search

project-fifo / fifo_dt

FiFo Datatypes
Other
1 stars 3 forks source link

how does read_fsm get rid of partially failed write? #2

Closed liuzhen closed 7 years ago

liuzhen commented 7 years ago

Say we have A, B, C vnodes for a datum, with "n-w-r" set to 3-2-2, which should be a "typical".

Step1, write only succeeded at A, failed at B & C, reporting to uplevel as failed; Step2, read request firstly reached to A & B, should we return A's view of data?

According to ft_obj's merge function, A gets returned to uplevel, but for Step1, the write already has been reported failed.

Any newly partial successful write predominates the result, no quorum or others gets in to ensure a consensus while reading.

Licenser commented 7 years ago

This is intended behaviuour

liuzhen commented 7 years ago

I start to guess there might be no such mechanism, while your quick response confirmed, thanks.