Informix 12.10 FC4 ON RHEL6
User encountered errno=16 when initializing Informix 12.10 FC4 raw chunks on RHEL6 using symbolic links to /dev/raw devices, despite correct permissions. Marcus Haarmann identified this as a known bug in version 12.10 FC4 and recommended upgrading to patch FC4W1 to resolve the issue and continue using raw chunks as the recommended storage method.
Auto-generated by DrWatson from the posts below — may be imperfect; read the full thread.
Hi Guys I need to know what's recomendation to use in chunks over RHEL6. Until RHEL5 i was using raw devices. I should use cooked files? Thanks Regards
When i used raw devices through udev i received this log message: "Cannot Open Primary Chunk "/opt/IBM/informix/dev/root_dbs", errno=16" root_dbs is a symbolic link to "/dev/raw/raw1", this device has right permissions and owner.
Hi, the recommendation did not change. Raw chunks are the fastest access method over files (even with direct io). But the version 12.10FC4 has a known bug when initializing raw chunks. Get the latest patch FC4W1 and you can go on like you used to work with Informix. We have also struggled a day with this bug, before we tried to initialize the DB using 11.70 release and this worked. Marcus Haarmann ----- Ursprüngliche Mail ----- Von: "LEONARDI CARDOZO" <leonardi.cardozo@intl.pepsico.com> An: ids@iiug.org Gesendet: Dienstag, 16. September 2014 01:25:37 Betreff: Re: Informix 12.10 FC4 ON RHEL6 [33785] When i used raw devices through udev i received this log message: "Cannot Open Primary Chunk "/opt/IBM/informix/dev/root_dbs", errno=16" root_dbs is a symbolic link to "/dev/raw/raw1", this device has right permissions and owner. ******************************************************************************* Forum Note: Use "Reply" to post a response in the discussion forum.
And in addition to fastest IO method, happens the following: when IFMX writes on a raw device, the write is garanteed as done, when IFMX writes on a cooked file, it writes to the buffered file system, which is not quite the same. In this case, in case of system crash, you may lose the write.... So the only way to commit the IO operations is to use RAW devices . Eric
I haven't seen the code, but that can't be true..... the write function has options to guarantee the write. Otherwise no database could assure consistency on cooked files which is not the case. Regards. Em 20/09/2014 20:19, "ERIC VERCELLETTO" <eric.vercelletto@begooden-it.com> escreveu: > And in addition to fastest IO method, happens the following: > when IFMX writes on a raw device, the write is garanteed as done, > when IFMX writes on a cooked file, it writes to the buffered file system, > which is not quite the same. In this case, in case of system crash, you may > lose the write.... > > So the only way to commit the IO operations is to use RAW devices . > > Eric > > > > ******************************************************************************* > Forum Note: Use "Reply" to post a response in the discussion forum. > > --089e013c66145bd43e050384803f
Fernando,
you are right, my mistake! I was stating something that was true before V4.00.
Cooked devices file chunks are opened by Informix in mode O_SYNC or O_DIRECT.
One can confirm that by running lsof +fg, and search for 'SYN'
(synchronized)in the FILE-FLAG command (this is linux).
COMMAND PID USER FD TYPE FILE-FLAG DEVICE SIZE/OFF NODE NAME
oninit 6922 informix 260u REG RW,SYN,DIR,LG 253,2 12778799104 5292035
/ids_chunks/ids_space01/v1210/bench_esql_data04.P.db
Therefore forget what I have written above, Informix writes are ALWAYS SAFE
after Informix v4.00 :-)
and ... thank you Art