Deal with response rate limiting #49

New Issue

Closed

opened 2020-11-30 20:13:34 +00:00 by projectmoon · 1 comment

projectmoon commented

2020-11-30 20:13:34 +00:00

Owner

Possible solution:

Store message response in a sled Tree.
The Tree has a watcher subscribed to all new entries.
When the watcher receives a new event, it will dump the message into a buffered queue, which is consumed by an async task. Possibly MPSC model.
The consumer has a single backoff time, which increases and decreases depending on ratelimit errors.
A message that failed to send is shoved back into the queue.
When message successfully sent, remove it from the sled Tree.
On bot startup, all messages still in the tree are immediately sent into the queue.

With this solution we may under rare circumstances get responses sent twice on a bot restart, but it's better than losing the messages. We can work on not duplicating message sends after basic rate limit handling.

If a response messages from the bot is rate-limited, the message will simply be dropped and the user will never get a response. An error is logged, but this is all. The solution is some kind of queue of unsent messages with exponential backoff. Sled's event watcher is a good candidate for this, as it allows us to store any unsent messages. Possible solution: - Store message response in a sled Tree. - The Tree has a watcher subscribed to all new entries. - When the watcher receives a new event, it will dump the message into a buffered queue, which is consumed by an async task. Possibly MPSC model. - The consumer has a single backoff time, which increases and decreases depending on ratelimit errors. - A message that failed to send is shoved back into the queue. - When message successfully sent, remove it from the sled Tree. - On bot startup, all messages still in the tree are immediately sent into the queue. With this solution we may under rare circumstances get responses sent twice on a bot restart, but it's better than losing the messages. We can work on not duplicating message sends after basic rate limit handling.

projectmoon commented

2021-05-26 22:33:01 +00:00

Author

Owner

The SDK now has built-in support for handling server rate limiting. So this original issue isn't much of a problem anymore.