postgres-xc-developers Mailing List for Postgres-XC

Hi All,

It'a a patch for fixing bug in gtm_serialize.c

hi 
  When I use pgxc_ctl tool to monitor datanode.  I found that pgxc_ctl monitor function have some problems as follow:
      1.If datanode is recovering, then pgxc_ctl monitor issue wrong state information that datanode is not runing. 
         Obviously, It is wrong. I hope pgxc_ctl tool tell me that datanode is recovering.
      2.If datanode is shutting down. pgxc_ctl mointor function has the same problem as above.

Now I have a design that it can provide detailed and precise information on the status of the database.
In datanode, there is a file in data folder. Name is pg_control,  Path is data/global.

This file records the current state information about the datanode as follow:
   Database cluster state:               in production
   Database cluster state:               starting up
   Database cluster state:               shut down
   Database cluster state:               shut down in recovery
   Database cluster state:               shutting down
   Database cluster state:               in crash recovery
   Database cluster state:               in archive recovery
     
Now if I know datanode master process is active. Then I can make sure that the stats information in pg_control is right.

Pseudocode as follows:
   1.See whether datanode master process is active with ssh command(ssh pgxc@192.168.198.101 ps -ef | grep  /home/pgxc/data/datanode)
   2.If master process is active, then we can get current state information with ssh command(ssh pgxc@192.168.198.101 /home/pgxc/postgres_xc/bin/pg_controldata                         /home/pgxc/data/dn_master/ | grep "Database cluster state")
   3.If master process is not active. then tell user that datanode service is shut down.

Please help to review this design, If any problem, please let me know.
Thank your for help




xu...@cm...

Hi All,

The gtm.patch1 is for TXN_START_PREPARED backup process.
The gtm.patch2 does some code refactory after apply gtm.patch1.

Hi All,

The patch is for Begin Transaction. It uses uninitialize timestamp for
backup.

Hi,

this patch reduce error level to LOG in ServerLoop() for GTM and GTM proxy.
    
Because PG_exception_stack not set in main thread, elog with
ERROR(eg. thread create failed) can cause whole program exit, see
pg_re_throw(). This patch refers to error process of postmaster,
which just log error.

Note, palloc still used in ServerLoop(), which will raise a ERROR
message if out of memory, fix it need another patch.


-- 
Best Regards,
Wang Diancheng