Kbase P136018: Database hanging periodically and user processes seem to be be performing a semop call.
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
11/19/2008 |
|
Status: Verified
SYMPTOM(s):
Database hanging periodically
Overall performance of the databases decreases
It takes a long time to login in the application when users already logged in experience delays in the aplication
Database appears to be hanging.
Periodically, user processes will all of a sudden seem to lock up or hang.
truss output produced from these processes indicate that these processes appear to be doing something with a semaphore call (semop).
Hang will be cleared up if DBA terminates all processes holding or waiting for a latch.
Promon shows blocked users
Blocked clients waiting on BHT (Buffer Hash Table)
Promon -> Status -> All Processes screen is showing multiple users waiting on the LKT (Lock Table latch)
Promon -> Status -> All Processes screen is showing multiple users waiting for a BUF (Buffer)
Promon shows high Waiting on semaphores
Cannot modify or tune -spin parameter within promon because the database was not started with -spin.
FACT(s) (Environment):
Operating System does not indicate a bottleneck on system resources (like disk IO or memory)
CPU is almost 100% idle
Machine where database is running has multiple CPUs.
Not making use of the -semsets database startup parameter.
Users can log into the database.
Enterprise Database License
UNIX
IBM AIX
Progress 9.x
OpenEdge 10.x
CAUSE:
Not making use of the -spin parameter, so all processes are using the semaphore set for locking.
FIX:
1. Modify database startup parameters to make use of the -spin parameter. Set -spin to 6000 or 10000 per CPU.
Example: If system contains 2 CPUs, set -spin either 12000 or 20000 and tune as needed.
2. Add -semsets parameter to database startup parameters. Set -semsets parameter to 1 semaphore set for every 50+ users in the database.
Example: 100 users, set -semsets 2