Open yugabyte-ci opened 3 months ago
Looks like the same external_hybrid_time
is set on every write_pair
and on the write_batch
causing more space usage on target WAL.
This only happens in YCQL direct write batches. Transactional writes use intents which do not set the external ht. YSQL uses transactional writes only.
Fix would be to skip the ht if the write batch had the same value. It won't handle all cases though
The Target write batch has 98 kv pairs. Each 7 of these has one external HT. The batch has 14 HTs in total. Today we record the HT in each one of the 98 records. The fix would be to remove the HT on the kvs and break the batch per HT.
Jira Link: DB-11828
WAL size on xCluster target universe is 20% higher than the source universe