On Thu, Sep 1, 2016 at 2:09 PM, Volker Armin Hemmann <[email protected]> wrote: > > a common misconception. But not true at all. Google a bit.
Feel free to enlighten us. My understanding is that data=journal means that all data gets written first to the journal. Completed writes will make it to the main filesystem after a crash, and incomplete writes will of course be rolled back, which is what you want. But simply disagreeing and saying to search Google is fairly useless, since you can find all kinds of junk on Google. You can't even guarantee that the same search terms will lead to the same results for two different people. And FWIW, this is a topic that Linus and the ext3 authors have disagreed with at points (not this specific question, but rather what the most appropriate defaults are). So, it isn't like there isn't room for disagreement on best practice, or that any two people with knowledge of the issues are guaranteed to agree. >> >> Now, I can still think of ways you can lose data in data=journal mode: >> >> * You mounted the filesystem with barrier=0 or with nobarrier; this can >> result > > not needed. Well, duh. He is telling people NOT to do this, because this is how you can LOSE data. >> >> * Your application didn't flush its writes to disk when it should have. > > not needed either. That very much depends on the application. If you need to ensure that transactions are in-sync with remote hosts (such as in a database) it is absolutely critical to flush writes. Applications shouldn't just flush on every write or close, because that causes needless disk thrashing. Yes, data will be lost if users have write caching enabled, and users who would prefer a slow system over one that loses more data when the power goes out should disable caching or buy a UPS. > > nope. Care to actually offer anything constructive? His advice was reasonably well-founded, even if I personally wouldn't do everything exactly as he prefers to do so. -- Rich
