| Register | FAQ | Calendar | Search | Today's Posts | Mark Forums Read |
|
#1
|
| At 3:12 PM +0200 5/18/07, Jean-Pierre Zuate wrote: >Hello Ingres Community, > >We have 5 DBMS definitions, all with cache_sharing = ON > >The five dmf_tbl_cache_size are equivalent (set to 150). > >If one of the DBMS definition of dmf_tbl_cache_size is : >- below the other values : session hang, and then entire installation hang. The closed-table info cache controlled by dmf_tbl_cache_size is stored in the shared buffer cache memory (the first one, if you have multiple caches with dmf_separate ON). The hang is caused by mismatch between the addresses computed by the various servers. I guess 2.6 must still have vestiges of the original 6.x cache-sharing code. In 6.x, because there was no config.dat, each dbms server had its own rundbms.opt and they had to match. With config.dat, the first server up can set up the shared cache and the other servers can simply attach, except that's not how the code worked until fairly recently. In 2.6 each server is still calculating most of the shared memory layout independently and they have to match. I think there is SOME attempt to force whatever cache parameters the cache was built with, but apparently dmf_tbl_cache_size isn't one of the things that get fixed up. This is fixed in 9.1.0 (maybe 9.0.4, not sure) such that servers attaching to an existing cache simply take what's already there, and ignore all the config cache parameters. Karl |
![]() |
| Thread Tools | |
| Display Modes | |