On Fri, Aug 30, 2013 at 09:44:28AM -0500, Eric Sandeen wrote: > On 8/29/13 3:19 PM, Chris Murphy wrote: > > > > On Aug 29, 2013, at 1:53 PM, Hugo Mills wrote: > > > >> On Thu, Aug 29, 2013 at 01:44:54PM -0600, Chris Murphy wrote: > >>> > >>> Certainly, if known for sure it won't be more than 30 seconds? > >> > >> Mmm... it'll depend on the setting of the commit period, which up > >> until a couple of weeks ago was always 30s, but someone posted a patch > >> to give it a config knob… > > > > > > > > "Proceeding will roll back the file system to a previous state, and > > may cause the loss of successfully written data since the last commit > > period (30 seconds by default). Proceed? (Y/N)" > > Is it just loss of data, or might this also result in a filesystem with inconsistent metadata, which then requires a fsck? No the metadata is always consistent (well, in theory, barring bugs and out-of-band corruption). > Above sounds like it's "just" reverting to a previous (consistent) state. Is that correct? Yes, it's dropping the log of accepted-but-uncommitted work. This is a Bad Thing in the sense that something that's reached the log is reported to the application as being successfully written. If the application critically relies on that (e.g. databases), then we've discarded durability from ACID. (Can you guess I've been marking Databases resit exam papers this morning? :) ) Hugo. > -Eric > > p.s. fwiw when the xfs_repair zero-log option "-L" is used, we say: > > "ALERT: The filesystem has valuable metadata changes in a log which is being\n" > "destroyed because the -L option was used.\n")); That's a reasonable wording too. Hugo. -- === Hugo Mills: hugo@... carfax.org.uk | darksatanic.net | lug.org.uk === PGP key: 65E74AC0 from wwwkeys.eu.pgp.net or http://www.carfax.org.uk --- We teach people management skills by examining characters in --- Shakespeare. You could look at Claudius's crisis management techniques, for example.