I migrated my MySQL database from GCP to Azure (both 5.7), but it seems to have affected performance.

Server before migration: 2 VCPUS with 7.5GB memory
Server after migration: 2 VCPUS with 8GB memory

Both servers run / ran version 5.7 of the MySQL server. My database is currently around 6GB in size, growing 100MB+ a day. It only consists of 32 tables, although a fraction of them tables enter the millions of rows category.

I read up on innodb_buffer_pool_size, GCP apparently sets it to around 80% of the memory, which would make it 6GB. I have set the innodb_buffer_pool_size on the new server to the same value.

Before updating this value (when I first noticed decreased performance), innodb_buffer_pool_size was set to 0.1 GB on the new server, I then decided to update this to the value the GCP server was set at hoping it would help.

Following this documentation I was able to update the buffer pool size.

How did I check the innodb_buffer_pool_size initially?

-- returned 0.111...
SELECT @@innodb_buffer_pool_size/1024/1024/1024;

How did I update innodb_buffer_pool_size?

SET GLOBAL innodb_buffer_pool_size=6442450944;

I checked the resize status with this query,

-- returned 'Completed resizing buffer pool at 200920 13:46:20.'
SHOW STATUS WHERE Variable_name='InnoDB_buffer_pool_resize_status';

I execute around 2 queries a second, peaking at 250k a day spread out. I can't be certain but this usage shouldn't be enough to halt performance?

How am I checking performance?

I have shown a list of queries ran, and the times it takes for the server to respond. I have tested these queries in Navicat, Datagrip, and CLI with similar results.

I wasn't sure what queries to include here to give as much information as possible, so if I haven't included anything useful I can update it upon request.

-- Fetching 100k rows from a 3.1m rows table
-- Time took: 21.248s
SELECT * FROM `profile_connections` LIMIT 100000;

-- (SECOND TIME) Fetching 100k rows from a 3.1m rows table
-- Time took: 1.735s
SELECT * FROM `profile_connections` LIMIT 100000;

- Fetching a random row from a 3.1m row table 
-- Time took: 0.857s
SELECT * FROM `profile_connections` WHERE `id` = 2355895 LIMIT 1;

-- (SECOND TIME) Fetching a random row from a 3.1m row table 
-- Time took: 0.850s
SELECT * FROM `profile_connections` WHERE `id` = 2355895 LIMIT 1;

-- Fetching all rows from a 20 row table
-- Time took: 40.010s
SELECT * FROM `profile_types`

-- (SECOND) Fetching all rows from a 20 row table
-- Time took: 0.850s
SELECT * FROM `profile_types`

But at times, I can run all of the above queries and get a response in 2 - 5 seconds. Performance seems to be hit or miss, there are huge differences in times taken for the same query, depending on when it is run which I am currently struggling to diagnose.

I ran mysqltuner and got these performance metrics back:

[--] Up for: 47m 39s (38K q [13.354 qps], 1K conn, TX: 403M, RX: 63M)
[--] Reads / Writes: 50% / 50%
[--] Binary logging is disabled
[--] Physical Memory     : 7.8G
[--] Max MySQL memory    : 146.8G
[--] Other process memory: 0B
[--] Total buffers: 6.0G global + 954.7M per thread (151 max threads)
[--] P_S Max memory usage: 72B
[--] Galera GCache Max memory usage: 0B
[!!] Maximum reached memory usage: 21.9G (281.61% of installed RAM)
[!!] Maximum possible memory usage: 146.8G (1888.34% of installed RAM)
[!!] Overall possible memory usage with other process exceeded memory
[OK] Slow queries: 3% (1K/38K)
[OK] Highest usage of available connections: 11% (17/151)
[OK] Aborted connections: 0.67%  (9/1342)
[!!] name resolution is active : a reverse name resolution is made for each new connection and can reduce performance
[OK] Query cache is disabled by default due to mutex contention on multiprocessor machines.
[OK] Sorts requiring temporary tables: 0% (0 temp sorts / 41 sorts)
[OK] No joins without indexes
[OK] Temporary tables created on disk: 4% (82 on disk / 1K total)
[OK] Thread cache hit rate: 98% (17 created / 1K connections)
[OK] Table cache hit rate: 63% (667 open / 1K opened)
[OK] table_definition_cache(1400) is upper than number of tables(302)
[OK] Open file limit used: 1% (55/5K)
[OK] Table locks acquired immediately: 100% (1K immediate / 1K locks)

Slow query logs I run a lot of the same queries, so I've truncated it to include just a few.

# Time: 2020-09-20T16:45:04.230173Z
# User@Host: root[root] @  []  Id:     7
# Query_time: 1.022011  Lock_time: 0.000084 Rows_sent: 1  Rows_examined: 1058161
SET timestamp=1600620304;
SELECT @id := `id`,`item`
                    FROM `queue_items`
                    WHERE `processed_at` IS NULL AND `completed_at` IS NULL AND `confirmed` = '1'ORDER BY `id` ASC
                    LIMIT 1
                    FOR UPDATE;
# Time: 2020-09-20T16:45:09.676613Z
# User@Host: root[root] @  []  Id:     5
# Query_time: 1.198063  Lock_time: 0.000000 Rows_sent: 0  Rows_examined: 0
SET timestamp=1600620309;
# Time: 2020-09-20T16:45:22.938081Z
# User@Host: root[root] @  []  Id:     4
# Query_time: 5.426964  Lock_time: 0.000133 Rows_sent: 0  Rows_examined: 1
SET timestamp=1600620322;
UPDATE `queue_items` SET `completed_at` = '2020-09-20 16:45:17', `updated_at` = '2020-09-20 16:45:17' WHERE `id` = 1818617;

Since asking, I have added the following variables to my config, although it doesn't seem to have fixed any of the performance issues.




SET GLOBAL innodb_buffer_pool_size=6442450944;

Lasts only until the server restarts. You must add it to the config file (typically my.cnf) to make it permanent.

Yes, 6G is probably a good value. If you have any swapping, lower it, starting with 5G.

It is quite typical for the second run to be 10x faster than the first. This is because the first had to do a lot of disk reads into the buffer_pool; then the second read could take advantage of the cached data.

Fetching a specific row (that's not a "random" row), took a long time because you don't have an index on id. Every table should have a PRIMARY KEY (which is an index).

40 seconds for a 20 row table -- No way! Even 0.85s is not real. Please provide more details.

The first entry from the slowlog says that a lot of the table needed to be scanned. This composite index will make it run much faster: INDEX(confirmed, completed_at, processed_at, id) -- Possibly 1ms instead of 1s.

Please provide SHOW CREATE TABLE queue_items; .

(For security and other reasons, I recommend you avoid "root" except for admin purposes. Create a login for the app that is limited to the database(s) it needs to touch.)

Did you change any other VARIABLES?

OK, I see it now. That SELECT probably scanned through more data than would fit in the buffer_pool. That bumped out the 20-row table, perhaps repeatedly. Add my suggested index. And, if you have any other slow queries, let's discuss them.

Also, consider purging old data from your "queue". Tips: http://mysql.rjweb.org/doc.php/deletebig

  • Thanks for replying. I was under the impression that primary keys (id on profile_connections) were indexes in MySQL? So the query I showed for fetching a random row (the query I usually use in my production application) already had an index? As well as id on queue_items? I could be wrong. You might find this interesting: pastebin.com/raw/vwUDtLsF - it is a dump of various outputs of variables and configuration and a report from a MySQL tuner. I have changed my configuration file to include the settings advised by the MySQL tuner and also added them to this question see my edit. – jacks Sep 22 '20 at 9:25
  • Yes, a PRIMARY KEY (in MySQL) is UNIQUE and an INDEX – Rick James Sep 22 '20 at 21:56
  • @jacks - You are at two cusps. The upgrade may have pushed your app over the cliff. (1) Swapping (easily fixed by lowering buffer_pool_size) (2) Doing full table scans that blows out RAM (maybe fixed by changing queries, changing indexes, getting more RAM) – Rick James Sep 22 '20 at 21:59

(separate "Answer" due to bulkiness)

This shows a few things that could be fixed, plus a suggestion that the slowlog may have useful info.


  • Version: 5.7.31-0ubuntu0.18.04.1-log
  • 7.5 GB of RAM
  • Uptime = 18:31:01; some GLOBAL STATUS values may not be meaningful yet.
  • You are not running on Windows.
  • Running 64-bit version
  • You appear to be running entirely (or mostly) InnoDB.

The More Important Issues:

Even after 18 hours of running, the buffer_pool is not full. Perhaps the dataset is smaller than the buffer_pool is capable of fully caching.

Shrink max_allowed_packet. It is so high that it threatens to cause swapping.

Digest the slowlog to see which queries are causing the most slowdowns.

SSD drives? If so, increase innodb_io_capacity (now 200) to 1000.

26% of queries are "admin" (eg, change db) -- Is this coming from some 3rd party software? It looks like 44 USE commands per connection!

Details and other observations:

( innodb_lru_scan_depth * innodb_page_cleaners ) = 1,024 * 4 = 4,096 -- Amount of work for page cleaners every second. -- "InnoDB: page_cleaner: 1000ms intended loop took ..." may be fixable by lowering lru_scan_depth: Consider 1000 / innodb_page_cleaners (now 4). Also check for swapping.

( innodb_page_cleaners / innodb_buffer_pool_instances ) = 4 / 6 = 0.667 -- innodb_page_cleaners -- Recommend setting innodb_page_cleaners (now 4) to innodb_buffer_pool_instances (now 6)

( innodb_lru_scan_depth ) = 1,024 -- "InnoDB: page_cleaner: 1000ms intended loop took ..." may be fixed by lowering lru_scan_depth

( Innodb_buffer_pool_pages_free / Innodb_buffer_pool_pages_total ) = 214,657 / 393216 = 54.6% -- Pct of buffer_pool currently not in use -- innodb_buffer_pool_size (now 6442450944) is bigger than necessary?

( innodb_io_capacity_max / innodb_io_capacity ) = 2,000 / 200 = 10 -- Capacity: max/plain -- Recommend 2. Max should be about equal to the IOPs your I/O subsystem can handle. (If the drive type is unknown 2000/200 may be a reasonable pair.)

( Uptime / 60 * innodb_log_file_size / Innodb_os_log_written ) = 66,661 / 60 * 780M / 1113788928 = 815 -- Minutes between InnoDB log rotations Beginning with 5.6.8, this can be changed dynamically; be sure to also change my.cnf. -- (The recommendation of 60 minutes between rotations is somewhat arbitrary.) Adjust innodb_log_file_size (now 817889280). (Cannot change in AWS.)

( innodb_flush_method ) = innodb_flush_method = -- How InnoDB should ask the OS to write blocks. Suggest O_DIRECT or O_ALL_DIRECT (Percona) to avoid double buffering. (At least for Unix.) See chrischandler for caveat about O_ALL_DIRECT

( Innodb_row_lock_time_avg ) = 1,635 -- Avg time to lock a row (millisec) -- Possibly conflicting queries; possibly table scans.

( Innodb_row_lock_time ) = 28,949,111 / 66661 = 434 /sec -- Total ms of waiting for row locks -- Possibly conflicting queries; possibly table scans.

( Innodb_row_lock_waits ) = 17,702 / 66661 = 0.27 /sec -- How often there is a delay in getting a row lock. -- May be caused by complex queries that could be optimized.

( Innodb_row_lock_waits/Innodb_rows_inserted ) = 17,702/43950 = 40.3% -- Frequency of having to wait for a row.

( innodb_flush_neighbors ) = 1 -- A minor optimization when writing blocks to disk. -- Use 0 for SSD drives; 1 for HDD.

( innodb_io_capacity ) = 200 -- I/O ops per second capable on disk . 100 for slow drives; 200 for spinning drives; 1000-2000 for SSDs; multiply by RAID factor.

( innodb_print_all_deadlocks ) = innodb_print_all_deadlocks = OFF -- Whether to log all Deadlocks. -- If you are plagued with Deadlocks, turn this on. Caution: If you have lots of deadlocks, this may write a lot to disk.

( max_allowed_packet ) = 999,999,488 / 7680M = 12.4% -- If you do not have large blobs (etc) to load, then decrease the value. Else decrease innodb_buffer_pool_size (now 6442450944) to make room. Swapping is terrible for performance.

( character_set_server ) = character_set_server = latin1 -- Charset problems may be helped by setting character_set_server (now latin1) to utf8mb4. That is the future default.

( local_infile ) = local_infile = ON -- local_infile (now ON) = ON is a potential security issue

( Select_scan / Com_select ) = 43,581 / 115824 = 37.6% -- % of selects doing full table scan. (May be fooled by Stored Routines.) -- Add indexes / optimize queries

( Com_admin_commands / Queries ) = 170,996 / 658652 = 26.0% -- Percent of queries that are "admin" commands. -- What's going on?

( Com__biggest ) = Com__biggest = Com_change_db -- Which of the "Com_" metrics is biggest. -- Normally it is Com_select (now 115824). If something else, then it may be a sloppy platform, or may be something else.

( Slow_queries / Questions ) = 9,208 / 484952 = 1.9% -- Frequency (% of all queries) -- Find slow queries; check indexes.

( log_slow_slave_statements ) = log_slow_slave_statements = OFF -- (5.6.11, 5.7.1) By default, replicated statements won't show up in the slowlog; this causes them to show. -- It can be helpful in the slowlog to see writes that could be interfering with Replica reads.

( Com_change_db / Connections ) = 171,209 / 3860 = 44.4 -- Database switches per connection -- (minor) Consider using "db.table" syntax

ulimit's open files is only 1024, but it has not tripped you up (yet).

Abnormally small:

Com_set_option = 13 /HR
Select_range = 0
Select_range / Com_select = 0

Abnormally large:

Com_kill = 0.97 /HR
Com_show_slave_hosts = 0.27 /HR
Com_show_storage_engines = 0.27 /HR
Handler_read_next = 543376 /sec
Handler_read_next / Handler_read_key = 2,178
Handler_read_rnd_next / Handler_read_rnd = 36,203
Performance_schema_file_instances_lost = 1
Ssl_accepts = 3,288
Ssl_finished_accepts = 3,288
Ssl_session_cache_misses = 6
Ssl_used_session_cache_entries = 5

Abnormal strings:

innodb_fast_shutdown = 1
optimizer_trace = enabled=off,one_line=off
optimizer_trace_features = greedy_search=on, range_optimizer=on, dynamic_range=on, repeated_subselect=on
slave_rows_search_algorithms = TABLE_SCAN,INDEX_SCAN

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.