curl: (16) Send failure: Connection was aborted

[marandaneto]

Hi, thanks for the hard work.

I'm experiencing very often connection errors when running this action on GH Actions - Windows.

Run 6a218f2
Run $GITHUB_ACTION_PATH/setup.sh stable Windows x64
Installing Dart SDK version "latest" from the stable channel (release) on windows-x64
Downloading https://storage.googleapis.com/dart-archive/channels/stable/release/latest/sdk/dartsdk-windows-x64-release.zip...
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed

0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
0 0 0 0 0 0 0 0 --:--:-- 0:00:09 --:--:-- 0
0 0 0 0 0 0 0 0 --:--:-- 0:00:11 --:--:-- 0
curl: (16) Send failure: Connection was aborted
End-of-central-directory signature not found. Either this file is not
a zipfile, or it constitutes one disk of a multi-part archive. In the
latter case the central directory and zipfile comment will be found on
the last disk(s) of this archive.
unzip: cannot find zipfile directory in one of /c/Users/runneradmin/dartsdk.zip or
/c/Users/runneradmin/dartsdk.zip.zip, and cannot find /c/Users/runneradmin/dartsdk.zip.ZIP, period.
Error: Download failed! Please check passed arguments.
Error: Process completed with exit code 1.

It's just flakiness but it's happening quite often, is there something we can improve about that?

Thanks.

[devoncarew]

Hi - thanks for the report!

I'd like to keep this issue open for a bit to see if we get reports from other users as well.

Do you have a very rough sense for how flaky or any other factors involved?

Wrt mitigating this on the action side, we do want to add more caching in this action (#2). This would mean that each job would have a lower likelihood of needing to download the SDK; it wouldn't address whatever the underlying source of flakiness is, but it would impact each job proportionally less.

[marandaneto]

@devoncarew thanks for your reply.

I've seen flakiness only on Windows, 1 out of ~7...8 runs?! it's quite often, usually rerunning helps.

No idea, maybe GH Action servers and storage.googleapis.com just don't get along so well, DNS resolving, not sure, I also experience a bit of flakiness with Maven Central servers as well, so it's not only this Action.

https://curl.se/libcurl/c/libcurl-errors.html

CURLE_HTTP2 (16)

A problem was detected in the HTTP2 framing layer. This is somewhat generic and can be one out of several problems, see the error buffer for details.

Maybe we can improve logging here?

Caching would definitely help.