On 2019/7/1 下午7:11, Pierre Couderc wrote: > > On 6/30/19 9:20 AM, Qu Wenruo wrote: >> >> On 2019/6/30 上午2:50, Pierre Couderc wrote: >>> 1- Is there a summary of btrfs recommendations for maintenance ? >>> >>> I have read somewhere that  a monthly  btrfs scrub is recommended. Is >>> there somewhere a reference,  an "official" (or not...) guide of all >>> that  is recommended ? >> I'd say scrub can tell you how bad your disks are. >> But at least, I'd recommend to do an offline check (btrfs check) and a >> scrub after every unclean shutdown. > OK, thank you ! >> >> For the maintenance recommends, Zygo Blaxell should has a pretty good >> ideas on this topic. > Is there some link to these "ideas"...? >>> I am lost in the wiki... >>> >>> 2- Is there a repair guide ? I see all these commands restore, scrub, >>> rescue. Is there a guide of what to do when a disk has some errors ? The >>> man does not say when use some command... >> If you're doing scrub routinely, it should give your a more reliable >> early warning than SMART. >> >> Normally for bad disk(s), you could replace them in advance. E.g when >> the disk begins to have unrecoverable errors suddenly, it is a good time ^^^ recoverable >> to replace it. >> >> If it's too late that the fs can't be mounted any more, my recommends >> are: >> 1. btrfs check --readonly and save the output >>     Sent the output to the mail list for help. The mail list will provide >>     much detailed solution to recover. >> >> 2. try to mount the fs RO and save the output >>     Just like step 1. >> >> 3. Btrfs-restore if you have enough space >>     The only generic and easy to use way to salvage data. >> >> The following methods are only for guys with some btrfs internal >> knowledge: >> - mount with usebackroot option >> - btrfs-find-root paired with btrfs check --chunk-root/--tree-root >>    Both methods are mostly the same, trying to use old roots. >>    Not reliable. >> - experimental kernel patches to skip extent tree at mount time >>    Kernel equivalent for btrfs-restore. Needs to recompile at least btrfs >>    kernel module. Only works for extent tree related corruption. >> > Thank you, I note that Although I hope you never need any of these recovery steps. Thanks, Qu