Need to reintroduce row insertion optimization
Arkatufus opened this issue · comments
Gregorius Soedharmo commented
So, the purpose of ConsumeSequenceForTagInsert
was as follows:
Because we are batching all of our writes, but bulk copy is -much- faster than row-by-row, the idea was that in cases of:
ItemWithTags ItemWithTags ItemNoTags ItemNoTags ItemNoTags ItemNoTags ItemWithTags ItemNoTags
etc... we would still be able to take advantage of bulk-copy for the no-tag cases, and at least lessen the write throughput penalty.
Originally posted by @to11mtm in #138 (comment)