Occasional Github Actions failures when calling wget

[clark-evans]

Occasionally (~10-20% of action collections?), Github Actions jobs will fail due to a wget error -- e.g., #151 (comment) -- in which a download fails due to exit code 4. Per the wget documentation, this indicates a network error. I suspect this means that the wget request is timing out. Note that when these failures occur, they usually occur with the GNU builds, such as autotools in the example linked above.

We currently get around this by manually rerunning the failed jobs; I've yet to encounter an instance where the same job failed twice. We may wish to modify our runners to add some failsafes around each wget call to more robustly address this issue; e.g.,

wget --tries=5 ...

@dustinswales @jderrico-noaa

[clark-evans]

Added note: #169 had four CI tests fail, #177 had four CI tests fail, and #192 had three CI tests fail tonight - all at the miniconda stage with the following error:

Warning: 
  CondaHTTPError: HTTP 503 Service Temporarily Unavailable for url <https://conda.anaconda.org/conda-forge/noarch/repodata.json>

I don't think the precise fix for this would be identical to that for wget, but the same general concept applies - the job fails because the downstream dependency is not satisfied.

[dustinswales]

 Service Temporarily Unavailable for url <https://conda.anaconda.org/conda-forge/noarch/repodata.json>

@clark-evans There seems to be a problem with conda.

[dustinswales]

@clark-evans I added --tries=5 to the CI scripts. I'm going to keep running it until it triggers a "timeout and retry". If all works as expected I will open PR.

[clark-evans]

Thanks, @dustinswales - appreciate it!

[clark-evans]

I haven't seen any of these issues since --tries = 5 was added to the CI script, so I'll go ahead and close this for now.