Обсуждение: Re: [GENERAL] postgres memory management
At 13:18 21/01/01 +0100, Alexander Jerusalem wrote: >Hi all, > >I'm experiencing some strange behaviour with postgresql 7.0.3 on Red Hat >Linux 7. I'm sending lots of insert statements to the postgresql server >from another machine via JDBC. During that process postgresql continues to >take up more and more memory and seemingly never returns it to the system. >Oddly if I watch the postmaster and it's sub processes in ktop, I can't >see which process takes up this memory. ktop shows that the postgresql >related processes have a constant memory usage but the overall memory >usage always increases as long as I continue to send insert statements. > >When the database connection is closed, no memory is reclaimed, the >overall memory usage stays the same. And when I close down all postgresql >processes including postmaster, it's the same. >I'm rather new to Linux and postgresql so I'm not sure if I should call >this a memory leak :-) >Has anybody experienced a similar thing? I'm not sure myself. You can rule out JDBC (or Java) here as you say you are connecting from another machine. When your JDBC app closes, does it call the connection's close() method? Does any messages like "Unexpected EOF from client" appear on the server side? The only other thing that comes to mine is possibly something weird is happening with IPC. After you closed down postgres, does ipcclean free up any memory? I'm cc'in the hackers list and the new jdbc list. Peter >thanks, >Alexander Jerusalemvknn
At 22:29 22.01.01, Alfred Perlstein wrote: >* Peter Mount <peter@retep.org.uk> [010122 13:21] wrote: > > At 13:18 21/01/01 +0100, Alexander Jerusalem wrote: > > >Hi all, > > > > > >I'm experiencing some strange behaviour with postgresql 7.0.3 on Red Hat > > >Linux 7. I'm sending lots of insert statements to the postgresql server > > >from another machine via JDBC. During that process postgresql > continues to > > >take up more and more memory and seemingly never returns it to the > system. > > >Oddly if I watch the postmaster and it's sub processes in ktop, I can't > > >see which process takes up this memory. ktop shows that the postgresql > > >related processes have a constant memory usage but the overall memory > > >usage always increases as long as I continue to send insert statements. > > > > > >When the database connection is closed, no memory is reclaimed, the > > >overall memory usage stays the same. And when I close down all postgresql > > >processes including postmaster, it's the same. > > >I'm rather new to Linux and postgresql so I'm not sure if I should call > > >this a memory leak :-) > > >Has anybody experienced a similar thing? > > > > I'm not sure myself. You can rule out JDBC (or Java) here as you say you > > are connecting from another machine. > > > > When your JDBC app closes, does it call the connection's close() method? > > Does any messages like "Unexpected EOF from client" appear on the > server side? > > > > The only other thing that comes to mine is possibly something weird is > > happening with IPC. After you closed down postgres, does ipcclean free up > > any memory? > >I don't know if this is valid for Linux, but it is how FreeBSD >works, for the most part used memory is never free'd, it is only >marked as reclaimable. This is so the system can cache more data. >On a freshly booted FreeBSD box you'll have a lot of 'free' memory, >after the box has been running for a long time the 'free' memory >will probably never go higher that 10megs, the rest is being used >as cache. > >The main things you have to worry about is: >a) really running out of memory (are you useing a lot of swap?) >b) not cleaning up IPC as Peter suggested. Thanks for your answer! I'm rather new to Linux, so I can't tell if it's that way on Linux. But I noticed that other programs free some memory when I quit them. But it's true that I'm not running out of memory. I have 300 MB of free RAM and no swap space is used. As I wrote in reply to Peters mail, ipcclean doesn't change anything. Alexander Jerusalem ajeru@gmx.net vknn
At 21:40 22.01.01, Peter Mount wrote: >At 13:18 21/01/01 +0100, Alexander Jerusalem wrote: >>Hi all, >> >>I'm experiencing some strange behaviour with postgresql 7.0.3 on Red Hat >>Linux 7. I'm sending lots of insert statements to the postgresql server >>>from another machine via JDBC. During that process postgresql continues to >>take up more and more memory and seemingly never returns it to the >>system. Oddly if I watch the postmaster and it's sub processes in ktop, I >>can't see which process takes up this memory. ktop shows that the >>postgresql related processes have a constant memory usage but the overall >>memory usage always increases as long as I continue to send insert statements. >> >>When the database connection is closed, no memory is reclaimed, the >>overall memory usage stays the same. And when I close down all postgresql >>processes including postmaster, it's the same. >>I'm rather new to Linux and postgresql so I'm not sure if I should call >>this a memory leak :-) >>Has anybody experienced a similar thing? > >I'm not sure myself. You can rule out JDBC (or Java) here as you say you >are connecting from another machine. > >When your JDBC app closes, does it call the connection's close() method? >Does any messages like "Unexpected EOF from client" appear on the server side? > >The only other thing that comes to mine is possibly something weird is >happening with IPC. After you closed down postgres, does ipcclean free up >any memory? > >I'm cc'in the hackers list and the new jdbc list. > >Peter Thanks for your answer! Yes I'm calling Connection.close(). I don't get any error messages but maybe I just don't see them because postgresql is started automatically at run level 3. I'm not sure where the output goes. (pg_log contains only garbage or maybe it's a binary file) I tried ipcclean right now and it doesn't free the memory but it gives me some messages that I cannot interpret: Shared memory 0 ... skipped. Process still exists (pid ). Shared memory 1 ... skipped. Process still exists (pid ). Shared memory 2 ... skipped. Process still exists (pid ). Shared memory 3 ... skipped. Process still exists (pid ). Semaphore 0 ... resource(s) deleted Semaphore 1 ... resource(s) deleted Oddly, when I try to run ipcclean a second time, it says: ipcclean: You still have a postmaster running. Which is not the case as ps -e proves. Alexander Jerusalem ajeru@gmx.net vknn
* Peter Mount <peter@retep.org.uk> [010122 13:21] wrote: > At 13:18 21/01/01 +0100, Alexander Jerusalem wrote: > >Hi all, > > > >I'm experiencing some strange behaviour with postgresql 7.0.3 on Red Hat > >Linux 7. I'm sending lots of insert statements to the postgresql server > >from another machine via JDBC. During that process postgresql continues to > >take up more and more memory and seemingly never returns it to the system. > >Oddly if I watch the postmaster and it's sub processes in ktop, I can't > >see which process takes up this memory. ktop shows that the postgresql > >related processes have a constant memory usage but the overall memory > >usage always increases as long as I continue to send insert statements. > > > >When the database connection is closed, no memory is reclaimed, the > >overall memory usage stays the same. And when I close down all postgresql > >processes including postmaster, it's the same. > >I'm rather new to Linux and postgresql so I'm not sure if I should call > >this a memory leak :-) > >Has anybody experienced a similar thing? > > I'm not sure myself. You can rule out JDBC (or Java) here as you say you > are connecting from another machine. > > When your JDBC app closes, does it call the connection's close() method? > Does any messages like "Unexpected EOF from client" appear on the server side? > > The only other thing that comes to mine is possibly something weird is > happening with IPC. After you closed down postgres, does ipcclean free up > any memory? I don't know if this is valid for Linux, but it is how FreeBSD works, for the most part used memory is never free'd, it is only marked as reclaimable. This is so the system can cache more data. On a freshly booted FreeBSD box you'll have a lot of 'free' memory, after the box has been running for a long time the 'free' memory will probably never go higher that 10megs, the rest is being used as cache. The main things you have to worry about is: a) really running out of memory (are you useing a lot of swap?) b) not cleaning up IPC as Peter suggested. -- -Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org] "I have the heart of a child; I keep it in a jar on my desk."