[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: ordered data



Stephen, 

Thanks for explaining that.  The reason we want something like this is
that we want a persistent record saying a file is (about to be)
modified.  We generate this upon the first write to a file. In this
way the replication doesn't overlook it in case the system crashes
before e.g. a close replicatin record is generated by InterMezzo.

BTW it's not that we are more religious about data than ext3, it's
just that we have multiple copies of data and equality is something we
are fairly religious about.

Here are 3 options:

(1) Calling sync is not that attractive (but possible).  

(2) Is the following possible: 

 strart_trans
   do a journaled write for our "record"
   call ext3_file_write (on ordered file system). 
 end_trans

Putting it differently can we achieve the ordering / making the record
part of the same transaction in this way.

(3) Another way would be to change the file to a data-journaled file
for the first write, but I remember you saying that this is a fairly
heavyweight operation too.

Is this making any sense? 

- Peter -





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]