RE: Shared Memory Segments are not removed

I did open up a case with Tech Support and they advised me post this problem on IIUG. I guess I am stuck!! > -----Original Message----- > From: Obnoxio The Clown [SMTP:obnoxio@hotmail.com] > Sent: Wednesday, March 08, 2000 9:37 AM > To: SArshad@purolator.com; informix-list@iiug.org > Subject: RE: Shared Memory Segments are not removed > > From: "Arshad, Shehla" <SArshad@purolator.com> > > > >Yes, we have tried ipcrm as root but that does not remove the segments. > >Unix > >complains that the segment id is invalid. That may be because the > instances > >have been marked as "to be removed" and the key of those segments looks > >like > >0x00000000. > > This does not sound even remotely good. Have you logged a case with Tech > Support? > > >No, we run onmode -ky once. Do you mean that running onmode -ky > repeatedly > >would remove the segments in question? > > It *might*. > > >There are two instances on that server so we noramlly have four > (Informix) > >shared memory segments when the instances are online (i.e two per > >instnace). > >After bringing down both instances we saw two segments still hanging in > >there. We could tell by the sizes of those segments that they both > belonged > >to one of the instances. > > Hm. Nope, no idea... > ______________________________________________________ > Get Your Private, Free Email at http://www.hotmail.com

The answer is that both ipcs and ipcrm are absolutely correct. Ipcs is reporting that the shared memory IDs still exist, this can be because some process is still attached to these segments (not neccessarily an Informix process but a shared memory client process) but has not tried to write to them recently and so has not encountered the 'invalid id' error and detached or exited (or perhaps it has gotten the error but is not trapping or is ignoring it). Ipcrm is correct that the id is invalid because the last ONINIT process removed all the segments on its way down. OK, so if the segment has been removed why does ipcs still show it? It does not. Ipcs just reports that there is a shared memory ID with some processes attached to it and as long as something is still attached to the defunct segment ids their header records, which is what ipcs reads, cannot be removed from the table. Not to worry, this condition will NOT prevent you from restarting your instances! I have it all the time. If an engine crashes I tell the console folk to move users to other servers and shutdown the application servers talking to the dead server and they move the users right away but are usually too busy to bounce the app servers. Meanwhile I cannot wait to restart the engines so I just destroy any remaining memory segments (yes it is usually enough to run onmode -ky again or failing that ipcrm) and restart the engine. No trouble and yes there are defunct segments hanging around. Art S. Kagel Arshad, Shehla wrote: > > I did open up a case with Tech Support and they advised me post this problem > on IIUG. I guess I am stuck!! > > > -----Original Message----- > > From: Obnoxio The Clown [SMTP:obnoxio@hotmail.com] > > Sent: Wednesday, March 08, 2000 9:37 AM > > To: SArshad@purolator.com; informix-list@iiug.org > > Subject: RE: Shared Memory Segments are not removed > > > > From: "Arshad, Shehla" <SArshad@purolator.com> > > > > > >Yes, we have tried ipcrm as root but that does not remove the segments. > > >Unix > > >complains that the segment id is invalid. That may be because the > > instances > > >have been marked as "to be removed" and the key of those segments looks > > >like > > >0x00000000. > > > > This does not sound even remotely good. Have you logged a case with Tech > > Support? > > > > >No, we run onmode -ky once. Do you mean that running onmode -ky > > repeatedly > > >would remove the segments in question? > > > > It *might*. > > > > >There are two instances on that server so we noramlly have four > > (Informix) > > >shared memory segments when the instances are online (i.e two per > > >instnace). > > >After bringing down both instances we saw two segments still hanging in > > >there. We could tell by the sizes of those segments that they both > > belonged > > >to one of the instances. > > > > Hm. Nope, no idea... > > ______________________________________________________ > > Get Your Private, Free Email at http://www.hotmail.com -- Art S. Kagel & Family kagel@erols.com