Thread: Backends staying around
Hi all, How long are backends supposed to hang around after a query is executed? We're running into trouble because they're not going away - this evening there were the full default complement of 32 sitting there, and the machine was refusing any more connections. I've increased -N to 64 and -B to 128 after reading the FAQ, but obvously I'd prefer if they quietly died after handling requests. We are running Postgres 7.0.2, accessed from an NT4 server via ADO 2.5 from IIS 4 (how's that for a mouthful of acronyms...?). Many thanks. --Ray O'Donnell --------------------------------------------------------- Raymond O'Donnell http://www.iol.ie/~rod/organ rod@iol.ie The Irish Pipe Organ Page ---------------------------------------------------------
On Tue, Dec 04, 2001 at 12:15:33PM -0000, Raymond O'Donnell wrote: > Hi all, > > How long are backends supposed to hang around after a query is > executed? We're running into trouble because they're not going > away - this evening there were the full default complement of 32 > sitting there, and the machine was refusing any more connections. > I've increased -N to 64 and -B to 128 after reading the FAQ, but > obvously I'd prefer if they quietly died after handling requests. Well, the backends die when the client closes the connection. Perhaps you should check on the number of open TCP connections to the backend and find the process that is neglecting to close it's connection. > We are running Postgres 7.0.2, accessed from an NT4 server via > ADO 2.5 from IIS 4 (how's that for a mouthful of acronyms...?). Maybe you have some form of persistant connections? At least in unix, a socket is closed automatically when all processes owning the endpoint die off or close it. HTH, -- Martijn van Oosterhout <kleptog@svana.org> http://svana.org/kleptog/ > Terrorists can only take my life. Only my government can take my freedom.
On 5 Dec 2001, at 0:00, Martijn van Oosterhout wrote: > Maybe you have some form of persistant connections? At least in unix, > a socket is closed automatically when all processes owning the > endpoint die off or close it. Thanks for your reply! I have connection pooling turned on in ODBC on the NT machine, but this is supposed to let clients re-use the connections, so in theory at least the number of backends shouldn't rise beyond a certain level - IIS always uses the same username/password to connect to the database, so connection pooling ought to be pretty efficient. I'll double-check how the connections are handled in the COM components called from IIS. --Ray. --------------------------------------------------------- Raymond O'Donnell http://www.iol.ie/~rod/organ rod@iol.ie The Irish Pipe Organ Page ---------------------------------------------------------