[pgpool-general: 2174] Re: read_startup_packet: incorrect packet length What does that mean?

[Tatsuo Ishii]

> Hi Tatsuo,
>> Hi Michal,
>>> Hi Tatsuo.
>>> Sorry for the delayed answer.
>>> In the meantime I've changed the configuration, but the message
>>> "read_startup_packet: incorrect packet length" didn't go away. I am 
>>> attaching configuration at the end of the e-mail and also a log file 
>>> right after I started Pgpool.
>>> 
>>> I think I understand the issue I've described in the last e-mail. I 
>>> mean the psql -c "show pool_nodes;" -d postgres -U postgres -p 9898 -> 
>>> the port should be 9999. Just silly typo.
>>> 
>>> I would like to summary of the issues I am experiencing now:
>>> 1. There are periodic ERROR messages in the Pgpool log file -
>>> "read_startup_packet: incorrect packet length (-1656006549)".
>>> 2. There are periodic messages in the PostgreSQL log file - 
>>> "incomplete startup packet".
>>> 3. If I halt or reboot primary node (soft shutdown), the secondary 
>>> node triggers fail-over immediatelly. --> NOT DESIRED BEHAVIOUR 4. If 
>>> I power off the primary node (unexpectedly), the secondary node 
>>> applies health_check and triggers fail-over after health_check 
>>> time-out. --> DESIRED BEHAVIOUR
>>> 
>>> I think, the first two issues are related. But don't know what could
> cause it. 
>>> We are using special services which connects to the Pgpool by using 
>>> JDBC. Can the services be the source of these error messages?
> 
>>No idea. However, you can enable "log_connections" which logs who is
> connecting to pgpool. This will reveal who is the client when the errors
> raise.
> Good point. I will try to investigate it and also asks our developers what
> could do that. From what I know the services are doing the "SELECT 1" query
> to determine if the backend is alive. But that is a normal thing. We'll see.
> 
>>> 
>>> According to issue number 3.. I can see in the log file, what happened: 
>>> "postmaster on DB node 0 was shutdown by administrative command".
>>> How can I force Pgpool to not fail-over after the administrative 
>>> command was issued? I would like to apply same health_check timouts 
>>> also in the situation when the administrative command was issued. It 
>>> is because in our environment one node is represented by more hosts 
>>> where the Postgresql services are migrating from one host to another
> automatically. It occurs also if the "halt"
>>> or "reboot" command is issued on that one node. This is not errorneous 
>>> condition for our environment, because if the "halt" is issued on one 
>>> host, PostgreSQL service will be brought up on another node. Virtual 
>>> IP address is in front of it.
> 
>>I understand your situation. What you can do is setting:
> 
>>backend_flag0 = 'DISALLOW_TO_FAILOVER'
> 
>>This will prevent your primary node from fail-over by "shutdown by
> administrative command". Please note that this will also prevent the primary
> from fail-over by the health check.
> This is what I was hoping to avoid. Because it will absolutly disallow to
> fail-over from primary node to the secondary. But thank you anyway. I am now
> certain that, there is no other way and I should include additional steps to
> maintenance manual, which should be taken in the process of restarting one
> host by administrative command where pgpool currently runs.
> 
> One more question.. do you plan to add the setting of "not to fail-over by
> administrative command" in the future releases or it is absolutely wrong
> idea?

IMO, that would be a good idea. I will add it to the TODO list.
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese: http://www.sraoss.co.jp