Re: pgAdmin 4 commit: Ensure we pick up the messages from the current query

Hello Khushboo,
The patch runs successfully in our CI with all tests passing.

I see the test that you created, and I do not understand why we need to
create tests that do HTTP requests in order to check something that is
executed against the database. What I was talking about in my previous
email was having tests that tested the function by itself.

(Copied from: https://jfiaffe.files.wordpress.com/2014/09/tests-pyramid.png)

This is the Testing Pyramid, there are a bunch of different drawings of it
and ways to explain it, but in broad stokes what is means is that we should
have the majority of tests around a Unit (that are some disagreements in
the community of what a Unit is) and a very small amount of Manual testing.
What Unit usually means is piece of code, it can be a function, it can be a
class or it can even be a module, but is something self contained. In
pgAdmin's case the majority of our tests go around the Integration Layer
because we are using HTTP requests in order to executing queries in the
database, so basically we are doing tests end to end in the backend, and
the cost time.

I do not want to held this patch back because of this, and I say this
because I have minimal confidence with the tests that you created, that
they would catch the majority of the problems, and hope that the majority
of the code is exercised by it.

Nevertheless I would like to challenge all the Hackers to think about
testing in a different way. The tests in our code are used to give us
confidence that the work we did is working as expected, this also makes it
much easier to refactor out bad patterns or very complicated ones into
something simple. A signal that our code is more complicated then it should
is when we need to test some behavior and we end up with a Stubbing Hell or
we need to test it End 2 End because it is to hard to isolate the part we
want to test. In the other hand we should not test all functions and every
class, because we might be coupling our tests to much to the implementation
and that will have the contrary effect, and we will not be able to refactor
and simply our code.
Like everything in life there need to be a balance.

Thanks
Joao

On Thu, Mar 1, 2018 at 12:56 AM Khushboo Vashi <
khushboo(dot)vashi(at)enterprisedb(dot)com> wrote:

> Hi Joao,
>
> Thanks for reviewing.
>
> On Wed, Feb 28, 2018 at 8:55 PM, Joao De Almeida Pereira <
> jdealmeidapereira(at)pivotal(dot)io> wrote:
>
>> Hello Khushboo,
>> After reviewing the patch I have the gut feeling that we do not have
>> enough test coverage on this issue, specially due to the intricate while
>> loop and conditions around the polling.
>> I think that this deserve Unit tests around it, When I say Unit Test I am
>> not talking about executing queries against the database, but do some
>> stubbing of the database so that we can control the flow that we want.
>>
> You are right. It needs more unit testing. I have checked below scenarios:
> 1. Returns 2 notices with data output
> 2. Returns 1000 notices with data output
> 3. No notices with data output
>
> By running above, I have checked, each time returned notices are accurate,
> no old notices are getting appended, it does not affect with the amount of
> messages (few, none or more). Also, with the updated patch, I have made
> sure that all these queries run with the single transaction id (same
> connection).
>
> So, please let me know if you think I can add more things to this.
>
>>
>>
> It is a temptation to try to always do a Feature Test to test what we want
>> because it is "easier" to write and ultimately it is what users see, but
>> while 1 Feature Test runs we can run 200 Unit Tests that give us much more
>> confidence that the code is doing what we expect it to do.
>>
>> Right, so added regression tests instead of feature tests.
>
> This being said, I run the tests on the CI Pipeline and all tests pass.
>> Running pycodestyle fails due to some line sizes on the
>> psycopg2/__init__py. I believe that it is not what you changed, but since
>> you were changing the file it can be fixed it is just:
>>
>> pgadmin/utils/driver/psycopg2/__init__.py:1276: [E501] line too long (81
>> > 79 characters)
>> pgadmin/utils/driver/psycopg2/__init__.py:1277: [E501] line too long (91
>> > 79 characters)
>> pgadmin/utils/driver/psycopg2/__init__.py:1282: [E501] line too long (81
>> > 79 characters)
>> pgadmin/utils/driver/psycopg2/__init__.py:1283: [E501] line too long (91
>> > 79 characters)
>> 4 E501 line too long (81 > 79 characters)
>>
>> Fixed. Thanks for pointing out.
>
>>
>> Thanks
>> Joao
>>
>>
>> On Wed, Feb 28, 2018 at 6:49 AM Khushboo Vashi <
>> khushboo(dot)vashi(at)enterprisedb(dot)com> wrote:
>>
>>> On Mon, Feb 26, 2018 at 10:02 PM, Dave Page <dpage(at)pgadmin(dot)org> wrote:
>>>
>>>> Argh, I ran some tests, but didn't spot any lost messages in the tests
>>>> I ran. I'll revert the patch.
>>>>
>>>> Khushboo;
>>>>
>>>> Please look at the following:
>>>>
>>>> - Fix the patch so it doesn't drop messages.
>>>>
>>> Fixed.
>>> By default, the notice attribute of the connection object of psycopg 2
>>> only stores 50 notices. Once it reaches to 50 it starts from 1 again.
>>> To fix this I have changed the notice attribute from list to deque to
>>> append more messages. Currently I have kept the maximum limit at a time of
>>> the notice attribute is 100000 (in a single poll).
>>>
>>>> - Add regression tests to make sure it doesn't break in the future.
>>>> This may require creating one or more functions the spew out a whole lot of
>>>> notices, and then running a couple of queries and checking the output.
>>>>
>>> Added. With this regression test, the current code is failing which has
>>> been taken care in this patch.
>>>
>>>> - Check the messages panel on the history tab. I just noticed it seems
>>>> to only be showing an even smaller subset of the messages.
>>>>
>>> Tested and no issues found.
>>>
>>>>
>>>>
>>> Thanks.
>>>>
>>>> On Mon, Feb 26, 2018 at 4:23 PM, Murtuza Zabuawala <
>>>> murtuza(dot)zabuawala(at)enterprisedb(dot)com> wrote:
>>>>
>>>>> Sent bit early,
>>>>>
>>>>> You can run 'VACUUM FULL VERBOSE' in query tool and verify the
>>>>> populated messages (pgAdmin3 vs. pgAdmin4).
>>>>>
>>>>>
>>>>> On Mon, Feb 26, 2018 at 9:48 PM, Murtuza Zabuawala <
>>>>> murtuza(dot)zabuawala(at)enterprisedb(dot)com> wrote:
>>>>>
>>>>>> Hi Khushboo/Dave,
>>>>>>
>>>>>> With given commit, I'm again seeing the issue raised in
>>>>>> https://redmine.postgresql.org/issues/1523 :(
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Regards,
>>>>>> Murtuza Zabuawala
>>>>>> EnterpriseDB: http://www.enterprisedb.com
>>>>>> The Enterprise PostgreSQL Company
>>>>>>
>>>>>>
>>>>>> On Mon, Feb 26, 2018 at 7:49 PM, Dave Page <dpage(at)pgadmin(dot)org> wrote:
>>>>>>
>>>>>>> Ensure we pick up the messages from the current query and not a
>>>>>>> previous one. Fixes #3094
>>>>>>>
>>>>>>> Branch
>>>>>>> ------
>>>>>>> master
>>>>>>>
>>>>>>> Details
>>>>>>> -------
>>>>>>>
>>>>>>> https://git.postgresql.org/gitweb?p=pgadmin4.git;a=commitdiff;h=08b3ccc01a4d57e8ea3657f8882a53dcd1b99386
>>>>>>> Author: Khushboo Vashi <khushboo(dot)vashi(at)enterprisedb(dot)com>
>>>>>>>
>>>>>>> Modified Files
>>>>>>> --------------
>>>>>>> web/pgadmin/utils/driver/abstract.py | 1 +
>>>>>>> web/pgadmin/utils/driver/psycopg2/__init__.py | 64
>>>>>>> +++++++++------------------
>>>>>>> 2 files changed, 21 insertions(+), 44 deletions(-)
>>>>>>>
>>>>>>>
>>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Dave Page
>>>> Blog: http://pgsnake.blogspot.com
>>>> Twitter: @pgsnake
>>>>
>>>> EnterpriseDB UK: http://www.enterprisedb.com
>>>> The Enterprise PostgreSQL Company
>>>>
>>>