Re: [Quickfix-developers] Semi OT : QF internal data hotbackup
Brought to you by:
orenmnero
|
From: Oren M. <or...@qu...> - 2007-01-04 16:38:17
|
> - if just backuping more regularly is considered enough, are there any > risks of copying a corrupted/incomplete file ? I did not check but > I guess > QF opens its file pointer when the session starts and closes it > when it > stops. QuickFIX does a write and flush with every message as it is stored. I'm not very well versed on how a hard drive or OS handles this scenario. I assume that it is unlikely but possible that the file can be caught in mid flush giving you a corrupted file. Modifying the FileStore to make regular backups might be a better solution since you're backup solution can coordinate with the rest of the store. > - if we go for NFS, how does QF handle the FS access to what is > stored in > FileStorePath ? Does it work in RAM and flushes to disk from time > to time > (with explicit fflushes or the OS decides) or are the messages > reloaded > from disk every so often ? As I said above a flush goes out with every sent message. If the message is particularly large, the OS could also force a flush in between. Explicit flushes guarantee a message is stored before it is sent. Once a message is sent, there is no need to ever read from the disk unless processing a resend request. During normal operation, this should not happen. Typically, resend requests are also not particularly performance intensive. Messages you receive do not get stored and so their performance will be generally unaffected by whatever storage you choose. The exception to this is to update the sequence number file. |