As a result, they get cleaned up by the cleanup job when it runs every 2 hours, since it thinks the peer is older than 31 minutes - Which it isn't - it's just that the recent announce didn't update in redis.
This is probably why the cache hits drop every 2 hours - since we delete peers who are active, and on a subsequent announce thinking we need to invalidate the cache for a "new seeder"
Fix: Even if there is no change to the cache, we need to call ZADD to update the seeder's timestamp.
ckcr4lyf
commented
1 year ago
Currently, if an announce can be served from cache (no "change"), we do so.
In the case of long term seeders, there is no change to
seed_countorleech_count, so we don'tZADDthem to the seeders: https://github.com/ckcr4lyf/kiryuu/blob/92611e8069c7205958cc475d9cf4478409db4e27/src/main.rs#L106-L125As a result, they get cleaned up by the cleanup job when it runs every 2 hours, since it thinks the peer is older than 31 minutes - Which it isn't - it's just that the recent announce didn't update in redis.
This is probably why the cache hits drop every 2 hours - since we delete peers who are active, and on a subsequent announce thinking we need to invalidate the cache for a "new seeder"
Fix: Even if there is no change to the cache, we need to call ZADD to update the seeder's timestamp.