[DocDB] The latency to update rows on colocated tables is higher than hash partitioned tables.

[shantanugupta-yb]

Jira Link: DB-13948

Description

Below are some of high level observation post comparing the flamegraphs for updating 1M rows from hash and colocated tables:

• In colocated tables we are spending lot more time in below stacks:
  • yb::docdb::DocRowwiseIteratorBase::GetTableTombstoneTime
  • yb::docdb::DocRowwiseIterator::FetchNextImpl<yb::dockv::PgTableRow*>
• Also in hash partitioned tables yb::docdb::(anonymous namespace)::FilteringIterator::FetchTuple is called from yb::docdb::PgsqlReadOperation::ExecuteBatchKeys where as in colocated it is called from yb::docdb::(anonymous namespace)::FetchTableRow

Perf flamegraph for updating 1M rows on hash partitioned tables
hnode2.svg

Perf flamegraph for updating 1M rows on colocated tables
node2.svg

Issue Type

kind/bug

Warning: Please confirm that this issue does not contain any sensitive information

• I confirm this issue does not contain any sensitive information.