"Richard" <husttripper@vip.sina.com> writes:
> 1.To add live HA to PG, I transfer WAL of a database instance(Primary node) to another database instance (standby
node)at real time, and keep startup alive in standby node to recovery WAL online,so that standby node can be a hot
standby.
> But I got some trouble. When standby node switch to primary mode to accept connections, of course after startup
initilizethe WAL and exit, the postgres process ereportERROR
> when mdread function enter ERROR branch, and I got message like this "could not read block X of relation X/X/X: read
onlyXof X bytes". I spent two days to figure out what happened, but it is too hard.Please help me.
I wonder whether your problem is related to this 8.3.8 bug fix:
* Force WAL segment switch during pg_start_backup() (Heikki)
This avoids corner cases that could render a base backup unusable.
I'd suggest updating to 8.3.latest and making sure that you are taking
the base backup according to the documented procedure. The symptoms you
are mentioning could definitely be caused by starting from an
inconsistent backup.
regards, tom lane
PS: Please do not post the same question to multiple mailing lists.