TSI & "error syncing wal"

Version 1.6.3. When using the “inmem” index, we received the error “limit exceeded … dropped …”. After switching to “tsi 1” index, writing to the database from Prometheus stops after some time (3-15 minutes) and we get the error “engine: error syncing wal”. The record is restored only after an “influxd” restart. After a while, the error repeats. What is the problem?

What parameters do you recommend to change? Most of the parameters in the configuration file have default values.