Rare and random HTTP 503 errors from JPL Horizon on parallax models

[ketozhang]

Less than 1% of the times when I am running a model with parallax (letting pyLIMA fetch ephemerides for me), I get errors from JPL Horizon.

The culprit might be here where too many queries are made to JPL: https://github.com/ebachelet/pyLIMA/blob/92b4f475214b693f8906b4f2ab16e2f69c2fd8d5/pyLIMA/parallax/JPL_ephemerides.py#L78-L110

However, it is also likely because I am running many models at the same time overloading the Horizon servers.

---

A fix that satisfies both issue is to add a "retry with backoff" algorithm (try to query again after waiting an exponentially increasing time)

It should be added to either here... https://github.com/ebachelet/pyLIMA/blob/92b4f475214b693f8906b4f2ab16e2f69c2fd8d5/pyLIMA/parallax/parallax.py#L181-L184

or as part of JPL_ephemerides.horizons_API itself.

[ketozhang]

I can contribute to this sometime soon.

My idea is to add one of these two libraries

• https://github.com/hynek/stamina
• https://github.com/litl/backoff

---

One thing to discuss is whether this is the right place to fix this or should the user handle this themselves (by handling whenever they see a requests.exceptions.HTTPError from pyLIMA)