greptimedb
greptimedb copied to clipboard
GreptimeTeam
Manually flush trigger
Current the flush will only be scheduled when memory threshold reached. It's a completely internal operation. This ticket request an external public interface to force flush a table or a region.
The requested API is helpful when
- Testing behaviors related to flush
- Scheduling a table (close in one instance and open in another one)
@waynexia I dig around, found this function: https://github.com/GreptimeTeam/greptimedb/blob/57979c9d3dba13beef5d63b4dc73fa1b3281e7e6/src/storage/src/flush.rs#L37
Should we add a new FlushStrategy like EventBasedStrategy which holding a channel receiver be good?
And this strategy also can be composed with SizeBasedStragety.
Glad you are interested!
found this function
Yes, in the write process we will check should_flush every time before writing. This is an async way to implement (if no write request then the flush event won't be executed neither). And I'm also wondering to exposing an active (sync) interface on RegionWriter to perform flush without depends on write request. Both active and passive manners are useful.
But consider the maybe use-case of the passive way, looks like FlushSwitch is able to cover the unit-test scenario. If we have EventBasedStrategy in the non-test path, a problem is where to put the tx of this strategy? So in my perspective we should implement the active way first.
