mysql lock_latency_Mysql 优化——分析表读写和sql效率问题

SELECT file_name AS file,

count_read,

sum_number_of_bytes_read AS total_read,

count_write,

sum_number_of_bytes_write AS total_written,

(sum_number_of_bytes_read + sum_number_of_bytes_write) AS total

FROM performance_schema.file_summary_by_instance

ORDER BY sum_number_of_bytes_read+ sum_number_of_bytes_write DESC;

-- 反映文件的延迟

SELECT (file_name) AS file,

count_star AS total,

CONCAT(ROUND(sum_timer_wait / 3600000000000000, 2), 'h') AS total_latency,

count_read,

CONCAT(ROUND(sum_timer_read / 1000000000000, 2), 's') AS read_latency,

count_write,

CONCAT(ROUND(sum_timer_write / 3600000000000000, 2), 'h')AS write_latency

FROM performance_schema.file_summary_by_instance

ORDER BY sum_timer_wait DESC;

-- table 的读写延迟

SELECT object_schema AS table_schema,

object_name AS table_name,

count_star AS total,

CONCAT(ROUND(sum_timer_wait / 3600000000000000, 2), 'h') as total_latency,

CONCAT(ROUND((sum_timer_wait / count_star) / 1000000, 2), 'us') AS avg_latency,

CONCAT(ROUND(max_timer_wait / 1000000000, 2), 'ms') AS max_latency

FROM performance_schema.objects_summary_global_by_type

ORDER BY sum_timer_wait DESC;

-- 查看表操作频度

SELECT object_schema AS table_schema,

object_name AS table_name,

count_star AS rows_io_total,

count_read AS rows_read,

count_write AS rows_write,

count_fetch AS rows_fetchs,

count_insert AS rows_inserts,

count_update AS rows_updates,

count_delete AS rows_deletes,

CONCAT(ROUND(sum_timer_fetch / 3600000000000000, 2), 'h') AS fetch_latency,

CONCAT(ROUND(sum_timer_insert / 3600000000000000, 2), 'h') AS insert_latency,

CONCAT(ROUND(sum_timer_update / 3600000000000000, 2), 'h') AS update_latency,

CONCAT(ROUND(sum_timer_delete / 3600000000000000, 2), 'h') AS delete_latency

FROM performance_schema.table_io_waits_summary_by_table

ORDER BY sum_timer_wait DESC ;

-- 索引状况

SELECT OBJECT_SCHEMA AS table_schema,

OBJECT_NAME AS table_name,

INDEX_NAME as index_name,

COUNT_FETCH AS rows_fetched,

CONCAT(ROUND(SUM_TIMER_FETCH / 3600000000000000, 2), 'h') AS select_latency,

COUNT_INSERT AS rows_inserted,

CONCAT(ROUND(SUM_TIMER_INSERT / 3600000000000000, 2), 'h') AS insert_latency,

COUNT_UPDATE AS rows_updated,

CONCAT(ROUND(SUM_TIMER_UPDATE / 3600000000000000, 2), 'h') AS update_latency,

COUNT_DELETE AS rows_deleted,

CONCAT(ROUND(SUM_TIMER_DELETE / 3600000000000000, 2), 'h')AS delete_latency

FROM performance_schema.table_io_waits_summary_by_index_usage

WHERE index_name IS NOT NULL

ORDER BY sum_timer_wait DESC;

-- 全表扫描情况

SELECT object_schema,

object_name,

count_read AS rows_full_scanned

FROM performance_schema.table_io_waits_summary_by_index_usage

WHERE index_name IS NULL

AND count_read > 0

ORDER BY count_read DESC;

-- 没有使用的index

SELECT object_schema,

object_name,

index_name

FROM performance_schema.table_io_waits_summary_by_index_usage

WHERE index_name IS NOT NULL

AND count_star = 0

AND object_schema not in  ('mysql','v_monitor')

AND index_name <> 'PRIMARY'

ORDER BY object_schema, object_name;

-- 糟糕的sql问题摘要

SELECT (DIGEST_TEXT) AS query,

SCHEMA_NAME AS db,

IF(SUM_NO_GOOD_INDEX_USED > 0 OR SUM_NO_INDEX_USED > 0, '*', '') AS full_scan,

COUNT_STAR AS exec_count,

SUM_ERRORS AS err_count,

SUM_WARNINGS AS warn_count,

(SUM_TIMER_WAIT) AS total_latency,

(MAX_TIMER_WAIT) AS max_latency,

(AVG_TIMER_WAIT) AS avg_latency,

(SUM_LOCK_TIME) AS lock_latency,

format(SUM_ROWS_SENT,0) AS rows_sent,

ROUND(IFNULL(SUM_ROWS_SENT / NULLIF(COUNT_STAR, 0), 0)) AS rows_sent_avg,

SUM_ROWS_EXAMINED AS rows_examined,

ROUND(IFNULL(SUM_ROWS_EXAMINED / NULLIF(COUNT_STAR, 0), 0))  AS rows_examined_avg,

SUM_CREATED_TMP_TABLES AS tmp_tables,

SUM_CREATED_TMP_DISK_TABLES AS tmp_disk_tables,

SUM_SORT_ROWS AS rows_sorted,

SUM_SORT_MERGE_PASSES AS sort_merge_passes,

DIGEST AS digest,

FIRST_SEEN AS first_seen,

LAST_SEEN as last_seen

FROM performance_schema.events_statements_summary_by_digest d

where d

ORDER BY SUM_TIMER_WAIT DESC

limit 20;