[MXS-2320] Basic MaxScale setup causes instability for applications. Created: 2019-02-07  Updated: 2019-10-28  Resolved: 2019-10-28

Status: Closed
Project: MariaDB MaxScale
Component/s: readwritesplit
Affects Version/s: 2.2.3, 2.3.3
Fix Version/s: N/A

Type: Bug Priority: Major
Reporter: William Edwards Assignee: Unassigned
Resolution: Cannot Reproduce Votes: 0
Labels: need_feedback
Environment:

Debian 9 (Stretch)
VM (KVM)



 Description   

I've been having strange issues where specific applications have queries fail at random. (Debug logging didn't output any useful results.)

Strange thing is that almost half of all queries failed with MaxScale 2.2.3. Then, I upgraded MaxScale to 2.3.3 (without changing the config or anything else), and now 'only' a few queries fail.

For example, I have OTRS running. It works perfectly fine when talking directly to db01.dmz (Galera node), but starts randomly failing with MaxScale. But, most of the time, it works just as well as it dose with that Galera node. This is not OTRS-specific, because it is happening with more applications.

I'm quite certain I have to do with a bug here, but I have the most basic setup:

  • Galera cluster with 3 nodes on Debian;
  • readwritesplit on MaxScale on Debian

The config:

[db03.dmz.cyberfusion.cloud]
priority=3
protocol=MariaDBBackend
type=server
port=3306
address=10.10.0.9
 
[db02.dmz.cyberfusion.cloud]
priority=2
protocol=MariaDBBackend
type=server
port=3306
address=10.10.0.8
 
[splitterlistener]
protocol=MariaDBClient
type=listener
port=3306
service=readwriteservice
address=0.0.0.0
 
[galeramonitor]
use_priority=true
module=galeramon
servers=db01.dmz.cyberfusion.cloud, db02.dmz.cyberfusion.cloud, db03.dmz.cyberfusion.cloud
available_when_donor=true
user=maxscale
password=x
type=monitor
 
[db01.dmz.cyberfusion.cloud]
priority=1
protocol=MariaDBBackend
type=server
port=3306
address=10.10.0.7
 
[maxscale]
threads=auto
 
[readwriteservice]
transaction_replay=true
master_accept_reads=True
use_sql_variables_in=master
servers=db01.dmz.cyberfusion.cloud, db02.dmz.cyberfusion.cloud, db03.dmz.cyberfusion.cloud
user=maxscale
router=readwritesplit
password=x
type=service
 
[maxadminlistener]
protocol=maxscaled
type=listener
socket=default
service=maxadminservice
 
[maxadminservice]
router=cli
type=service



 Comments   
Comment by markus makela [ 2019-02-07 ]
  • How does the failure manifest itself?
  • Do you get an error in the application? If so, what is the error?
  • Can you attach the MaxScale log or tell us whether it contains any errors?
  • Does the problem persists if you remove any of the transaction_replay, master_accept_reads or use_sql_variables_in parameters?
Generated at Thu Feb 08 04:13:16 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.