rockeet

In most cases, secondary index values are empty, thus `rocks_iter->value()` is not needed in such cases. We realized an SST with **lazy load** of value, this is a big improvement...

Rdb_iterator_base::next_with_direction: too many compares for eof check

2

in `Rdb_iterator_base::next_with_direction`, it checks prefix and lower bound and upper bound, this consuming too many cpu in our cases, when running tpcc, the flame graph showing these 3 compare takes...

In tpcc, optimize `bmsql_history` immediately return and has no any effect, which `bmsql_history` has a hidden pk, no primary index and any other index.

Currently creating index by `alter table add index ...` will scan table one pass for each index, this waste cpu time if there are multi indexes are created in one...

Feature Request: Create Index in Parallel

3

`inplace_populate_sk` scans tables, if a table is too large, this consumes too much time. The table can be divided into multiple parts, one thread per-part, to speed up the scaning....

Commit sst in dedicated threads

4

Execute `Rdb_sst_info::commit_sst_file` in dedicated threads, this improves performance: 1. `Rdb_sst_file_ordered::commit` may use stack to reverse input data, this is time consuming 2. `m_sst_file_writer->Finish` may be consuming, at least it need...

The `else` conditional compile branch of `HAVE_POLL` has a redundant `{`. This PR fix this issue

m_auto_incr_map: reduce a redundant search operation

1

This PR reduced redundant searchs on `m_auto_incr_map`.

properties_collector: (un)materialize: add missing m_name

2

This PR bump stats version to `INDEX_STATS_VERSION_WITH_NAME = 3` for added `m_name`.

Add refresh_iter to release pinned MemTable and SST

2

During long running trx, there is a rocksdb::Version object is referenced by iterator, thus the **MemTable**s and **SST**s referenced by **Version** can not be released even the **MemTable**s are flushed...