[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference orarep::nomahs::rdb_60

Title:	Oracle Rdb - Still a strategic database for DEC on Alpha AXP!
Notice:	RDB_60 is archived, please use RDB_70..
Moderator:	NOVA::SMITHISON

Created:	Fri Mar 18 1994
Last Modified:	Fri May 30 1997
Last Successful Update:	Fri Jun 06 1997
Number of topics:	5118
Total number of notes:	28246

4974.0. "Hang on COMMIT - "writing to ruj block"" by M5::BLEHLBAC (RDB: 34% better than real life) Tue Jan 28 1997 13:25

rdb 6.1a
axp,vms 6.1

Hi,

  This is similar to note 3224.

  Customer did an insert transaction that built and ruj file of 46k.

  A commit was issued and the process hung.  The customer let things hang
for 1 1/2 hours.

  rmu/show stat - stall messages showed "writing to ruj block".

  sho/locks/mode=blocking didn't show anything.

  show/locks/full/mode=waiting showed 2 processes - the user and the
monitor - the monitor was waiting.

  The monitor log didn't show anything of interest.

  The customer recovered by killing the process  - which forced a rollback
and then reran the job - which worked ok.

  Any further ideas or enlightenments on this??

                             Thanks,  Barry

T.R	Title	User	Personal Name	Date	Lines
4974.1		DUCATI::LASTOVICA	Is it possible to be totally partial?	`Tue Jan 28 1997 14:38`	6
	> rmu/show stat - stall messages showed "writing to ruj block". Rdb was waiting for an I/O to the RUJ file to finish. We've fixed a couple of possible hangs on Alpha systems, but I would have expected them to have been fixed by 6.1A. So I don't know what to tell ya.
4974.2		NOVA::R_ANDERSON	Oracle Corporation (603) 881-1935	`Tue Jan 28 1997 14:51`	7
	We've fixed a large number of I/O hangs in Rdb7 related to interrupts from RMS. There is no pre-Rdb7 work-around available other than set your DCL "SET OUTPUT_RATE=n" really large (and cross your fingers). The changes made to Rdb7 to fix this problem CANNOT be back-ported. Rick
4974.3	Confused about RMS hangs	NOMAHS::SECRIST	Rdb WWS; [email protected]	`Thu Mar 06 1997 16:40`	32
	re: .1 ; Rdb was waiting for an I/O to the RUJ file to finish. We've ; fixed a couple of possible hangs on Alpha systems, but I would ; have expected them to have been fixed by 6.1A. Would they have been fixed by 6.1 eco 4 ? re: .2 ; We've fixed a large number of I/O hangs in Rdb7 related to ; interrupts from RMS. Is there any place we can read about these ? I didn't see them in the preliminary 7.0 eco 1 release notes. ; ... ; There is no pre-Rdb7 work-around available other than set ; your DCL "SET OUTPUT_RATE=n" really large (and cross your ; fingers). The changes made to Rdb7 to fix this problem ; CANNOT be back-ported. What exactly does setting the output rate do ? With a default rate of 60 seconds I'm really confused. Are there any known issues regarding these problems and hangs of the monitor process ? Regards, rcs
4974.4	disregard the release notes for now	NOVA::MCGEE	Oracle Rdb Mission Critical Engineering	`Thu Mar 06 1997 17:07`	9
	>Is there any place we can read about these ? I didn't see them >in the preliminary 7.0 eco 1 release notes. The release notes for ECO1 have not been built yet so I included the release notes for some other release just as a placeholder for now. In other words, the release notes in the preliminary kit do not represent fixes in this kit. Steve.
4974.5	How do these problems apply to V6.* ?	NOMAHS::SECRIST	Rdb WWS; [email protected]	`Fri Mar 07 1997 05:34`	41
	re: .4 ; ...release notes for ECO1 have not been built yet... Thanks, Steve. What I'm really after is whether how these bugs affect V6.1 eco 4 or V6.1-1. re: .1 ; Rdb was waiting for an I/O to the RUJ file to finish. We've ; fixed a couple of possible hangs on Alpha systems, but I would ; have expected them to have been fixed by 6.1A. Would they have been fixed by 6.1 eco 4 ? re: .2 ; We've fixed a large number of I/O hangs in Rdb7 related to ; interrupts from RMS. Is there any place we can read about these ? How do these problems affect V6.* ? ; ... ; There is no pre-Rdb7 work-around available other than set ; your DCL "SET OUTPUT_RATE=n" really large (and cross your ; fingers). The changes made to Rdb7 to fix this problem ; CANNOT be back-ported. What exactly does setting the output rate do ? With a default rate of 60 seconds I'm really confused. Are there any known issues regarding these problems and hangs of the monitor process in V6.* ? Regards, rcs
4974.6		NOVA::R_ANDERSON	Oracle Corporation (603) 881-1935	`Fri Mar 07 1997 06:39`	18
	While the RMS-interrupt problems are possible in Rdb v6.0 and v6.1, we don't normally encounter these types of problems. The basically reason Rdb7 has such problems with interrupts is that ALL I/O (rootfile, AIJ, RUJ, storage areas, etc) is asynchronous. Also, Rdb7 uses many more timers (for various performance algorithms). All this means is that Rdb7 relies on AST delivery for notification of event completion, whereas Rdb v6.1 and earlier used event flags and synchronous operations for event completion notification. And, since every in Rdb7 is asynchronous, RMS is almost guaranteed to interrupt an Rdb7 AST delivery. So, the answer is "yes" these problems are possible in Rdb v6.1 but not as common. The changes made to Rdb7 to correct these problems are extensive and extremely complex, and cannot be back-ported to earlier versions. Rick
4974.7	Large SET OUTPUT_RATE workaround ?	NOMAHS::SECRIST	Rdb WWS; [email protected]	`Sat Mar 08 1997 20:23`	15
	re: .2 ; ... ; There is no pre-Rdb7 work-around available other than set ; your DCL "SET OUTPUT_RATE=n" really large (and cross your ; fingers). The changes made to Rdb7 to fix this problem ; CANNOT be back-ported. Since the default for SET OUTPUT RATE is 60 seconds why does making it large make any difference ? Thanks, rcs
4974.8		HOTRDB::LASTOVICA	Is it possible to be totally partial?	`Sat Mar 08 1997 20:39`	2
	it reduces the chances that DCL will be trying to do a log file flush when an Rdb AST is delivered.