Bug #30932 slave_net_timeout for connect() to master is not optimal.
Submitted: 10 Sep 2007 14:09 Modified: 24 Oct 2007 12:46
Reporter: Domas Mituzas Email Updates:
Status: Closed Impact on me:
None 
Category:MySQL Server: Replication Severity:S2 (Serious)
Version:5.0, 5.1, .. OS:Any
Assigned to: Andrei Elkin CPU Architecture:Any
Tags: replication

[10 Sep 2007 14:09] Domas Mituzas
Description:
Connections to master from slave use slave_net_timeout as connect() timeout. 
There're multiple issues with it:

a) Default value is 3600, so connect() will be terminated by OS rather than MySQL
b) Reducing slave_net_timeout to reasonable amount for connect() (like 10s) would cause disconnects in case of idle replication, as well as log rotation (see Bug#20435)

As there is no way to fail fast with connects (unlike in client library), it seems to be buggy implementation of replication networking. 

How to repeat:
SLAVE START;

Suggested fix:
decouple connect timeout from network timeout, make it settable properly or at least set to reasonable value.
[13 Oct 2007 11:19] Andrei Elkin
With fixes for bug#20435 introducing a new heartbeat feature `slave_net_timeout'
can be set to a small values e.g the same suggested 10 seconds.
The heartbeat replication event - to be sent in master's idling time - resets `slave_net_timeout'. Provided that the heartbeat period would be set to a value
less that `slave_net_timeout' there is a very slim chance for the latter to elapse and force reconnecting.
[16 Oct 2007 20:59] Lars Thalmann
This is fixed by the fix for BUG#20435:

> a) Default value is 3600, so connect() will be terminated by OS
>    rather than MySQL

With heartbeats enabled, there will be no termination of connect().

> b) Reducing slave_net_timeout to reasonable amount for connect()
>    (like 10s) would cause disconnects in case of idle replication, as
>    well as log rotation (see Bug#20435)

With heartbeats enabled, there will be no idle replication.
[24 Oct 2007 9:54] Jon Stephens
Documented fix in mysql-5.1.22-ndb-6.3.4 changelog; closed.
[24 Oct 2007 10:15] Domas Mituzas
This is generic bug, not related to NDB - is it going to be added to mainstream versions?
[8 Nov 2007 14:08] Lars Thalmann
Domas, yes it will appear in 6.0.
[28 Nov 2007 20:21] Jon Stephens
Also documented for 6.0.4 release - see also Bug #20435.