Hello friends,
*Today I decided to tune MySQL Master Server by adjusting some
variables:*
-join_buffer bigger (1-2MB)
-read_buffer bigger (8MB)
-read_rnd_buffer bigger (2MB)
-connect_timeout bigger (20-25)
-tmp_table_size 4x bigger (512MB)
-sort_buffer bigger (2MB)
-table_cache 2x bigger (512)
Server* is not* on heavy load so I can afford high values.
4cpu-8cores, 8gb ram, sas 15k hdd.
Local application installed.
I tested MonYog for monitoring and I use MySQL report on linux. Both
installed on Linux. Before I could expect MySQL
improvement after service restart and wait at least a day to see
status variables Replication crashes.
*The error was like few years ago. Error description:*
Error 1236 on slave. Could not read binary log (corrupt)?
SLAVE_IO is on not running. SLAVE_SQL running.
MySQL master version : 5.0.51b win64
MySQL slave version: 5.0.51b win36 and 5.0.67 (linux x64)
Since nothing has cheanged at least a year and I mostly do not
restart replication for at least 6 months period. This occurred
an hour or more after tunning.
Now my question is do some of these variables really affect replication?
*Slave servers are connected to master using following:*
MASTER (3306) - SERVER01 (on direct line, static ip) - 3309 (ip
filter, using CCproxy) - SLAVE (global, location DE)
Master and Server01 are inside the company on a LAN (1Gbps). Direct
Line speed is 5/5 Mbps. SLAVE is 100Mbps (Web Server). I doubt that its
related to network problems because it works perfect.
Slave connects to proxy on 3309, Server01 forwards request to MASTER.
* Reason for optimize:*
-table locks, waited(Select queries), state DEAD displayed in
process list - query does finish in a lot of time, concurrent query's
while user presses F5 in browser makes it a little worse
-table scans
-joins
-tryed to improve server by tunning because of those reasons
-query cache not used because of big query's
-index hit rate 99,96%
-slow query time: *200s,* lock times:* 60s*
* Query status:*
-lot's of slow queries which I can't rewrite because its not in
my jurisdiction
-notified the programmers and sent slow queries
-indexes used
-tryed to make some status with mk-query-filter
-Explain
Generally this Question may not be useful but I only wanted to know
replication connections with
these variables that i modified and gave myself additional work to
fix replication.
* Conclusion:
*Replication does not like* high buffer* values and high
max_allowed_packet on master. Our max_allowed_packet
stayed as default at 1M which is ok and works fine (this variable
was not changed in my tunning).
max_allowed_packet=1M (because of LOAD DATA command), high values
broke replication with same error 1236.
MySQL phenomenon errors once more.
Thank you.
Sorry for a long long description.
--
Andrej Pintar
email : api984@stripped
andrej@stripped
api984@stripped
web: http://www.api984.net
contact cell: 00385 98 790 639
home server: http://api984.ath.cx
ICQ: 191748772
Skype: api9841
MSN: fatallord@stripped
IRC: api984, freenode.net
::Software is like sex: it's better when it's free::
