Re: The behavior of CheckRequiredParameterValues() - Mailing list pgsql-hackers

From Haribabu Kommi
Subject Re: The behavior of CheckRequiredParameterValues()
Date
Msg-id CAJrrPGe2aoBSK+zFD2jK8UoWn+Uy3JMVz-G9H_V=1SzD-GD8=w@mail.gmail.com
Whole thread Raw
In response to The behavior of CheckRequiredParameterValues()  (Sawada Masahiko <sawada.mshk@gmail.com>)
List pgsql-hackers
On Wed, Mar 5, 2014 at 4:09 AM, Sawada Masahiko <sawada.mshk@gmail.com> wrote:
Hi all,

I had doubts regarding behavior of CheckRequiredParameterValues() function.

I could not start standby server which is created by pg_basebackup
with following scenario.
1. Start the master server with 'wal_level = archve' , 'hot_standby =
on' and other settings of replication.
2. Create the standby server from the master server by using pg_basebackup.
3. Change the wal_level value of both master and standby server to
'hot_standby'.
4. Restarting the master server.
5. Starting the standby server.

In #5, I got following error even if I set wal_level to 'hot_standby'.

FATAL:  hot standby is not possible because wal_level was not set to
"hot_standby" or higher on the master server

I tried to investigate this behaviour.
Currently CheckRequiredParameterValues() function uses wal_level value
which is got from ControlFile when comparing between wal_level and
WAL_LEVEL_HOT_STANDBY as following code.

xlog.c:6177
 if (ControlFile->wal_level < WAL_LEVEL_HOT_STANDBY)
     ereport(ERROR,
             (errmsg("hot standby is not possible because wal_level was not

So we have to start and stop standby server with changed
wal_level(i.g., hot_standby) if we want to enable hot standby.
In this case, I think that the standby server didn't need to confirm
wal_level value of ControlFile.
I think that it should confirm value which is written in postgreql.conf.

The snapshot of running transaction information is written to WAL only when the wal_level is set to 'hot_standby'.
This information is required on the standby side to recreate the running transactions. 
  
Regards,
Hari Babu
Fujitsu Australia

pgsql-hackers by date:

Previous
From: Bruce Momjian
Date:
Subject: Re: Securing "make check" (CVE-2014-0067)
Next
From: Craig Ringer
Date:
Subject: Re: Row-security on updatable s.b. views