d-dd
commented
10 years ago After some tests, I found out that the database chat logging itself wasn't the reason for the restart. When there are too many chatMsg frames, Naoko must process each one, and if it takes too long, it reaches the (heartbeat) TIMEOUT before the heartbeat is felt (because it's so far behind the queue).
The database does slow things down; after a spam session all database operations are blocked until the chat messages are written, so this may still be good to implement.
When the chat goes too fast (spam) Naoko has trouble writing each line to its sqlite database. It uses up a lot of disk and CPU, until it can no longer catch up and times out to the server. Chat logging could be limited to at most once every second. When there is more than one line in a second, multiple lines will be cached and wait to be inserted in one transaction. In the event of a crash, the most data loss that will happen (because of this change) is one second of chat.