Hello,
We are running 9.2 w/ streaming replication.
The slave is used for heavy tsearch based data mining.
Apparently depending on the business of the master the slave queries fail with different frequency with the following
message—
ERROR: canceling statement due to conflict with recovery
DETAIL: User query might have needed to see row versions that must be removed.
Initially we resolved the problem by introducing a retry (wait 1 second, try), which works in some cases, and fails on
otherdue to too many retries necessary (10+).
After that we introduced a second slave, and told the try to switch slaves (try here, if not, wait 1 second and try
there).
Which works almost every time.
So apparently the issues are not synchronous (probably logical as the streaming replication is asynchronous).
Has anybody else experience such an issue and are there different approaches to it?
Than you,
T.E.L.