|
Problem IT12508
|
Status: Closed |
IPL THROTTLING ON SYSTEMS WITH LARGE LOGICAL LOG SPACE CAN SEE THE IPL THREAD YIELD FOR LONG PERIODS OF TIME
|
| product: |
INFORMIX SERVER / 5725A3900 / B70 - IDS 11.70 |
| Problem description: |
The IPL thread breaks down an index copy into bite sized pieces
of work to avoid the transfer turning into a long transaction.
At the commit phase of each of these sub transactions, a check
is made to make sure the IPL thread is not flooding the logical
logs, and prematurely forcing other unrelated transactions into
a long transaction state. To do this, a check is made to see
whether the instance has advanced LTXHWM * 0.5 beyond the oldest
transaction, and if it has, the IPL thread will go into a wait
state (mt_yield) in ipl_commit.
Example stack of IPL thread waiting for other transactions to
catch up:
Stack for thread: 33527735 ipl
base: 0x07000001cff08000
len: 69632
pc: 0x00000001000620f0
tos: 0x07000001cff18740
state: sleeping
vp: 9
(oninit)yield_processor_mvp
(oninit)mt_yield
(oninit)ipl_commit
(oninit)ipl_traverseidx
(oninit)ipl_logidxpgs
(oninit)ipl_logidxpgs_fd
(oninit)ipl_logidxpgs_partn
(oninit)ipl_startup
(oninit)th_init_initgls
(oninit)startup
Problems can occur if you have sessions which start a
transaction, and then become idle, which means the IPL thread
can become blocked until those sessions are either terminated by
the client or hit the LTXHWM and are rolled back by the engine.
If the instance is configured with a large amount of logical log
space, the IPL thread can become blocked for a long time. |
| Problem Summary: |
****************************************************************
* USERS AFFECTED: *
* Users enabling Index Page Logging *
****************************************************************
* PROBLEM DESCRIPTION: *
* See Error Description *
****************************************************************
* RECOMMENDATION: *
* Update to IBM Informix Server 11.70.xC9 *
**************************************************************** |
| Local Fix: |
To allow the IPL thread to continue, you might consider finding
those sessions and forcing a commit or rollback, or terminate
them with an 'onmode -z'. |
| Solution |
|
| Workaround |
not known / see Local fix |
| Timestamps |
Date - problem reported :
Date - problem closed :
Date - last modified :
| 01.12.2015
09.06.2017
09.06.2017 |
| Problem solved at the following versions (IBM BugInfos) |
11.70.xC9
|
| Problem solved according to the fixlist(s) of the following version(s) |
