Bug submitted. Since the Trace file is 1.9Mb, I only posted a portion of
it. The data looks pretty much the same from that point on, but if you
would like the entire file I will be happy to email it to you. I
increased ZUNDOPAGESIZE to 1024, recompiled and copied the binaries to
all my servers, but still it crashes.
Mikael Ronström wrote:
> Hi Devananda,
> 2004-07-08 kl. 21.04 skrev Devananda:
>> Jonas Oreland wrote:
>>> A better way is to instead bulk rows to about 64k and the commit.
>>> So, each insert should contain roughly 64k and there should be
>>> autocommit on.
>> The insert statements are the direct result of mysqldump; I merely
>> grabbed the first complete line. In looking at the options for
>> mysqldump I dont see anything allowing me to control how long each
>> insert statement that it creates will be. If the optimum size of an
>> insert is 64k, is there a way to get mysqldump to create inserts of
>> just that size?
> The problem you are experiencing sounds very much like your disk is
> much slower compared to your CPU as Jonas mentioned.
> The problem you are experiencing is that the NDB kernel produces UNDO
> log records faster than the disk can swallow them. Since
> for inserts the UNDO log records are created at the time of insert and
> not at commit that points to this problem even more. At the same
> time there is a protection barrier that should be invoked to ensure
> this problem doesn't occur but rather that the behaviour is aborted
> So I suggest you file a bug report on that one together with the trace
> file so that we can check it up. Please also describe the HW and
> the query so that we get an idea of the environment. How long is very
> long and how many inserts is pressed into this single line.
> If the problem persists there is a compile time parameter at line 197
> in Dbacc.hpp
> ZUNDOPAGESIZE set to 64 (number of 32 kByte pages). You can set this
> to a higher
> value like 128, 256, 512, 1024. At the moment it is 1 MByte. So far we
> haven't seen the
> need to make it a configurable parameter.
> Rgrds Mikael
>> MySQL Cluster Mailing List
>> For list archives: http://lists.mysql.com/cluster
>> To unsubscribe: http://lists.mysql.com/cluster?unsub=1
> Mikael Ronström, Senior Software Architect
> MySQL AB, www.mysql.com