Upload progress bar jumps weirdly in production

[pastewka]

The progres bar jumps back to zero at intermediate steps. This is identical to what we see for the analysis progress bars. It does appear to only happen in production, not when developing locally.

[pastewka]

See #755

[mcrot]

Now I've seen the following: During upload, the browser requests the current progress state by short intervals. Normally, these HTTP requests result in JSON data with keys size and received, but sometimes the result is just null:

I think this null result marks the moment when the bar vanished for a second. When looking into the code of uploadprogressbar this means, that the cache key, which is built from the progress_id, was not found in the cache. I think this could be a hint for the other cache problems.

[mcrot]

It seems that this jumping was induced by using multiple cache instances. Sometimes the progress ID was included, sometimes not. The progress bar no longer jumps with a single cache server. Nevertheless, the upload of a big files stops with an exception on the server side. But that is another problem. So I think this here is solved. I'll keep it open until the compose file is changed accordingly.

[pastewka]

Okay, great! The larger file issue seems to be related to size limitations of memcached. I think those should go away if we switch to redis, which does not have those limitations.

[mcrot]

It seems to also happen with smaller files. Somehow the memcached server is terminated from outside, probably by Docker Swarm, maybe due to instabilities in the internal network. If this is the case, Redis would be frequently terminated, too. But I'll give it a try, we will know more afterwards.

[mcrot]

This seems to be solved after simplified setup with redis and probably more relevant, fixed a wrong configuration with supervisord.