In D3056#74217, @olasd wrote:

Jeez. The check_rpc_serve context manager is so backwards it broke my brain a little. Took me quite a while to figure out that the thread actually runs the test rather than the server. I also think the synchronization between the test and the server initialization should be done more explicitly.

(but I don't really have a better thing to suggest right now)

rebase

Add a commit to "normalize" the get_journal_client() function

In D3044#73903, @ardumont wrote:

I'd say it'd be neat to add some docstrings and type at the same time, don't you think?

In D3040#73644, @ardumont wrote:

Technically, this sounds fine ;)

What is the new use case? Will we merge back the content topics into one or something?

improve commit message

attempt to fix the mess

Also, @ardumont I agree with your other comments, but these will come later, if you may :-)

In D3043#73818, @ardumont wrote:

It seems we duplicate the objects data (CONTENT, etc...) both in the pytest_plugin and in the journal_data module.
is that expected?

Thanks a lot

Fine for me. I agree the semantic of this date in the context of a deposit could be documented somehow (a simple comment may be enough).

fix the test test_generation_blacklist

typo in the commit message

Set the default value for blacklist_types to ("origin_visit_status",)

rebase *and* pick all the revisions involved

rebase

In D3010#72991, @douardda wrote:

In D3010#72943, @ardumont wrote:

In test_replay, get rid of test_storage_play and
test_storage_play_with_collision because these tests do require a proper
kafka setup to run, which we do not want to require kafka in the
swh.storage package.

hu ho...
i'm not fine with getting rid of test_storage_play*.

See for example D3015 which moves the hash collision policy within the retry proxy.
To ensure this continued to be ok, i've kept the test_storage_play_with_collision as is, only changed the test config with retry proxy and the test is green... (and it is, well locally because right now it's in diff).
My point is those are important tests.

I understand, however, these 2 tests, as currently written, are too far "integration tests" rather than "unit tests". And it makes the testing stack too complex.

So IMHO we want "unit" tests dedicated to parts of this stack that handle collisions 9typically without kafka being involved, swh-storage side) where there are useful, and integration tests (in the docker-dev project typically) to check the whole stack behaves properly.

rebase

In D3010#72943, @ardumont wrote:

In test_replay, get rid of test_storage_play and
test_storage_play_with_collision because these tests do require a proper
kafka setup to run, which we do not want to require kafka in the
swh.storage package.

hu ho...
i'm not fine with getting rid of test_storage_play*.

See for example D3015 which moves the hash collision policy within the retry proxy.
To ensure this continued to be ok, i've kept the test_storage_play_with_collision as is, only changed the test config with retry proxy and the test is green... (and it is, well locally because right now it's in diff).
My point is those are important tests.

add the commit "cli: rename the command 'backfiller' as 'backfill'" in the diff

rebase and improved to make it also compatible with swj.journal 0.0.29

rephrase the commit msg