Обсуждение: question on backends
Hi all
As I understand every time there is a request to postgres a new backend is made, and when the request is finished, even if the connection is already active the backend dies. I wonder if is there any parameter that allow backends to remain beyond a transaction. Creating a new backend every time a transaction is made means forking the code and reallocating sort_memory. Although it is not a high resource usage, on short transactions as OLTPs it is a relevant work time, I think it would be interesting that a predefined number of backends were allowed to remain active beyond the transaction.
Thanks and Regards
Just use persistent connections.
Chris
----- Original Message -----Sent: Monday, July 29, 2002 5:32 PMSubject: [HACKERS] question on backendsHi allAs I understand every time there is a request to postgres a new backend is made, and when the request is finished, even if the connection is already active the backend dies. I wonder if is there any parameter that allow backends to remain beyond a transaction. Creating a new backend every time a transaction is made means forking the code and reallocating sort_memory. Although it is not a high resource usage, on short transactions as OLTPs it is a relevant work time, I think it would be interesting that a predefined number of backends were allowed to remain active beyond the transaction.Thanks and Regards
How?
----- Original Message -----From: Christopher Kings-LynneSent: Monday, July 29, 2002 12:36 PMSubject: Re: [HACKERS] question on backendsJust use persistent connections.Chris----- Original Message -----Sent: Monday, July 29, 2002 5:32 PMSubject: [HACKERS] question on backendsHi allAs I understand every time there is a request to postgres a new backend is made, and when the request is finished, even if the connection is already active the backend dies. I wonder if is there any parameter that allow backends to remain beyond a transaction. Creating a new backend every time a transaction is made means forking the code and reallocating sort_memory. Although it is not a high resource usage, on short transactions as OLTPs it is a relevant work time, I think it would be interesting that a predefined number of backends were allowed to remain active beyond the transaction.Thanks and Regards
libpq has a function pconnect as opposed to connect that will do it. PHP and most other interfaces will let you use persistent connections.
Chris
----- Original Message -----Sent: Monday, July 29, 2002 7:00 PMSubject: Re: [HACKERS] question on backendsHow?----- Original Message -----From: Christopher Kings-LynneSent: Monday, July 29, 2002 12:36 PMSubject: Re: [HACKERS] question on backendsJust use persistent connections.Chris----- Original Message -----Sent: Monday, July 29, 2002 5:32 PMSubject: [HACKERS] question on backendsHi allAs I understand every time there is a request to postgres a new backend is made, and when the request is finished, even if the connection is already active the backend dies. I wonder if is there any parameter that allow backends to remain beyond a transaction. Creating a new backend every time a transaction is made means forking the code and reallocating sort_memory. Although it is not a high resource usage, on short transactions as OLTPs it is a relevant work time, I think it would be interesting that a predefined number of backends were allowed to remain active beyond the transaction.Thanks and Regards
On Mon, Jul 29, 2002 at 11:28:54PM +0800, Christopher Kings-Lynne wrote: > libpq has a function pconnect as opposed to connect that will do it. libpq has neither function, AFAIK. As for persistent backends, it's on the TODO list, but I'm not aware that anyone has put any work into implementing it. At the moment, a backend connects to a single database for its entire lifecycle -- so you'd either need one pool of persistent backends for each database in use, or you'd need to allow a backend to change the database it is connected to. Cheers, Neil -- Neil Conway <neilconway@rogers.com> PGP Key ID: DB3C29FC
On Mon, 2002-07-29 at 11:32, Luis Alberto Amigo Navarro wrote: > Hi all > As I understand every time there is a request to postgres a new backend > is made, and when the request is finished, even if the connection is > already active the backend dies. I think you have misunderstood it. A new backend is forked only when a new connection is made, not for every transaction. There may be some frontends that do make a new connection for each http request or such, but most of them allow for persistent connections, either as an option or by default. > I wonder if is there any parameter > that allow backends to remain beyond a transaction. Creating a new > backend every time a transaction is made means forking the code and > reallocating sort_memory. ------------------ Hannu
if i put debug_level=1 i get for one connect and several inserts on backend die after each insert ----- Original Message ----- From: "Hannu Krosing" <hannu@tm.ee> To: "Luis Alberto Amigo Navarro" <lamigo@atc.unican.es> Cc: <pgsql-hackers@postgresql.org> Sent: Monday, July 29, 2002 8:42 PM Subject: Re: [HACKERS] question on backends > On Mon, 2002-07-29 at 11:32, Luis Alberto Amigo Navarro wrote: > > Hi all > > As I understand every time there is a request to postgres a new backend > > is made, and when the request is finished, even if the connection is > > already active the backend dies. > > I think you have misunderstood it. A new backend is forked only when a > new connection is made, not for every transaction. > > There may be some frontends that do make a new connection for each http > request or such, but most of them allow for persistent connections, > either as an option or by default. > > > I wonder if is there any parameter > > that allow backends to remain beyond a transaction. Creating a new > > backend every time a transaction is made means forking the code and > > reallocating sort_memory. > > ------------------ > Hannu > >
On Mon, 2002-07-29 at 20:21, Luis Alberto Amigo Navarro wrote: > if i put debug_level=1 i get for one connect and several inserts on backend > die after each insert What client do you use ? ---------- Hannu
libpq PQsetdb(........ ----- Original Message ----- From: "Hannu Krosing" <hannu@tm.ee> To: "Luis Alberto Amigo Navarro" <lamigo@atc.unican.es> Cc: <pgsql-hackers@postgresql.org> Sent: Monday, July 29, 2002 9:40 PM Subject: Re: [HACKERS] question on backends > On Mon, 2002-07-29 at 20:21, Luis Alberto Amigo Navarro wrote: > > if i put debug_level=1 i get for one connect and several inserts on backend > > die after each insert > > What client do you use ? > > ---------- > Hannu > > > ---------------------------(end of broadcast)--------------------------- > TIP 4: Don't 'kill -9' the postmaster > >
On Mon, 2002-07-29 at 20:50, Luis Alberto Amigo Navarro wrote: > libpq > PQsetdb(........ > ----- Original Message ----- > From: "Hannu Krosing" <hannu@tm.ee> > To: "Luis Alberto Amigo Navarro" <lamigo@atc.unican.es> > Cc: <pgsql-hackers@postgresql.org> > Sent: Monday, July 29, 2002 9:40 PM > Subject: Re: [HACKERS] question on backends > > > > On Mon, 2002-07-29 at 20:21, Luis Alberto Amigo Navarro wrote: > > > if i put debug_level=1 i get for one connect and several inserts on > backend > > > die after each insert > It should not happen. I've run several websites using both php and python (which use libpq to connect to backend), and they have been up for months connected to the _same_ backend, doing inserts, updates, deletes and selects. I guess there is some error in you code that manifests itself as dying backends. You can test this by trying to do several inserts in one transaction. and see if the data is still there after the backend closes after the first insert. -------------- Hannu
Ah yes - that was me making an unfortunate exptrapolation without thinking it through. Of course, PHP implements persistent connections for you, etc., etc., not the postgres client library. Chris > -----Original Message----- > From: Neil Conway [mailto:nconway@klamath.dyndns.org] > Sent: Tuesday, 30 July 2002 12:11 AM > To: Christopher Kings-Lynne > Cc: Luis Alberto Amigo Navarro; pgsql-hackers@postgresql.org > Subject: Re: [HACKERS] question on backends > > > On Mon, Jul 29, 2002 at 11:28:54PM +0800, Christopher Kings-Lynne wrote: > > libpq has a function pconnect as opposed to connect that will do it. > > libpq has neither function, AFAIK. > > As for persistent backends, it's on the TODO list, but I'm not aware > that anyone has put any work into implementing it. At the moment, > a backend connects to a single database for its entire lifecycle -- so > you'd either need one pool of persistent backends for each database > in use, or you'd need to allow a backend to change the database it is > connected to. > > Cheers, > > Neil > > -- > Neil Conway <neilconway@rogers.com> > PGP Key ID: DB3C29FC >