RE: bzr commit into mysql-6.0-falcon:mysql-6.0-falcon branch (vvaintroub:2685) Bug#36396

[Vladislav Vaintroub]

> -----Original Message-----
> From: Kevin Lewis [mailto:klewis@stripped]
> Sent: Tuesday, June 10, 2008 12:05 AM
> To: 'Vladislav Vaintroub'; 'Vladislav Vaintroub';
> commits@stripped
> Cc: 'Chris Powers'
> Subject: RE: bzr commit into mysql-6.0-falcon:mysql-6.0-falcon branch
> (vvaintroub:2685) Bug#36396


> So if I understand it, what used to happen is that there were
> infrequent
> errors on CREATE with "file already exists" message. 

Exactly, we've seen this messages sometimes happen on pushbuild, typically
after DROP TABLESPACE IF EXISTS foo engine=falcon; CREATE TABLESPACE foo
engine=falcon sequence.


> Now, it may get the same message, but only after waiting 10 seconds.
Given the infrequency
> of the problem.  And the fact that the failed create tablespace can be
retried with probably success.  I agree that simple is better.

Yes. Also, polling/waiting  happens only, if there is at least one "drop
tablespace" in progress. It does not always take 10 seconds.The polling
frequency is 1 sec and in all configuration I tried client thread waited for
at most 1 sec .Except for one case, when I deliberately suspended DROR
TABLESPACE gopher using debugger and had the "file already exists" message
in client thread after 10 seconds wait.