Hello Machiel,
(I am guessing you can only process top-posts?)
When you stop only the IO thread, you may leave the last event recorded
to the Relay Log incomplete.
When it gets to that part of the Relay Log, the SQL thread may only be
part-way through a transaction. It will keep that transaction alive
waiting for the IO thread to finish downloading the rest of the event
from the master's copy of the binary log. That partially-complete
transaction is most likely blocking the ability of your other commands
to operate more efficiently for several reasons:
* MVCC
* InnoDB history length
* Incomplete transactions to secondary indexes forcing those commands
to scan the table instead of using the index (related to MVCC)
We made the SQL thread wait so that intermittent networks (a real thing
years ago) would not "break" replication. We would wait for connectivity
to resume so that replication could continue.
A safer plan is to stop both threads at the same time. Just use the
basic STOP SLAVE command instead of the more specific STOP SLAVE IO_THREAD.
Yours,
--
Shawn Green
MySQL Senior Principal Technical Support Engineer
Oracle USA, Inc. - Integrated Cloud Applications & Platform Services
Office: Blountville, TN
Become certified in MySQL! Visit https://www.mysql.com/certification/
for details.
On 10/26/2018 2:09 AM, Machiel Richards wrote:
Hi Shawn
Thank you for the response.
In order to pause the slave , the stop the sql_io_thread, and to unpause
they simply start the thread.
I have run "show engine innodb status" yes and the threads show 90% as
sleeping and then a few selects , all from the same table as it does a lot of
authentications for dial outs.
I will have a look at the results from SELECT * FROM
information_schema.INNODB_TRX; during the day as we get this issue regularly
and will provide feedback.
Regards
(earlier thread below... )
From: shawn l.green
Sent: Thursday, 25 October 2018 9:54:10 PM
To: mysql@lists.mysql.com
Subject: Re: High cpu usage
Hello Machiel,
On 10/25/2018 6:09 AM, Machiel Richards wrote:
Good day all
Hoping this mail finds you well.
I am hoping someone can perhaps give us some guidance here as we now seem
to be stuck on a problem and have not been able to find a solution after more
than a month.
We are running an opensips server on Centos 6.5 , using mysql 5.7.13 which
was installed via Tarball file.
The server is setup as a slave and master and receives updates from
opensips config nodes as well as registrations from workers.
Replication is paused during the day and forward replication (master) is
disabled at the moment.
However , we are getting an issue every day on mysql side in terms of
mysql pushing up server load.
During the day the server is running fine with a load avg not going above
1.5 during peak times.
However in the evening , replication is unpaused, and completes
processing and catchup within about 15 minutes and is paused again about 30
minutes after the unpause.
Give or take 45 minutes to an hour after the replication is paused
again, mysql starts to cause high cpu usage with no apparent processes running
as can be seen on full processlist (maybe one or two selects which completes
fairly quickly)
The higher load, causes queries to slow down however and opensips to
start timing out on db connections, causing clients to resubmit.
The resubmits , then obviously causes even more load spiking the mysql
load to increase as well as the server load and eventually opensips kills
itself.
I have looked at the disks, iowaits, memory usage, all is fine.
We do not see any strange queries or stick queries, no deadlocks, etc...
only the increase in selects after mysql starts to push up the cpu load.
We have added all indexes we can find, but even this has made no
difference at all.
Currently we are at a loss so I am hoping someone else can assist in
explaining how else we can find out why mysql is eating up the cpu ...
The same behaviour can also be seen the moment any new feature is added to
the server that requires mysql processing to be done, so this does not seem to
be specifically related to replication, however it does seem like the current
load from replication causes mysql to act up.
the server is currently running on SSD (recently replaced) , and 8Gb of
memory with 1 x quadcore CPU.
should any more info be required, please feel free to ask.
When you say pause replication, what command are you executing on the
slave?
Which end of the system is experiencing the high CPU usage: the master
or the slave?
Have you checked these resources to see what the InnoDB main or
background threads are doing when your CPU starts to spike? (you could
be in a massive rollback)
SHOW ENGINE INNODB STATUS
SELECT
