Upload failed in docker logs and lot less traffic

A little info that might be relevant. It used to be that out of 95 upload transfers only 80 were finished and 15 cancelled. Roughly 84%. If you used to see success rates higher than that, you did better than average on the test traffic which came mostly from one location. Please note that this means other nodes did worse than average back then.

Since then, 3 things changed:

  • Test traffic has stopped and we now see traffic from all around the world. Nodes that used to do better than average likely saw a drop because their advantage with the test source doesn’t necessarily apply with other customers.
  • Instead of 95, now 110 transfers are started, but still only 80 finish, so 30 are cancelled. This means the average success rate has dropped to 73%.
  • The lower amount of total traffic may have caused previously bandwidth constraint nodes to become more competitive. They would have lost more races when there were multiple transfers going on on their node, but with only a single transfer they are better able to keep up. This may cause some high bandwidth connections or faster hardware to see less of an advantage than they used to.

Unfortunately for me, I saw a big drop as well. I did exceptionally well with the test traffic. About 97-99% success rates. I have now dropped to 47%. I’m located in the Netherlands, so with test traffic from Germany I used to do pretty well. My guess is we now see quite a few US based customers testing as the two users reporting higher results were both from the US and they both scored above the new average.

Please also note that from the reports here it may seem like almost everyone saw a drop. This isn’t necessarily the case. People go to a forum to post about an issue. So drops in success rate are by definition over represented.

Edit: In order to make this post more complete I wanted to add that an additional change in how the uplink closes connections more aggressively after uploads are complete can cause failed uploads in the logs while the transfer actually completed. This can be seen in some cases because cancelled uploads are later deleted in the logs. So the piece was definitely there. The impact may be a lot smaller than it seems as a result of this. More info here.

In my experience update 0.33.4 improved the success rates a little again, but not back to the old level. So keep this in mind. Your node is probably working just fine. It looks worse than it is.