Run conformance tests with granian, gunicorn, uvicorn too #33
Conversation
conformance/test/test_server.py
Outdated
| "--skip", | ||
| "**/bidi-stream/full-duplex/**", | ||
| ] | ||
| # Servers often run out of file descriptors on macOS due to low default ulimit. |
There was a problem hiding this comment.
Previously with hypercorn, it would fail due to ulimit issue in a default macos shell without --parallel=1. For some reason with granian it would have issues even with the parallelism. But I realized it's not difficult to automatically increase ulimit like this, so I did so and removed the parallel limit.
|
|
||
| _skipped_tests_async = [ | ||
| "--skip", | ||
| # There seems to be a hypercorn bug with HTTP/1 and request termination. |
There was a problem hiding this comment.
This seems to only have been happening with the low ulimit
anuraaga
changed the title
anuraaga
force-pushed
the
server-subprocess
branch
from
820b1c4 to
2176f4c
Compare
anuraaga
force-pushed
the
server-subprocess
branch
2 times, most recently
from
23251a3 to
be6d91f
Compare
|
While I could work through some flakiness locally in docker, looks like there is still some issue will check it out |
anuraaga
force-pushed
the
server-subprocess
branch
3 times, most recently
from
f3ba171 to
be0d36f
Compare
|
Got 3 runs in a row to succeed so hopefully this is fine now but let's keep an eye out for flakiness. I will also see if I can understand why granian would have issues with concurrency since in other testing I've found it to work pretty well. |
| ): | ||
| # Granian seems to have a bug that it prints out 0 rather than the resolved port, | ||
| # so we need to determine it ourselves. If we see race conditions because of it, | ||
| # we can set max-servers=1 in the runner. |
There was a problem hiding this comment.
want to add a ref: https://github.com/emmett-framework/granian/issues/711 to this comment so we can revisit when fixed?
conformance/test/server.py
Outdated
| env=_server_env(request), | ||
| ) | ||
| stdout = proc.stdout | ||
| assert stdout is not None # noqa: S101 |
There was a problem hiding this comment.
should we just disable the ruff rule that checks for assert usage in conformance/ or conformance/test?
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
anuraaga
force-pushed
the
server-subprocess
branch
from
7b1c7d4 to
3f54d06
Compare
Signed-off-by: Anuraag Agrawal <[email protected]>
anuraaga
force-pushed
the
server-subprocess
branch
from
ed148af to
4da1ff5
Compare
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
Signed-off-by: Anuraag Agrawal <[email protected]>
anuraaga
force-pushed
the
server-subprocess
branch
from
7ddb53c to
887de92
Compare
|
Sorry for the notification spam - I had good luck yesterday but very much not so today. In the end
It means I can't find a good solution for http/2 on python yet. Will keep on trying |
2 participants
In preparation for writing a doc on deployment, I wanted to make sure the common app servers are all tested. This adds granian, gunicorn, and uvicorn.
While previously hypercorn was started directly, I switched to just using subprocess for all the servers. It was already pretty tedious to do programmatic start for hypercorn, and some servers like granian it's not practical since they need to control the main thread. Instead of having many different ways of starting a server by code, we get to instead have a consistent pattern with just different CLI flags so overall it seems like a win.