Description:
SysQA (Philip) ran iuds6 (the main stress test, involves fairly large
transactions and runs with 100 threads for 24 hours) on Summit v0.2.0
and found that the default InnoDB buffer pool sized is too small. Results
shows that the next version of Summit needs a default InnoDB buffer pool
of 1GB.
How to repeat:
Run iuds6 with 100 threads for 24 hours. This test passes with the default mysqld settings on our standard releases. With Summit v0.2.0, however, after
2 hours, Innodb returned:
o 090130 15:09:33 InnoDB: WARNING: over 67 percent of the buffer pool is occupied by lock heaps or the adaptive hash index! Check that your ...
o 090130 15:24:51 InnoDB: ERROR: over 95 percent of the buffer pool is occupied by lock heaps or the adaptive hash index! Check that your transactions do not set too many row locks. Your buffer pool size is 8 MB. Maybe you should make the buffer pool bigger? We intentionally generate a seg fault to print a stack trace on Linux!
o 090130 15:24:51 InnoDB: Assertion failure in thread 1155688768 in file buf/buf0lru.c line 388
Philip's comment:
I have never seen this error before on this particular test. When I re-ran
the test with innodb-buffer-pool=1GB, the test passed without issues and
nothing was printed in the log. This means that some workloads that worked
with the standard releases and default settings will now cause Innodb to
crash. People that are running badly-configured Innodb servers may be in for
a surprise. In order to avoid having to blame the users for not adjusting the
default upwards, maybe we should ship this version with a higher default?