', 'lookup.cache.ttl' = '30s') */
FOR SYSTEM_TIME AS OF f.proctime ff  on ff.rule_type = 0 and f.ne_ip = ff.ip
We submit to flink 1.17.1 cluster with sql-gateway. We found that the execution plan ignores the filter condition (rule_type=0),  and we could not get the expected query results

   +- [1196]:LookupJoin(table=[mysql_bnpmp.gem_bnpmp.f_a_alarm_filter], joinType=[LeftOuterJoin], lookup=[ip=ne_ip], select=[event_id, severity

We submit same sql to flink 1.17.1 cluster with sql-client. We found that the execution plan contains the filter condition ( lookup=[rule_type=0, ip=ne_ip], where=[(rule_type = 0)] ), and we got the expected query results

   +- [3]:LookupJoin(table=[mysql_bnpmp.gem_bnpmp.f_a_alarm_filter], joinType=[LeftOuterJoin], lookup=[rule_type=0, ip=ne_ip], where=[(rule_type = 0)], select=[event_id, severity,

We think this issue is related to https://issues.apache.org/jira/browse/FLINK-32321