New "deadlock" type DB_SQLError in 9.5.0 compared to 9.2.5?

We recently upgraded all our Tango systems to 9.5.0 from 9.2.5a. Now we can see that when booting a system with multiple servers configured, some of them (nondeterministically) don't come up on the first try. The error is the following:

DevFailed: DB_SQLError: Failed to query TANGO database (error=Deadlock found when trying to get lock; try restarting transaction) (in DataBase::db_put_device_property())

Retrying to start the affected server(s) usually works without problems.

For background, our servers are written so that they pump all configuration for all devices (which is stored in a file) into the Tango database at server startup. Therefore, there are a lot of such put_device_property calls during the initialization phase, and a lot of them in parallel when multiple servers (about 20) start up at once.

Of course I can put a big retry-loop around these calls, but that seems the wrong place to fix this. The database should be able to handle concurrent calls, and 20 servers isn't such a huge number.

Especially, the curious thing is that this was never a problem with Tango 9.2. Has anything changed from the database point of view? Any lock timeouts introduced or lowered? Is there anything else I can do to help debug this further?

Thanks!