Hema,
...and my replies to you.
> > - Why is the backup_views test being deleted? Please take
> that out of
> > the patch.
>
> Hema: I created backup_views.test in backup/suite for an
> internal sample test that I was performing (not a valid test
> scenario and I was using for a quick check). Its not related
> to the original backup_views.test that is in main suite.
Ok, but this is a good example of why you need to get your gcommit working
so you can add per-file comments. A comment placed in the patch would have
prevented me from having to ask this (and similar) questions. :)
> > - The backup_partition test needs clarification. Why are
> character set
> > specifics included?
> > - The backup_datatypes test is incomplete and also needs
> > clarification. Why character set specifics?
>
> Hema: When I checked with Bar about this, he mentioned that I
> need to use accented characters for this WL. Anyways I can
> remove it if you feel unnecessary and will create a new patch
> with the changes.
No, Bar is the undisputed expert in that area. Go with what he recommends
but please add a comment explaining and documenting the decision.
> > - Running tests against all storage engines? Why?
>
> Hema: As discussed in IRC, I feel that it's good to execute
> some test cases with all storage engines. We need to decide
> the tests that need to be run with all storage engines.
I think we need to create a policy to govern/guide this decision. Please
make a stab at creating what you think is the proper guidelines and let's
add this to the next Backup Team meeting.
> Hema: Ok, I will remove accented characters from partition
> test. I have separate test for special characters and
> accented characters which are failing now, so included those
> tests in the bug report (developer who is working on this
> bug should push those test cases)
Good.
> > Likewise, I don't think we need to combine special
> character testing in the
> > datatypes test. Besides, it is just the table and database
> names that use
> > the characters. That isn't a complete test and doesn't
> belong in a test that
> > is (seems to be) designed to test how backup can handle
> different data
> > types. In that case, the test is incomplete because it
> doesn't contain all
> > possible data types.
>
> Hema: I am not very clear here. I had included all datatypes
> that I obtained from manuals. Do you see anything missing
> here? Can you please explain me more clearly on what are the
> other test scenarios that you are looking for?
Yes, lots of datatypes are missing. What about text and binary and all the
forms of each? Please see
http://dev.mysql.com/doc/refman/6.0/en/data-type-overview.html and go
through each class (numeric, dte, string) to verify you're checking all data
types. However, I think I should point out here that this is and example of
the type of test that does not warrant running against all storage engines.
> Hema: I have separate test for special characters and
> accented characters which are failing now, so attached them in the
> bug report. As said before, intention is to have separate
> test case for better coverage and triage in case of failures.
Ok. Please document this in the tests so that someone doesn't ask this
question again in the future.
> Hema: My intention is to unveil the defects pertinent to
> different datatypes / different engines with respect of backup
> and restore (if exists). So I would definitely prefer to run
> at least few selected tests with all engines and we need to
> decide on that. Partition test - ** Fails for Falcon storage
> engine only while performing backup **. So its imperative
> that I verify partition test with all storage engines.
The backup system has nothing to do with how storage engines store their
data. With the (unlikely) possible exception of blobs, the backup system
uses the standard MySQL replication internal storage for records processed
by the default drivers. The default drivers are the only part of the backup
system that works with the actual storage of the data in the storage
engines. However, I must stress that this is a layer above the storage
engine and not connectect whatsoever to backup. Backup uses the internal
record format. What happens beyond that is not a concern (for backup).
The error for Falcon and backup has to do with the drivers and not the data
types or the partition engine (specifically).
Please do not run the test against all storage engines -- it isn't needed
and the logic used to justify is nullified by the conditions of the failure
with Falcon and the fact that the backup system has nothing to do with how
datatypes are stored or processed in the storage engines.
Chuck
