CI: Travis failure in master because of GBQ table

[datapythonista]

Looks like there is a problem with GBQ. A test is finding that a table already exists. It's making Travis fail master. Didn't check in detail, the log is here: https://travis-ci.org/pandas-dev/pandas/jobs/629453901#L2297

@tswast mind having a look?

CC: @jreback

[jreback]

#30447 could have broken this

@datapythonista if you want to xfail this for now would be great.

[alimcmaster1]

Yep looks like the table added isn't removed in the teardown_class

In test_gbq.py :
destination_table = DESTINATION_TABLE + "2"

I can potentially take a look

[jreback]

ideally should fixtureze these tables to avoid this

[jreback]

hmm this is still running this test (maybe there were 2) https://travis-ci.org/pandas-dev/pandas/jobs/629507853

[charlesdong1991]

ahh, sorry about CI failure! I just take a look, and the test failed when if_exists=='fail'. And as i looked at pandas-gbq and found out the problem.

From pandas.to_gbq documentation (https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.to_gbq.html), when if_exists=='fail', it says If table exists, do nothing., so in the test, i leave the table size as is.
But in pandas_gbq documentation (https://pandas-gbq.readthedocs.io/en/latest/writing.html#writing-to-an-existing-table), it says when if_exists=='fail', should raise TableCreationError and this is why CI failed.

I did not double check the documentation from pandas_gbq when working on it, and somehow, this error did not show up in my local and PR test, so was overlooked. Very sorry about that.

changing the result to TableCreationError for fail should fix the problem

@datapythonista @jreback @alimcmaster1

could be good to change the documentation as well in #30478 to make the documentation clearer for fail.

[datapythonista]

Can you open a PR with this @charlesdong1991 ? Thanks!

[charlesdong1991]

yeah, of course, but i saw a PR related to it was already opened by @alimcmaster1

Or do you mean that you want a PR for doc change? @datapythonista

[datapythonista]

Sorry, didn't see the PR. But yes, creating one for the doc changes you propose sounds good.

[charlesdong1991]

sure!

[alimcmaster1]

Our docs are a copy of the gbq ones, i've updated the pandas side here. Would be good to do the same in pandas_gbq.

Thinking we should maybe restore a little documentation around this. Since these tests do not run on PRs. (Only master)