[erlang-questions] mnesia sync_transactions not fsynced?

[Emile Joubert]

On 27/10/11 13:17, Håkan Mattsson wrote:

> Anyway, the subset of the commits that may be lost are the last written ones
> before a crash/shutdown. It is not a random subset.

In that case, would you expect consecutive records with ID numbers
starting at 0 to possibly survive the VM shutdown in the test script I
supplied? I always see a single record with much higher ID.

In both the original RabbitMQ code where this problem was discovered and
the distilled test script the missing transaction was never the last
one, but a random older one.



Emile