perl-modperl mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jonathan Vanasco <modperl-l...@2xlp.com>
Subject DBD:Pg (Postgres) + Apache::DBI
Date Sun, 19 Mar 2006 20:13:25 GMT
I'm finishing up my migration to Postgres, and i noticed some odd  
behavior

I was hoping that someone could point me in the right direction:

a - 	i often lose a connection to the server for an apache child.  i  
can't seem to trace when where or why this happens
	i just get this in the log "DBD::Pg::st execute failed: no  
connection to the server"
	the other children are unaffected.  i'm wondering if this is a ping  
error, postgres shutting things down from too many invalid queries  
(i'm still on my dev box) , or something else

b- 	several times a handle seems to have been corrupted.  usually it  
seems to happen with a failed sequence operation ( adding a new  
record or retreiving a value )
	when this happens, the handle seems to not be reset by the cleanup  
handler ( ie - the automagic Apache::DBI issued rollback ).  i have  
to restart the server to clean the handle
	has this happened to anyone else?

c-	i was getting a bunch of "message type 0x49 arrived from server  
while idle" -> "message type 0x53 arrived from server while idle"
	looking on the net, i wasn't able to find much about it.  a lot of  
people had the issue on the perl DB lists, the Postgres lists, but  
there was no mention of what caused it.
	Then I finally came across somethig from the bricolage changelog:
		http://bricolage.sourceforge.net/docs/Bric/Changes.html
		Removed queries added in 1.7.2 that were running at Bricolage  
startup time. They could cause DBI to cache a database handle and  
return it after Apache forks, leading to strange errors such as  
``message type 0x49 arrived from server while idle'', and  
occaisionally a frozen server. [David]

	It turns out that I had a handle that was cached by accident ( i  
hadn't made a config only user yet, and the config handle was  
connecting with the same params as my writing handle).  Chances are ,  
thats probably the issue.  But does anyone know if there is a  
definitive cause for this?

Thanks.
Jonathan

Mime
View raw message