[Bug]: feed with errors aren't refreshed

[jvoisin]

Bug Summary

Feeds that fail in some ways aren't refreshed regularly

Description

Some websites might be offline/broken for whatever reasons. For example, lowtechmagazine runs on a solar panel, and thus might be offline when the weather is shite for a couple of days.

Unfortunately, miniflux is prompt to give up refreshing the feed:

I would expected something like an exponential backoff with a maximum of 48h or whatever.

This issue was reported to me by @magne4000

Steps to Reproduce

1. Subscribe to a feed
2. The feed goes offline/broken for a couple of days
3. Miniflux will stop refreshing the feed
4. The feed comes back online
5. Miniflux didn't refresh it, the feed is now dead to miniflux :<

Expected Behavior

Miniflux should keep refreshing broken feeds, with an exponential backoff.

Actual Behavior

Miniflux gives up.

Version

2.2.8

Browser

No response

Relevant Logs or Error Output

Additional Context

No response

Checklist

• I have searched existing issues to ensure this bug hasn't been reported before.

[fguillot]

This is currently the expected behavior, and it's documented here: https://miniflux.app/docs/configuration.html#polling-parsing-error-limit

POLLING_PARSING_ERROR_LIMIT

The maximum number of parsing errors that the program will try before stopping polling a feed. Once the limit is reached, the user must refresh the feed manually. Set to 0 for unlimited.

Default is 3.

While I agree that the current implementation is pretty basic, this is not a bug but a feature request to implement a more sophisticated exponential back-off.

Alternatively, you can set POLLING_PARSING_ERROR_LIMIT=0 to avoid that. There is also the possibility to reset the feed errors periodically with the command line: miniflux -reset-feed-errors as documented here https://miniflux.app/docs/cli.html#reset-feed-errors

[jvoisin]

Would something like a default exponential back-off with an upper-bound of 1 week be acceptable instead?

[fguillot]

Would something like a default exponential back-off with an upper-bound of 1 week be acceptable instead?

Sounds acceptable. Some folks might want the upper bound to be configurable.