Shulou(Shulou.com)06/01 Report--

Slave_rows_search_algorithms consists of a combination of three values: TABLE_SCAN,INDEX_SCAN, HASH_SCAN.

TABLE_SCAN,INDEX_SCAN (default configuration, which means index is used if there is an index, otherwise a full table scan is used)

HASH_SCAN can partially solve the replication delay problem caused by non-primary key tables.

When there is no primary key or unique key on the table, then for the DML done on the table, if it is replicated in ROW mode, it is possible for each row to be mirrored in the standby database before recording.

To generate a full table scan (or a secondary index scan), in most cases, this overhead is very unacceptable and can cause a lot of latency.

The realization method of hash_scan

To put it simply, in apply rows_log_event, updates to rows in log_event are cached in two structures, respectively

Yes: m_hash, m_distinct_key_list. M_hash: mainly used to cache the starting position of updated row records

Is a hash table; m_distinct_key_list: push the value of the index to m_distinct_key_list if there is an index, if the table does not have an index

This List structure is not used; the whole call process of the pre-scan is as follows: Log_event::apply_event

Rows_log_event::do_apply_event

Rows_log_event::do_hash_scan_and_update

Rows_log_event::do_hash_row (add entry info of changed records)

If (m_key_index

< MAX_KEY) (index used instead of table scan) Rows_log_event::add_key_to_distinct_keyset () 当一个event 中包含多个行的更改时，会首先扫描所有的更改，将结果缓存到m_hash中，如果该表有索引，则将索引的值 缓存至m_distinct_key_list List 中，如果没有，则不使用这个缓存结构， 而直接进行全表扫描。 执行stack如下： #0 handler::ha_delete_row #1 0x0000000000a4192b in Delete_rows_log_event::do_exec_row #2 0x0000000000a3a9c8 in Rows_log_event::do_apply_row #3 0x0000000000a3c1f4 in Rows_log_event::do_scan_and_update #4 0x0000000000a3c5ef in Rows_log_event::do_hash_scan_and_update #5 0x0000000000a3d7f7 in Rows_log_event::do_apply_event #6 0x0000000000a28e3a in Log_event::apply_event #7 0x0000000000a8365f in apply_event_and_update_pos #8 0x0000000000a84764 in exec_relay_log_event #9 0x0000000000a89e97 in handle_slave_sql #10 0x0000000000e341c3 in pfs_spawn_thread #11 0x0000003a00a07851 in start_thread () #12 0x0000003a006e767d in clone () 执行过程说明： Rows_log_event::do_scan_and_update open_record_scan() do next_record_scan() if (m_key_index >

MAX_KEY)

Ha_rnd_next ()

Else

Ha_index_read_map (m_key from m_distinct_key_list)

Entry= masked hash-> get ()

Masked hash-> del (entry)

Do_apply_row ()

While (masked hash-> size > 0)

As can be seen from the execution process, when using hash_scan, the whole table will be scanned only once, although the m_hash hash table will be traversed many times, but this scan

It is O (1), so the cost is very small, so the number of scans can be reduced and the execution efficiency can be improved.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.